Latest / Deep Tech Germany – AI, Robotics & Frontier Innovation by Startuprad.io™ / AWS's Jennifer Grün: AI Monetization from POC to Proof of Value
Transcript
- WEBVTT 00:00:00.017 --> 00:00:04.317 <v Speaker1>That's a really great question and a very critical one for any founder, 00:00:04.517 --> 00:00:06.537 <v Speaker1>especially as they start off in scale. 00:00:06.777 --> 00:00:10.917 <v Speaker1>So it really depends on your business, the market, and the problem you're solving. 00:00:11.217 --> 00:00:16.277 <v Speaker1>So let's start with Gen.AI as a feature. That is often the safest and most effective strategy. 00:00:16.497 --> 00:00:21.017 <v Speaker1>So you should use it when you have a strong existing user base and a clear use 00:00:21.017 --> 00:00:23.357 <v Speaker1>case for AI to improve a current workflow. 00:00:24.017 --> 00:00:29.177 <v Speaker1>So instead of building a new product from scratch, you embed AI directly into 00:00:29.177 --> 00:00:31.477 <v Speaker1>your existing offering to enhance its value. 00:00:31.657 --> 00:00:35.197 <v Speaker1>So you typically can't charge a premium for a single feature, 00:00:35.257 --> 00:00:37.877 <v Speaker1>but instead you can use tier pricing, 00:00:38.217 --> 00:00:43.317 <v Speaker1>offering AI features in higher price plans, or a freemium model where basic 00:00:43.317 --> 00:00:48.837 <v Speaker1>AI functions are free, but advanced capabilities, imagine more queries or larger 00:00:48.837 --> 00:00:50.957 <v Speaker1>outputs, are meted with a credit system. 00:00:51.440 --> 00:01:11.280 <v Music> 00:00:56.197 --> 00:01:03.837 <v Speaker0>Welcome to StartupRad.io, your podcast and YouTube blog covering the German 00:01:03.837 --> 00:01:08.737 <v Speaker0>startup scene with news, interviews and live events. 00:01:10.817 --> 00:01:15.497 <v Speaker0>AWS is proud to sponsor this week's episode of StartupRad.io. 00:01:15.497 --> 00:01:22.037 <v Speaker0>The AWS team compromises former founders, CTOs, venture capitalists, 00:01:22.457 --> 00:01:27.237 <v Speaker0>angel investors and mentors, ready to help you prove what's possible. 00:01:27.697 --> 00:01:36.037 <v Speaker0>Since 2013, AWS has supported over 280,000 startups across the globe and provided 00:01:36.037 --> 00:01:42.477 <v Speaker0>$7 billion in credits through the AWS Active program. 00:01:42.477 --> 00:01:44.977 <v Speaker0>Big ideas feel at home at AWS. 00:01:45.457 --> 00:01:49.997 <v Speaker0>And with access to cutting-edge technologies like generative AI, 00:01:50.337 --> 00:01:54.277 <v Speaker0>you can quickly turn those ideas into marketable products. 00:01:55.157 --> 00:01:59.117 <v Speaker0>Want your own AI-powered assistant? Try Amazon Q. 00:01:59.897 --> 00:02:05.317 <v Speaker0>Want to build your own AI products? Privately customize leading foundation models 00:02:05.317 --> 00:02:09.457 <v Speaker0>on Amazon Badrock. Want to reduce the cost of AI workloads? 00:02:09.777 --> 00:02:13.157 <v Speaker0>AWS Tranium is the silicon you're looking for. 00:02:13.920 --> 00:02:17.000 <v Speaker0>Whatever your ambitions you've already had 00:02:17.000 --> 00:02:20.280 <v Speaker0>the idea now prove it's possible on aws 00:02:20.280 --> 00:02:23.820 <v Speaker0>visit aws.amazon.com forward 00:02:23.820 --> 00:02:26.520 <v Speaker0>slash startups to get started if you're a 00:02:26.520 --> 00:02:32.140 <v Speaker0>founder here's the challenge generative ai is everywhere but turning it into 00:02:32.140 --> 00:02:38.020 <v Speaker0>actual revenue is far harder than the hype suggests jennifer gruen senior specialist 00:02:38.020 --> 00:02:43.680 <v Speaker0>for generative ai at aws has worked with startups software vendors and enterprises 00:02:43.680 --> 00:02:45.840 <v Speaker0>to unlock billions in business value. 00:02:46.040 --> 00:02:51.580 <v Speaker0>Today we'll break down AI monetization frameworks, return on investment metrics 00:02:51.580 --> 00:02:56.640 <v Speaker0>that investors demand and real startup case studies so you can turn AI from 00:02:56.640 --> 00:02:58.120 <v Speaker0>a buzzword into business. 00:02:58.640 --> 00:03:02.440 <v Speaker0>Jennifer Grun is a senior specialist for generative AI and machine learning 00:03:02.440 --> 00:03:07.700 <v Speaker0>at AWS helping startups, software vendors and enterprises across Germany and 00:03:07.700 --> 00:03:12.880 <v Speaker0>in Central Europe to define, scale and monetize their AI use cases. 00:03:13.240 --> 00:03:17.740 <v Speaker0>She led go-to-market strategies for generative AI beyond AWS. 00:03:18.160 --> 00:03:23.700 <v Speaker0>She's passionate about democratizing machine learning and runs initiatives to 00:03:23.700 --> 00:03:27.120 <v Speaker0>bring knowledge to underrepresented groups and universities. 00:03:27.480 --> 00:03:32.920 <v Speaker0>Today, she'll share what founders need to know about monetization opportunities with AI. 00:03:32.920 --> 00:03:37.520 <v Speaker0>From metrics like customer acquisition costs and customer lifetime value, 00:03:37.720 --> 00:03:44.440 <v Speaker0>to Canva's upsell playbook, to why ROI storytelling can make or break an investor 00:03:44.440 --> 00:03:46.120 <v Speaker0>pitch. Jennifer, welcome. 00:03:46.880 --> 00:03:50.140 <v Speaker1>Thank you, Jörn. Really excited to be here today. 00:03:50.780 --> 00:03:54.920 <v Speaker0>Totally. My pleasure. We had some issues with the setup, but now we're totally 00:03:54.920 --> 00:04:00.220 <v Speaker0>good and everything should work out. Welcome to our episode. 00:04:00.500 --> 00:04:09.660 <v Speaker0>You often describe the trillion dollar question as not what to build with AI, but how to monetize it. 00:04:09.880 --> 00:04:12.580 <v Speaker0>Why is this gap so persistent? 00:04:13.686 --> 00:04:18.366 <v Speaker1>Thanks, Joe. So that's a really great question. So for me, the main reason for 00:04:18.366 --> 00:04:20.466 <v Speaker1>this gap comes down to three things. 00:04:20.746 --> 00:04:23.566 <v Speaker1>So first, you have the economics of Gen AI. 00:04:24.046 --> 00:04:28.606 <v Speaker1>So unlike traditional software, where adding a new user has a very low marginal 00:04:28.606 --> 00:04:30.906 <v Speaker1>cost, generative AI is different. 00:04:31.206 --> 00:04:35.486 <v Speaker1>So every time a user interacts with a model, so every query, 00:04:35.766 --> 00:04:39.946 <v Speaker1>every word that is generated, it has a real tangible operational cost. 00:04:39.946 --> 00:04:42.066 <v Speaker1>And this also creates a paradox. 00:04:42.366 --> 00:04:47.346 <v Speaker1>So if you have a user who gets a ton of value from your service and uses it 00:04:47.346 --> 00:04:50.926 <v Speaker1>frequently, that might actually become less profitable for you. 00:04:51.086 --> 00:04:55.926 <v Speaker1>And this makes it really difficult to achieve the high gross profit margins 00:04:55.926 --> 00:05:00.746 <v Speaker1>that investors have come to expect, especially, for instance, for SaaS startups. 00:05:01.246 --> 00:05:06.466 <v Speaker1>And the second issue that I see founders are facing is often defaulting to familiar 00:05:06.466 --> 00:05:09.766 <v Speaker1>pricing models that just don't fit the technology. 00:05:10.366 --> 00:05:15.486 <v Speaker1>So, we are seeing a lot of startups try to use a per-seed or per-user model, 00:05:15.706 --> 00:05:17.166 <v Speaker1>but that can also backfire. 00:05:17.426 --> 00:05:21.186 <v Speaker1>So, imagine if your AI product makes the team more efficient, 00:05:21.486 --> 00:05:25.786 <v Speaker1>they might actually need fewer people, and that means also fewer seats. 00:05:26.758 --> 00:05:30.138 <v Speaker1>And this means that it would actually shrink your addressable market. 00:05:30.738 --> 00:05:34.418 <v Speaker1>And also, there's a market perception issue that I can observe. 00:05:34.638 --> 00:05:39.398 <v Speaker1>So when a lot of DEI features are given away for free to drive adoption, 00:05:39.818 --> 00:05:42.958 <v Speaker1>Customers can actually start to see them as stable stakes. 00:05:43.138 --> 00:05:47.018 <v Speaker1>So this makes it really tough to convince them to pay for what they believe 00:05:47.018 --> 00:05:51.918 <v Speaker1>should be a standard feature, where startups can't afford to raise to the bottom 00:05:51.918 --> 00:05:54.138 <v Speaker1>because every dollar counts in a startup. 00:05:54.138 --> 00:05:59.298 <v Speaker1>And then thirdly, many founders are so focused on the technical what, 00:05:59.538 --> 00:06:04.518 <v Speaker1>building the next great model or creating a really interesting feature that 00:06:04.518 --> 00:06:07.258 <v Speaker1>they fail to properly consider the business how. 00:06:07.478 --> 00:06:11.958 <v Speaker1>So in many of the conversations that I have with startups, I mentioned that 00:06:11.958 --> 00:06:15.038 <v Speaker1>you can't just build technology and hope for the best. 00:06:15.058 --> 00:06:19.518 <v Speaker1>So you really have to start with an outcome oriented mindset from day one. 00:06:19.518 --> 00:06:23.878 <v Speaker1>So this means understanding the specific customer pain point you're solving 00:06:23.878 --> 00:06:25.518 <v Speaker1>and how you will measure success. 00:06:25.778 --> 00:06:32.258 <v Speaker1>And a great example of this is a company that has navigated this beautifully is Lovable. 00:06:32.418 --> 00:06:35.478 <v Speaker1>So we are also going to have them soon in a hackathon here in Munich. 00:06:35.798 --> 00:06:40.398 <v Speaker1>And they're building a platform that allows anyone to create full stack web 00:06:40.398 --> 00:06:43.298 <v Speaker1>applications just by typing in a simple prompt. 00:06:43.298 --> 00:06:48.138 <v Speaker1>So they're solving a huge problem for non-technical founders who want to build 00:06:48.138 --> 00:06:51.758 <v Speaker1>an MVP quickly without the high cost of a development team. 00:06:52.158 --> 00:06:56.758 <v Speaker1>And what they did here is that Lovable has aligned their monetization with the 00:06:56.758 --> 00:06:57.938 <v Speaker1>value that they provide. 00:06:58.038 --> 00:07:02.158 <v Speaker1>So they use a credit-based pricing model where users pay for what they use, 00:07:02.298 --> 00:07:03.918 <v Speaker1>which directly aligns with their 00:07:03.918 --> 00:07:07.018 <v Speaker1>operational cost and the value of the application they are generating. 00:07:07.018 --> 00:07:11.738 <v Speaker1>And this really shows, again, an example of a company that has successfully 00:07:11.738 --> 00:07:15.658 <v Speaker1>moved beyond the what of the technology, but also figured out the how. 00:07:16.387 --> 00:07:19.707 <v Speaker0>Again Madhu, many startups get stuck in proof of concept. 00:07:20.067 --> 00:07:27.687 <v Speaker0>Why do POC technology choices fail to scale and how does this kill monetization potential? 00:07:28.007 --> 00:07:32.467 <v Speaker1>And that's an excellent question and it's something that we see actually all the time. 00:07:33.107 --> 00:07:37.427 <v Speaker1>So a significant reason for failure is really the fundamental difference in 00:07:37.427 --> 00:07:40.767 <v Speaker1>purpose between a POC and a production ready system. 00:07:41.427 --> 00:07:45.947 <v Speaker1>So many AI pilots that I have observed are driven by a technology-first mindset, 00:07:46.207 --> 00:07:49.587 <v Speaker1>where a solution is developed in search of a problem to solve. 00:07:49.767 --> 00:07:52.607 <v Speaker1>So think about a cool technology that people want to build. 00:07:52.867 --> 00:07:59.167 <v Speaker1>But to really transform the organization, the product requires buy-in from the entire organization. 00:07:59.507 --> 00:08:04.247 <v Speaker1>So we've seen startups where a lack of internal alignment on business priorities, 00:08:04.567 --> 00:08:09.247 <v Speaker1>funding, or success metrics completely stores a project. And this is why it's 00:08:09.247 --> 00:08:11.247 <v Speaker1>so important to use a structured approach. 00:08:11.587 --> 00:08:17.287 <v Speaker1>So I've worked here on an AI canvas that is forcing a shift in thinking beyond technology. 00:08:17.427 --> 00:08:21.627 <v Speaker1>So this canvas helps you to clearly articulate the customer needs, 00:08:21.827 --> 00:08:26.307 <v Speaker1>the value proposition and the metrics for success up front to make sure that 00:08:26.307 --> 00:08:30.147 <v Speaker1>the solution is tied to a real business problem with a measurable outcome. 00:08:30.147 --> 00:08:35.587 <v Speaker1>And the shift from a proof of concept to a proof of value, how we call it, 00:08:35.647 --> 00:08:38.487 <v Speaker1>is really crucial in addressing this issue that I just mentioned. 00:08:38.867 --> 00:08:43.347 <v Speaker1>So the proof of value really focuses on delivering and measuring a tangible 00:08:43.347 --> 00:08:47.787 <v Speaker1>business impact rather than really simply demonstrating technical feasibility. 00:08:47.787 --> 00:08:52.847 <v Speaker1>And what we've seen here is without a clear quantifiable metric for success, 00:08:53.067 --> 00:08:55.687 <v Speaker1>imagine cost savings or revenue growth, 00:08:55.987 --> 00:09:01.527 <v Speaker1>even a technically sound POC that wows your team will struggle to secure the 00:09:01.527 --> 00:09:04.727 <v Speaker1>continued investment required for full scale development. 00:09:05.127 --> 00:09:10.467 <v Speaker1>So this requires strong cross-functional collaboration among business, 00:09:10.667 --> 00:09:15.207 <v Speaker1>engineering and product teams to ensure alignment on a unified strategy. 00:09:16.370 --> 00:09:21.770 <v Speaker1>And also, what I like to outline as well is that the technology choices made 00:09:21.770 --> 00:09:25.650 <v Speaker1>for POC are often not suitable for a production application. 00:09:26.050 --> 00:09:30.190 <v Speaker1>So, for instance, POC might use a popular general purpose model, 00:09:30.470 --> 00:09:34.890 <v Speaker1>but a production application requires choosing the right model that balances 00:09:34.890 --> 00:09:37.810 <v Speaker1>capabilities, performance and cost. 00:09:38.050 --> 00:09:43.190 <v Speaker1>And also, it might need to be fine-tuned on proprietary data to create a truly 00:09:43.190 --> 00:09:44.310 <v Speaker1>differentiated product. 00:09:45.190 --> 00:09:50.750 <v Speaker1>Also, what I observe a lot is that POCs are often designed for a few users only, 00:09:50.910 --> 00:09:54.810 <v Speaker1>and they are really not built to handle thousands or millions of queries. 00:09:55.170 --> 00:09:59.530 <v Speaker1>So this means that they lack the necessary architecture for resilience, 00:09:59.790 --> 00:10:01.650 <v Speaker1>security, and global scalabilities. 00:10:02.590 --> 00:10:08.490 <v Speaker1>And that brings me a little bit to the challenges and the startup's monetization potential. 00:10:08.650 --> 00:10:14.210 <v Speaker1>That is quite direct and critical. So, if we think about the generative AI space, 00:10:14.410 --> 00:10:21.050 <v Speaker1>a unique characteristic is that the cost of goods sold is often tied directly to user consumption. 00:10:21.510 --> 00:10:26.010 <v Speaker1>So, if the underlying technology is not designed for efficiency and cost optimization, 00:10:26.530 --> 00:10:31.270 <v Speaker1>every user interaction can become an expense that quickly erodes your profitability. 00:10:31.870 --> 00:10:36.870 <v Speaker1>So imagine an inefficient, unoptimized POC may function really on the small 00:10:36.870 --> 00:10:38.730 <v Speaker1>scale that we discussed, but its 00:10:38.730 --> 00:10:42.490 <v Speaker1>per unit cost can make it financially unsustainable at a larger volume. 00:10:42.690 --> 00:10:47.070 <v Speaker1>So when a startup fails to address these issues, they end up with a product 00:10:47.070 --> 00:10:49.830 <v Speaker1>that is slow, expensive and unreliable. 00:10:50.110 --> 00:10:54.470 <v Speaker1>So their monetization potential is killed because you simply cannot charge for 00:10:54.470 --> 00:10:58.750 <v Speaker1>a service that doesn't work consistently and that can't handle a growing user base. 00:10:58.750 --> 00:11:03.710 <v Speaker1>So as a result, the startup is unable to leverage its technology for monetization 00:11:03.710 --> 00:11:09.330 <v Speaker1>through methods like premium features, tiered subscription or revenue sharing models. 00:11:10.239 --> 00:11:13.479 <v Speaker0>I see. So what do you see? 00:11:13.599 --> 00:11:20.419 <v Speaker0>What's a common mistake that startups make when defining monetization models for AI features? 00:11:20.739 --> 00:11:25.639 <v Speaker1>Really, the most common mistake that I see startups make is creating a pricing 00:11:25.639 --> 00:11:29.179 <v Speaker1>model that doesn't align with the underlying cost and the value of AI. 00:11:29.659 --> 00:11:34.659 <v Speaker1>So many startups that I've spoken with try to use a traditional perceived model 00:11:34.659 --> 00:11:37.879 <v Speaker1>to start with, which works well for conventional software. 00:11:37.879 --> 00:11:42.839 <v Speaker1>But if we take, for instance, agentic AI models, especially an autonomous one 00:11:42.839 --> 00:11:45.999 <v Speaker1>where we're developing towards, this may be completely ineffective. 00:11:46.539 --> 00:11:51.699 <v Speaker1>So an AI agent doesn't really need a user to be active, so you can't charge per seat. 00:11:52.039 --> 00:11:57.479 <v Speaker1>So unlike traditional software with its low marginal cost, every interaction 00:11:57.479 --> 00:12:00.439 <v Speaker1>with a generative AI model has a real tangible cost. 00:12:00.699 --> 00:12:03.439 <v Speaker1>So as a result, startups that charge 00:12:03.439 --> 00:12:07.459 <v Speaker1>a flat fee for unlimited usage find themselves in a dangerous paradox. 00:12:07.879 --> 00:12:12.199 <v Speaker1>So their most engaged and arguably most valuable customers are actually making 00:12:12.199 --> 00:12:16.939 <v Speaker1>them less profitable because their high usage drives operational expenses and 00:12:16.939 --> 00:12:18.139 <v Speaker1>eats into the bottom line. 00:12:18.399 --> 00:12:22.439 <v Speaker1>So this is, of course, the exact opposite of what you want in a business model. 00:12:23.279 --> 00:12:28.739 <v Speaker1>Beyond the cost paradox, another significant mistake is a failure to properly 00:12:28.739 --> 00:12:31.179 <v Speaker1>package offerings for different customer segments. 00:12:31.499 --> 00:12:35.099 <v Speaker1>So a good pricing strategy isn't just about the price itself. 00:12:35.099 --> 00:12:38.199 <v Speaker1>It's really about how you differentiate your product. 00:12:38.639 --> 00:12:43.119 <v Speaker1>For instance, you can create a basic package that offers a generic AI agent 00:12:43.119 --> 00:12:47.719 <v Speaker1>with broad expertise, and then a premium package that gives customers access 00:12:47.719 --> 00:12:51.439 <v Speaker1>to specialized agents trained on specific high-value topics, 00:12:51.799 --> 00:12:53.959 <v Speaker1>let's imagine legal or finance. 00:12:54.479 --> 00:12:59.859 <v Speaker1>And you can also package in enterprise-grade features that are crucial for larger 00:12:59.859 --> 00:13:05.179 <v Speaker1>clients, but not necessarily for every user. So this could mean single sign-on, 00:13:05.319 --> 00:13:09.119 <v Speaker1>audit logs, and data privacy assurances in your top-tier packages. 00:13:09.459 --> 00:13:13.439 <v Speaker1>So by doing this, you ensure that the customers with the highest willingness 00:13:13.439 --> 00:13:19.239 <v Speaker1>to pay are getting the most expensive product, which is a key principle of effective monetization. 00:13:20.556 --> 00:13:25.536 <v Speaker1>Another big mistake is choosing the wrong price metric, so the unit of measurement 00:13:25.536 --> 00:13:27.236 <v Speaker1>you use to charge for your product. 00:13:27.656 --> 00:13:30.956 <v Speaker1>Charging per token can be confusing and meaningless to a customer. 00:13:31.416 --> 00:13:35.496 <v Speaker1>Instead, you can use a more value-based metric, such as charging per completed 00:13:35.496 --> 00:13:38.356 <v Speaker1>task, per generated image, or per video minute. 00:13:38.756 --> 00:13:43.816 <v Speaker1>And these metrics are more tangible and directly relate to the value the customer is receiving. 00:13:44.156 --> 00:13:49.596 <v Speaker1>So ultimately, the goal is to tie your price to the actual business outcome you deliver. 00:13:50.436 --> 00:13:54.776 <v Speaker1>Imagine the percentage of cost savings or fixed fee per support ticket resolved. 00:13:55.236 --> 00:14:00.676 <v Speaker1>However, in the practical space, it is often too complex to attribute a specific 00:14:00.676 --> 00:14:02.576 <v Speaker1>outcome to a single AI agent. 00:14:02.756 --> 00:14:06.516 <v Speaker1>So especially if you think about multi-agent environments, so you might need 00:14:06.516 --> 00:14:08.656 <v Speaker1>to find a simpler proxy for value. 00:14:09.096 --> 00:14:14.396 <v Speaker1>And also, lastly, many startups make the mistake of not building a go-to-market 00:14:14.396 --> 00:14:16.516 <v Speaker1>strategy that can sell a new business model. 00:14:16.516 --> 00:14:20.916 <v Speaker1>So without a clear and compelling story that justifies a new pricing structure 00:14:20.916 --> 00:14:25.856 <v Speaker1>and quantifies its value, startups will struggle to get adoption and generate revenue. 00:14:26.076 --> 00:14:30.656 <v Speaker0>I have to tell my audience, you shared a lot of data, a lot of information with 00:14:30.656 --> 00:14:32.576 <v Speaker0>me for the preparation of this interview. 00:14:32.756 --> 00:14:38.076 <v Speaker0>In one of the monetization decks, you've outlined metrics like customer acquisition 00:14:38.076 --> 00:14:44.396 <v Speaker0>costs, monthly returning revenue, churn, customer lifetime value, and so on and so forth. 00:14:44.396 --> 00:14:48.616 <v Speaker0>Which are the most convincing for investors and customers? 00:14:48.936 --> 00:14:53.776 <v Speaker1>I think what is key about this question is to remember that investors and customers 00:14:53.776 --> 00:14:56.376 <v Speaker1>are looking for fundamentally different things. 00:14:56.676 --> 00:15:01.556 <v Speaker1>So if we take an investor, the most compelling metrics are those that prove 00:15:01.556 --> 00:15:05.456 <v Speaker1>your business is a sustainable, scalable and efficient growth machine, really. 00:15:05.856 --> 00:15:11.116 <v Speaker1>So they are looking for signals that the investment will lead to a predictable and high return. 00:15:11.986 --> 00:15:16.986 <v Speaker1>So if we take Monthly Recurring Revenue, or MRR for short, that shows investors 00:15:16.986 --> 00:15:20.326 <v Speaker1>that your business has a predictable and repeatable revenue stream. 00:15:20.406 --> 00:15:22.546 <v Speaker1>So not just one-off projects. 00:15:22.906 --> 00:15:25.786 <v Speaker1>And it's really the cornerstone for me of a healthy business. 00:15:25.806 --> 00:15:31.786 <v Speaker1>It demonstrates that your customers are continuously willing to pay for the value you provide. 00:15:32.326 --> 00:15:38.166 <v Speaker1>Then, Customer Lifetime Value, or CLV, demonstrates the long-term value of a 00:15:38.166 --> 00:15:39.266 <v Speaker1>customer to your business. 00:15:39.266 --> 00:15:44.106 <v Speaker1>So when an investor sees a high customer lifetime value, they see a company 00:15:44.106 --> 00:15:48.986 <v Speaker1>with a durable competitive advantage and a clear path to long-term profitability. 00:15:49.906 --> 00:15:54.686 <v Speaker1>Also, even better if you have a strong ratio between the lifetime value of a 00:15:54.686 --> 00:15:56.526 <v Speaker1>customer and the cost to acquire them. 00:15:56.666 --> 00:16:00.866 <v Speaker1>So this shows that your business model is fundamentally sound and that your 00:16:00.866 --> 00:16:05.166 <v Speaker1>go-to-market strategy is working efficiently. It tells an investor that you 00:16:05.166 --> 00:16:11.106 <v Speaker1>can pour money into a customer acquisition and get more back over the lifetime of these customers. 00:16:11.726 --> 00:16:15.046 <v Speaker1>Lastly, and I find this also a very important metric to consider, 00:16:15.306 --> 00:16:19.906 <v Speaker1>a low churn rate tells investors that customers are finding so much value in 00:16:19.906 --> 00:16:24.206 <v Speaker1>your solution that they are staying with you, which really reinforces a high 00:16:24.206 --> 00:16:27.966 <v Speaker1>customer lifetime value and proves that your business model is working. 00:16:28.686 --> 00:16:35.186 <v Speaker1>If we take customers, however, so they don't really care about your MRR or CLV. 00:16:35.446 --> 00:16:39.626 <v Speaker1>They really care about their own business. So the number one thing that will 00:16:39.626 --> 00:16:44.266 <v Speaker1>convince a customer is a clear, compelling demonstration of value and a positive 00:16:44.266 --> 00:16:45.526 <v Speaker1>return on their investment. 00:16:45.866 --> 00:16:50.986 <v Speaker1>So your pitch and your decks must really focus on the tangible business benefits you deliver. 00:16:51.406 --> 00:16:54.906 <v Speaker1>So customers want to know if you're reducing their manual processes, 00:16:55.406 --> 00:16:57.626 <v Speaker1>saving them money or increasing their revenue. 00:16:57.966 --> 00:17:03.086 <v Speaker1>And these factors directly impact their bottom line and are what they truly care about. 00:17:03.806 --> 00:17:08.326 <v Speaker1>Secondly, customers are convinced by the results that they can see with their own eyes. 00:17:08.346 --> 00:17:12.546 <v Speaker1>So they need to know that your product can deliver on its promises and that 00:17:12.546 --> 00:17:14.806 <v Speaker1>it is accurate, scalable and cost effective. 00:17:15.246 --> 00:17:19.506 <v Speaker1>And these factors then ultimately build their confidence in your solution and 00:17:19.506 --> 00:17:22.286 <v Speaker1>give them the evidence they need to justify the investment. 00:17:23.206 --> 00:17:28.106 <v Speaker1>Lastly, also the price metric has to be tied to the value you provide. 00:17:28.386 --> 00:17:33.446 <v Speaker1>So charging per completed task or a direct outcome rather than per user makes 00:17:33.446 --> 00:17:36.306 <v Speaker1>the value proposition very clear to the customer. 00:17:36.426 --> 00:17:40.066 <v Speaker1>So it shows them that they are paying for a solution that delivers a measurable 00:17:40.066 --> 00:17:42.086 <v Speaker1>result, but not just a tool. 00:17:42.286 --> 00:17:46.846 <v Speaker0>Not sure where I've seen this. Maybe even in your DAX, Canvas, 00:17:47.086 --> 00:17:51.666 <v Speaker0>AI Feature, Upsells are well-known case studies. 00:17:51.666 --> 00:17:58.466 <v Speaker0>What can SaaS founders learn from, for example, how Canva monetizes Gen AI? 00:17:59.717 --> 00:18:04.457 <v Speaker1>So the core lesson from Canva is really to avoid building a separate AI product. 00:18:04.797 --> 00:18:10.657 <v Speaker1>So instead, they embedded Gen AI capabilities directly into the core offering as a value booster. 00:18:11.077 --> 00:18:16.857 <v Speaker1>So to be more specific here, Canva doesn't put all its AI features behind a paywall from the start. 00:18:16.857 --> 00:18:22.497 <v Speaker1>So they give users a limited number of credits to use features like Magic Eraser 00:18:22.497 --> 00:18:24.857 <v Speaker1>or Magic Edit for free for their presentations. 00:18:24.857 --> 00:18:30.037 <v Speaker1>And this freemium approach allows users to experience what one source calls 00:18:30.037 --> 00:18:33.657 <v Speaker1>the AI-managed JIC without any upfront commitment. 00:18:33.917 --> 00:18:39.377 <v Speaker1>So this strategy is highly effective because it demonstrates the value proposition firsthand. 00:18:39.537 --> 00:18:44.857 <v Speaker1>So if you take a designer or a small business owner, if they are able to remove 00:18:44.857 --> 00:18:49.917 <v Speaker1>a background or generate an image, that can save 15 to 20 minutes on a single task. 00:18:50.097 --> 00:18:54.597 <v Speaker1>And that's also what I call often the 10 times value. So it means that for a 00:18:54.597 --> 00:18:59.417 <v Speaker1>customer to justify the cost, the risk, and the effort of adopting a new solution, 00:18:59.657 --> 00:19:03.877 <v Speaker1>the value it provides has to be at least 10 times greater than the price they pay. 00:19:04.097 --> 00:19:09.077 <v Speaker1>So in this case, once the user has experienced that level of value and hits 00:19:09.077 --> 00:19:12.137 <v Speaker1>their credit limit, the decision to upgrade is easy. 00:19:12.517 --> 00:19:17.417 <v Speaker1>The cost of a paid subscription, perhaps take a few dollars a month, 00:19:17.617 --> 00:19:21.697 <v Speaker1>becomes insignificant compared to the hours of work that they have saved. 00:19:21.697 --> 00:19:26.257 <v Speaker1>And so they're not just paying for a tool, they're paying to unlock the value 00:19:26.257 --> 00:19:27.977 <v Speaker1>they've already experienced. 00:19:28.417 --> 00:19:33.557 <v Speaker1>And another critical lesson from Canva is the need to align your pricing model 00:19:33.557 --> 00:19:35.537 <v Speaker1>with the underlying cost of GenEI. 00:19:36.017 --> 00:19:41.017 <v Speaker1>So unlike traditional model, traditional software, every interaction with a 00:19:41.017 --> 00:19:45.357 <v Speaker1>GenEI model, as we said earlier, has a real tangible compute cost. 00:19:45.637 --> 00:19:51.397 <v Speaker1>So Canva's credit-based system is a perfect example of a usage-based pricing model. 00:19:51.697 --> 00:19:57.337 <v Speaker1>Which is then again crucial for monetizing AI and SaaS. It really directly matches 00:19:57.337 --> 00:19:59.517 <v Speaker1>the model's cost with the value delivered. 00:20:00.057 --> 00:20:04.117 <v Speaker1>And it really ensures that the compute-heavy features remain a profitable part 00:20:04.117 --> 00:20:05.857 <v Speaker1>of the business as it scales on. 00:20:06.297 --> 00:20:11.237 <v Speaker1>So by charging for usage after a certain free limit, companies can manage their 00:20:11.237 --> 00:20:16.337 <v Speaker1>unit economics and ensure that the most engaged high-usage customers are also 00:20:16.337 --> 00:20:17.777 <v Speaker1>their most profitable ones. 00:20:17.777 --> 00:20:22.457 <v Speaker1>So this strategy prevents the cost paradox that I discussed earlier, 00:20:22.737 --> 00:20:27.537 <v Speaker1>where a flat fee, unlimited usage model can become financially unsustainable 00:20:27.537 --> 00:20:29.177 <v Speaker1>as customer engagement increases. 00:20:29.537 --> 00:20:34.237 <v Speaker1>So this approach really offers flexibility and a lower barrier to entry for 00:20:34.237 --> 00:20:38.017 <v Speaker1>customers who want to test the product without a large upfront commitment, 00:20:38.237 --> 00:20:42.457 <v Speaker1>while also still at the same time providing a clear path to monetization for 00:20:42.457 --> 00:20:44.817 <v Speaker1>high volume users. So really a great use case. 00:20:44.817 --> 00:20:52.157 <v Speaker0>When listening to this, I was wondering, how do you help startups connect operational savings? 00:20:53.109 --> 00:20:58.689 <v Speaker0>For example, like faster development to hard revenue metrics that resonate with, 00:20:58.849 --> 00:21:01.249 <v Speaker0>that resonates, for example, with boards or VCs. 00:21:01.629 --> 00:21:05.909 <v Speaker1>And that's a challenge really that I see a lot of the times when talking to startups. 00:21:06.049 --> 00:21:09.629 <v Speaker1>And it's a major hurdle, generally speaking, on the path to production. 00:21:09.909 --> 00:21:13.909 <v Speaker1>So you can't just go into a board meeting and say, we are building things faster. 00:21:14.149 --> 00:21:17.809 <v Speaker1>So you really have to speak the language of the board, which is revenue, 00:21:18.009 --> 00:21:22.309 <v Speaker1>valuation and return on investment. So the key here is to build a bridge between 00:21:22.309 --> 00:21:26.049 <v Speaker1>your operational metrics and your business financial health. 00:21:26.489 --> 00:21:31.149 <v Speaker1>So let's explore a three-step framework for how you can do just that. 00:21:31.629 --> 00:21:36.629 <v Speaker1>So first, you have to get granular and define the current reality and the future 00:21:36.629 --> 00:21:39.829 <v Speaker1>reality, as you can't improve what you can't measure, obviously. 00:21:40.309 --> 00:21:45.029 <v Speaker1>So the first step here is to define the exact time or resource you're saving. 00:21:45.029 --> 00:21:49.809 <v Speaker1>For example, a startup might be able to shorten its delivery timeline from 12 00:21:49.809 --> 00:21:54.869 <v Speaker1>months to just two months by using AI, or reduce the time for a machine learning 00:21:54.869 --> 00:21:57.729 <v Speaker1>model to go live from six months to under two weeks. 00:21:57.949 --> 00:22:02.429 <v Speaker1>So you have to define these metrics first, whether it's lines of code written 00:22:02.429 --> 00:22:06.569 <v Speaker1>per day, bugs found per sprint, or the number of support tickets that are handled, 00:22:06.689 --> 00:22:07.969 <v Speaker1>for instance, by an AI agent. 00:22:08.189 --> 00:22:12.429 <v Speaker1>So this establishes your baseline and provides a clear point of comparison. 00:22:12.429 --> 00:22:17.929 <v Speaker1>And once you have gathered all this operational data, you also need to connect it to your finances. 00:22:18.329 --> 00:22:22.789 <v Speaker1>So this is really where you translate the what into the so what for investors. 00:22:23.089 --> 00:22:27.729 <v Speaker1>So, for instance, faster development means fewer engineering hours on a project, 00:22:27.889 --> 00:22:30.829 <v Speaker1>which really translates to reduced personal cost. 00:22:30.969 --> 00:22:34.249 <v Speaker1>And this improves your burn rate and extends your cash runway, 00:22:34.509 --> 00:22:39.209 <v Speaker1>both of which are critical metrics for VCs. And when you can develop and launch 00:22:39.209 --> 00:22:42.929 <v Speaker1>new features faster, you are also accelerating your time to market. 00:22:43.149 --> 00:22:47.049 <v Speaker1>And this means you can start generating revenue from those features sooner. 00:22:47.209 --> 00:22:51.869 <v Speaker1>And it also means you can respond to competitive threats and customer demands 00:22:51.869 --> 00:22:55.169 <v Speaker1>more quickly, which can lead to a higher customer satisfaction, 00:22:55.589 --> 00:23:00.489 <v Speaker1>reduced churn and ultimately what we discussed earlier, a higher customer lifetime value. 00:23:01.356 --> 00:23:07.536 <v Speaker1>And finally, you have to package the story in a way that resonates with VCs and board members. 00:23:07.756 --> 00:23:12.936 <v Speaker1>So this is where you bring the operational and financial metrics together in 00:23:12.936 --> 00:23:14.396 <v Speaker1>a single powerful statement. 00:23:14.736 --> 00:23:19.256 <v Speaker1>For example, you can say, by reducing our development cycle by 50%, 00:23:19.256 --> 00:23:24.076 <v Speaker1>we were able to launch Feature X, for instance, a quarter ahead of schedule, 00:23:24.316 --> 00:23:26.796 <v Speaker1>which is already contributing, let's say, 00:23:27.156 --> 00:23:29.936 <v Speaker1>200K in new monthly recurring revenue. 00:23:29.936 --> 00:23:34.596 <v Speaker1>And this is a repeatable process that will allow us to continuously improve 00:23:34.596 --> 00:23:39.256 <v Speaker1>our product and our go-to-market efficiency, which then in turn will reduce 00:23:39.256 --> 00:23:42.996 <v Speaker1>our customer acquisition cost and increase our customer lifetime value. 00:23:43.236 --> 00:23:47.636 <v Speaker1>And if you follow this framework, you're not just selling a product or technology, 00:23:47.636 --> 00:23:52.836 <v Speaker1>you're really selling a quantifiable financial improvement that directly impacts 00:23:52.836 --> 00:23:55.436 <v Speaker1>your company's valuation and growth potential. 00:23:55.436 --> 00:23:58.916 <v Speaker1>And this is really the language that boards and investors understand. 00:23:59.056 --> 00:24:04.356 <v Speaker1>And it's the key also in terms of fundraising based on what I've seen so far. 00:24:07.256 --> 00:24:13.536 <v Speaker0>Guys, we'll be back after a short ad break talking about monetization models here. 00:24:18.376 --> 00:24:23.116 <v Speaker0>Guys, welcome back to our interview with Jennifer from AWS. 00:24:24.396 --> 00:24:28.436 <v Speaker0>AWS works with a lot of very complex customers. 00:24:29.056 --> 00:24:32.716 <v Speaker0>That's why we decided to pick a very big example. Jennifer, 00:24:32.876 --> 00:24:37.256 <v Speaker0>can you, for example, share how Pfizer used Gen.AI for scientific applications 00:24:37.256 --> 00:24:45.596 <v Speaker0>to unlock something around $1 billion annual savings and what that teaches early-stage 00:24:45.596 --> 00:24:48.976 <v Speaker0>founders about return on investment storytelling? Erling? 00:24:50.196 --> 00:24:54.576 <v Speaker1>That's really a fantastic example because it perfectly illustrates the power 00:24:54.576 --> 00:24:59.076 <v Speaker1>of moving beyond a simple proof of concept and into a production mindset with 00:24:59.076 --> 00:25:01.296 <v Speaker1>a clear focus on tangible business value. 00:25:01.536 --> 00:25:06.516 <v Speaker1>So the core of the Pfizer story is that they apply Gen.E.I. to their most complex 00:25:06.516 --> 00:25:08.736 <v Speaker1>and time-consuming internal processes. 00:25:09.036 --> 00:25:12.236 <v Speaker1>So let's think about scientific research and development. 00:25:12.576 --> 00:25:17.676 <v Speaker1>So if you imagine developing a single drug can generate over 20,000 documents 00:25:17.676 --> 00:25:23.196 <v Speaker1>and working through this information manually is an enormous drain on a scientist's time. 00:25:23.476 --> 00:25:28.416 <v Speaker1>So in the Pfizer-Amazon collaboration team, or PACT for short, 00:25:28.696 --> 00:25:33.876 <v Speaker1>we implemented rapid prototyping with a fail-fast cycle. So this typically takes 00:25:33.876 --> 00:25:38.456 <v Speaker1>no more than six weeks for a project that would have taken mums to do internally. 00:25:39.408 --> 00:25:43.428 <v Speaker1>So their first high ROI use case was data discovery. 00:25:43.688 --> 00:25:47.848 <v Speaker1>They built an internal platform called Box using generative AI, 00:25:48.108 --> 00:25:49.528 <v Speaker1>which allowed scientists to 00:25:49.528 --> 00:25:53.588 <v Speaker1>search a massive repository of documents using natural language queries. 00:25:53.948 --> 00:25:59.348 <v Speaker1>So Pfizer estimates this has the potential to save its scientists up to 16,000 00:25:59.348 --> 00:26:04.568 <v Speaker1>hours of search time annually and reduce infrastructure costs by 55%. 00:26:04.568 --> 00:26:09.748 <v Speaker1>However, they didn't stop there with us. So they also applied machine learning 00:26:09.748 --> 00:26:13.628 <v Speaker1>and Gen AI to other critical processes like manufacturing. 00:26:13.928 --> 00:26:18.968 <v Speaker1>So by using services like Amazon SageMaker, they developed a prototype to detect 00:26:18.968 --> 00:26:23.768 <v Speaker1>anomalies in their continuous manufacturing processes, which also helps predict 00:26:23.768 --> 00:26:26.728 <v Speaker1>maintenance needs and reduce equipment downtime. 00:26:27.048 --> 00:26:32.668 <v Speaker1>So this is really a perfect example of applying AI to a valuable and unique business process. 00:26:32.748 --> 00:26:37.368 <v Speaker1>So in this case, ensuring the quality and consistency of drug production. 00:26:38.408 --> 00:26:42.488 <v Speaker1>And especially if we tie it back to the learnings for early stage founders, 00:26:42.848 --> 00:26:46.528 <v Speaker1>it isn't the specific technology or the scale of the savings. 00:26:46.528 --> 00:26:48.328 <v Speaker1>It's really the story that they told. 00:26:48.628 --> 00:26:54.768 <v Speaker1>So the $70, $50 million to $1 billion in annual savings, that wasn't just a number for them. 00:26:54.908 --> 00:27:00.508 <v Speaker1>So it was the result of a clear, cohesive narrative that allowed them to redirect 00:27:00.508 --> 00:27:04.908 <v Speaker1>resources, shorten development cycles, and ultimately get their products to 00:27:04.908 --> 00:27:06.568 <v Speaker1>patients and the market sooner. 00:27:07.508 --> 00:27:12.308 <v Speaker1>And to tie it back to what we said earlier, Pfizer didn't just tell its stakeholders, 00:27:12.608 --> 00:27:14.548 <v Speaker1>our research and development is faster. 00:27:14.768 --> 00:27:19.648 <v Speaker1>They said, because our research and development is faster, we can bring new 00:27:19.648 --> 00:27:26.128 <v Speaker1>medicine to market sooner, which will unlock this much annual revenue and cost savings. 00:27:26.488 --> 00:27:32.028 <v Speaker1>So they connected really an internal operational gain to a clear external business outcome. 00:27:32.328 --> 00:27:35.608 <v Speaker1>So for Founder, this is the blueprint for a compelling pitch. 00:27:35.608 --> 00:27:39.788 <v Speaker1>So you need to show your board investors how an operational improvement, 00:27:40.088 --> 00:27:44.808 <v Speaker1>imagine what we said earlier, the past development cycles, translates into a 00:27:44.808 --> 00:27:48.708 <v Speaker1>higher customer lifetime value or lower customer acquisition cost. 00:27:49.108 --> 00:27:53.308 <v Speaker1>So by focusing your generative AI application and your most valuable and unique 00:27:53.308 --> 00:27:58.048 <v Speaker1>business processes, you create actually a snowball effect of value that becomes 00:27:58.048 --> 00:27:59.368 <v Speaker1>central to how you operate. 00:28:00.068 --> 00:28:05.148 <v Speaker1>Also, you should apply Gen AI to your most valuable and unique business processes. 00:28:05.608 --> 00:28:08.248 <v Speaker1>For Pfizer, in that case, that was research and development. 00:28:08.268 --> 00:28:11.808 <v Speaker1>For a SaaS company, it might be your product development, your go-to-market 00:28:11.808 --> 00:28:13.368 <v Speaker1>strategy or your customer support. 00:28:13.708 --> 00:28:18.368 <v Speaker1>So really, the production mindset is about focusing on these areas to create 00:28:18.368 --> 00:28:22.828 <v Speaker1>a snowball effect of value that becomes really the fabric of how you operate. 00:28:23.228 --> 00:28:28.608 <v Speaker0>For our audience, you've seen startups crash because they scale Gen.AI features 00:28:28.608 --> 00:28:30.288 <v Speaker0>without a monetization plan. 00:28:30.748 --> 00:28:35.928 <v Speaker0>Could you share a raw story of what went wrong? The full story is in the founder's fault. 00:28:36.879 --> 00:28:40.919 <v Speaker1>Unfortunately, due to kind of the confidential nature of this question, 00:28:41.039 --> 00:28:42.519 <v Speaker1>I can't really answer to that. 00:28:42.619 --> 00:28:46.619 <v Speaker1>But I've definitely observed a number of startups that initially started without 00:28:46.619 --> 00:28:51.859 <v Speaker1>a clear monetization plan, but where I actually helped them to develop that. 00:28:52.059 --> 00:28:53.299 <v Speaker1>So, for instance, we've been doing. 00:28:54.259 --> 00:28:58.819 <v Speaker0>Jennifer, we don't need that. It's just for our audience. Okay. Okay. 00:28:59.819 --> 00:29:03.359 <v Speaker0>Jennifer, let's talk a little bit about monetization models here. 00:29:04.199 --> 00:29:11.219 <v Speaker0>Do you see startups better off charging for AI as a feature added to an existing 00:29:11.219 --> 00:29:16.199 <v Speaker0>SaaS or a standalone enabler or like a new product line? 00:29:16.459 --> 00:29:20.979 <v Speaker1>That's a really great question and a very critical one for any founder, 00:29:21.219 --> 00:29:23.199 <v Speaker1>especially as they start off in scale. 00:29:23.439 --> 00:29:27.579 <v Speaker1>So it really depends on your business, the market and the problem you're solving. 00:29:27.839 --> 00:29:32.939 <v Speaker1>So let's start with Gen AI as a feature. That is often the safest and most effective strategy. 00:29:33.359 --> 00:29:37.679 <v Speaker1>So you should use it when you have a strong existing user base and a clear use 00:29:37.679 --> 00:29:40.019 <v Speaker1>case for AI to improve a current workflow. 00:29:41.029 --> 00:29:45.829 <v Speaker1>So instead of building a new product from scratch, you embed AI directly into 00:29:45.829 --> 00:29:48.129 <v Speaker1>your existing offering to enhance its value. 00:29:48.309 --> 00:29:51.849 <v Speaker1>So you typically can't charge a premium for a single feature, 00:29:52.009 --> 00:29:57.129 <v Speaker1>but instead you can use tiered pricing, offering AI features in higher price 00:29:57.129 --> 00:30:01.229 <v Speaker1>plans, or a freemium model where basic AI functions are free. 00:30:01.369 --> 00:30:05.989 <v Speaker1>But advanced capabilities, imagine more queries or larger outputs, 00:30:06.189 --> 00:30:10.309 <v Speaker1>are meted with a credit system. So, the good example is Canvas Magic Studio 00:30:10.309 --> 00:30:11.529 <v Speaker1>that we discussed before. 00:30:11.769 --> 00:30:17.029 <v Speaker1>So, it's not a new separate product. It's a suite of tools that makes the existing 00:30:17.029 --> 00:30:18.769 <v Speaker1>design process faster and better. 00:30:19.189 --> 00:30:23.829 <v Speaker1>It's also a great way to get customer feedback for iterative experimentation 00:30:23.829 --> 00:30:25.489 <v Speaker1>and continuous improvement. 00:30:26.009 --> 00:30:30.469 <v Speaker1>So, the main advantage here are that it reduces the friction of adoption because 00:30:30.469 --> 00:30:32.469 <v Speaker1>customers are already in your product. 00:30:32.709 --> 00:30:36.809 <v Speaker1>And the value is immediately clear and it can be a powerful upsell driver. 00:30:36.809 --> 00:30:40.669 <v Speaker1>On the other hand, you can't charge a premium for a single feature, 00:30:40.729 --> 00:30:44.109 <v Speaker1>so your monetization has to be tied to a different value metric. 00:30:44.429 --> 00:30:48.049 <v Speaker1>Imagine a premium model with a credits or a tiered pricing structure. 00:30:48.389 --> 00:30:52.809 <v Speaker1>If you move to the second part of your question, so generative AI is a standalone 00:30:52.809 --> 00:30:57.309 <v Speaker1>enabler, it is the path often I see with startups that are building foundational 00:30:57.309 --> 00:30:59.049 <v Speaker1>technology for other developers. 00:30:59.049 --> 00:31:03.889 <v Speaker1>So you're not selling a finished product. You're selling, let's call it the 00:31:03.889 --> 00:31:07.489 <v Speaker1>picks and shovels for others to build their own AI applications. 00:31:08.029 --> 00:31:14.469 <v Speaker1>So this is when your core intellectual property is in the AI model or a specific repeatable AI process. 00:31:14.829 --> 00:31:19.669 <v Speaker1>So you're giving other companies the ability to add product innovation to their 00:31:19.669 --> 00:31:21.829 <v Speaker1>own platforms by integrating your API. 00:31:22.509 --> 00:31:26.409 <v Speaker1>The main advantage I see here is that this model is highly scalable. 00:31:26.409 --> 00:31:31.229 <v Speaker1>So your market could be potentially every developer or company that needs your 00:31:31.229 --> 00:31:32.769 <v Speaker1>specific AI capabilities. 00:31:33.189 --> 00:31:37.729 <v Speaker1>And it can lead to rapid adoption and significant revenue without the overhead 00:31:37.729 --> 00:31:39.929 <v Speaker1>of building a complete end-user product. 00:31:40.741 --> 00:31:45.821 <v Speaker1>The most common approach here is usage-based pricing. So you can charge here 00:31:45.821 --> 00:31:49.681 <v Speaker1>per API call, per token, or per unit of data processed. 00:31:50.081 --> 00:31:54.861 <v Speaker1>It's also known as the bait-as-you-go model. So it aligns the cost directly 00:31:54.861 --> 00:31:57.881 <v Speaker1>with the value a developer gets from your technology. 00:31:58.501 --> 00:32:02.621 <v Speaker1>And this is also where you might see a hybrid model with a fixed subscription 00:32:02.621 --> 00:32:08.761 <v Speaker1>fee for a certain level of usage and an additional usage fee once a customer surpasses that usage. 00:32:08.761 --> 00:32:14.581 <v Speaker1>So while this model is very scalable, it faces also direct competition from 00:32:14.581 --> 00:32:19.161 <v Speaker1>big companies like OpenAI, and they can have less predictable revenue streams 00:32:19.161 --> 00:32:20.401 <v Speaker1>than a flat subscription. 00:32:20.921 --> 00:32:27.041 <v Speaker1>And for me, this is a model for companies with a very unique or highly specialized AI capability. 00:32:27.401 --> 00:32:32.501 <v Speaker1>To give you an example of a startup that I worked with over here is DeepSet. 00:32:32.501 --> 00:32:38.761 <v Speaker1>So they don't create the end user facing product, but they build the tools, 00:32:39.141 --> 00:32:44.321 <v Speaker1>infrastructure and platforms that other companies use to create their own Gen AI applications. 00:32:45.021 --> 00:32:49.781 <v Speaker1>So DeepSat's business model is based on an open source or freemium model combining 00:32:49.781 --> 00:32:54.961 <v Speaker1>their open source framework haystack with a commercial SaaS platform called DeepSat AI Platform. 00:32:54.961 --> 00:33:00.501 <v Speaker1>So this platform provides enterprise-grade features and support that goes beyond 00:33:00.501 --> 00:33:04.341 <v Speaker1>the free open-source version, including managed infrastructure, 00:33:04.581 --> 00:33:08.981 <v Speaker1>advanced tools, and enterprise-grade security, as well as dedicated support. 00:33:09.541 --> 00:33:14.241 <v Speaker1>And then, closing the loop to the new product line. For me, this is really the 00:33:14.241 --> 00:33:15.601 <v Speaker1>most ambitious strategy. 00:33:16.001 --> 00:33:20.641 <v Speaker1>So you're building brand-new end-to-end products with Gen.AI at its core. 00:33:21.281 --> 00:33:26.321 <v Speaker1>So when the AI provides a fundamentally new experience or solves a problem that 00:33:26.321 --> 00:33:31.001 <v Speaker1>couldn't be solved before, the AI really isn't an add-on. It's the product itself. 00:33:31.381 --> 00:33:35.461 <v Speaker1>So this approach offers the highest potential for revenue and market disruption. 00:33:35.501 --> 00:33:40.561 <v Speaker1>So if you succeed, you can create a completely new market category and build 00:33:40.561 --> 00:33:43.281 <v Speaker1>a defensible model around your business. 00:33:44.649 --> 00:33:48.389 <v Speaker1>Nevertheless, it carries the highest risk as well, because you have to solve 00:33:48.389 --> 00:33:52.769 <v Speaker1>a real problem, build a product from scratch, and also deal with the high cost 00:33:52.769 --> 00:33:56.229 <v Speaker1>per inference and unit economics of running complex AI models. 00:33:56.529 --> 00:34:01.529 <v Speaker1>You also need to get to market before your end of runway, which is a key metric for startups. 00:34:02.069 --> 00:34:06.769 <v Speaker1>An example of a new product line in this context would be our partner Entropic. 00:34:07.249 --> 00:34:10.189 <v Speaker1>So as you might have used their cloud model as well. 00:34:10.189 --> 00:34:14.169 <v Speaker1>So rather than being an enabler for others to build applications, 00:34:15.069 --> 00:34:19.609 <v Speaker1>Antropic's business model is really centered on creating and selling the core, 00:34:20.029 --> 00:34:23.989 <v Speaker1>let's call it going back to the analogy, picks and shovels themselves. 00:34:24.189 --> 00:34:29.189 <v Speaker1>So they create and continuously improve their co-product, in this case, 00:34:29.329 --> 00:34:31.309 <v Speaker1>the Claude family of large language models. 00:34:32.069 --> 00:34:36.349 <v Speaker1>So Entropic monetizes its models through various tiers, offering different levels 00:34:36.349 --> 00:34:40.529 <v Speaker1>of intelligence, speed, and cost to meet a range of user needs. 00:34:40.849 --> 00:34:45.309 <v Speaker1>So these models are offered on a tiered pricing model, often based on token 00:34:45.309 --> 00:34:49.089 <v Speaker1>usage with different plans for individuals, teams, and enterprises. 00:34:49.709 --> 00:34:54.209 <v Speaker1>As you might know, Entropic also created specialized products like CloudCode, 00:34:54.409 --> 00:34:58.329 <v Speaker1>which is specifically designed for developers and their workflows. 00:34:58.329 --> 00:35:04.529 <v Speaker1>So this is a form of new product features that are aimed at a specific customer segment. 00:35:05.149 --> 00:35:09.789 <v Speaker1>So Entropic's strategy involves making their models available through multiple 00:35:09.789 --> 00:35:14.469 <v Speaker1>channels, including their own web interface, direct API access, 00:35:14.649 --> 00:35:16.969 <v Speaker1>and partnerships with cloud providers like us. 00:35:17.129 --> 00:35:21.609 <v Speaker1>So this ensures that their product can reach a wide range of customers from 00:35:21.609 --> 00:35:23.789 <v Speaker1>individual developers to large enterprises. 00:35:23.789 --> 00:35:28.249 <v Speaker1>So, this approach has the highest potential for market disruption revenue, 00:35:28.249 --> 00:35:33.369 <v Speaker1>but it also carries the highest risk and requires significant investment in 00:35:33.369 --> 00:35:36.989 <v Speaker1>building a new product from the ground up, as well as, as I mentioned, 00:35:37.209 --> 00:35:40.729 <v Speaker1>managing that high cost per inference of running complex model. 00:35:41.009 --> 00:35:46.349 <v Speaker1>So, really, to answer the question, the best approach depends on your specific circumstances. 00:35:46.349 --> 00:35:50.709 <v Speaker1>So, for most startups, especially those that already have an existing customer 00:35:50.709 --> 00:35:55.389 <v Speaker1>base, adding AI as a feature is a great way to start because it offers a clear 00:35:55.389 --> 00:35:58.589 <v Speaker1>path to value realization. But it really depends where you're standing. 00:36:01.469 --> 00:36:09.809 <v Speaker0>What role do usage-based versus subscription monetization models play in AI? 00:36:10.009 --> 00:36:13.349 <v Speaker0>And how do you get founders in choosing between them? 00:36:14.869 --> 00:36:19.169 <v Speaker1>That's really, for me, a fundamental question that I like to take very early 00:36:19.169 --> 00:36:22.489 <v Speaker1>because it's one of the most critical ones that a founder can make. 00:36:22.669 --> 00:36:27.449 <v Speaker1>So it's not really just about a pricing model. It's a statement about your company 00:36:27.449 --> 00:36:31.009 <v Speaker1>values and how you see your relationship with your customers. 00:36:31.509 --> 00:36:35.949 <v Speaker1>So in the world of AI, this decision is even more critical because it directly 00:36:35.949 --> 00:36:39.129 <v Speaker1>ties to the unique unit economics of your product. 00:36:39.929 --> 00:36:43.949 <v Speaker1>So what we're seeing in the market is that successful markets are moving beyond 00:36:43.949 --> 00:36:48.969 <v Speaker1>the traditional flat fee subscription and even the pure usage-based model to 00:36:48.969 --> 00:36:50.349 <v Speaker1>embrace a hybrid approach. 00:36:50.489 --> 00:36:52.829 <v Speaker1>So this gives you the best of both worlds. 00:36:52.989 --> 00:36:57.729 <v Speaker1>So you can start with a predictable stable subscription that includes a certain 00:36:57.729 --> 00:37:01.049 <v Speaker1>amount of AI credits or usage, as we've seen in the Canva example. 00:37:01.689 --> 00:37:06.669 <v Speaker1>And then once a customer becomes a powerful user and exceeds that included amount, 00:37:06.929 --> 00:37:09.729 <v Speaker1>you automatically move them to a usage-based model. 00:37:09.989 --> 00:37:13.889 <v Speaker1>And this really ensures that you're compensated for the value you provide. 00:37:13.889 --> 00:37:18.429 <v Speaker1>And it also helps manage the high variable cost of your AI workload that we've 00:37:18.429 --> 00:37:20.789 <v Speaker1>just seen in what I shared earlier. 00:37:22.149 --> 00:37:28.069 <v Speaker1>And this hybrid approach also opens up a number of specific monetization avenues 00:37:28.069 --> 00:37:31.029 <v Speaker1>for founders that have Gen.E.I. 00:37:31.209 --> 00:37:35.849 <v Speaker1>As the end product, where you can, for instance, charge per image generated or per query answered. 00:37:36.449 --> 00:37:41.069 <v Speaker1>And on the other hand side, Gen.E.I. is a super tier means that for an existing 00:37:41.069 --> 00:37:45.609 <v Speaker1>product, you can create a higher price subscription tier that offers exclusive 00:37:45.609 --> 00:37:47.829 <v Speaker1>access to your advanced AI capabilities. 00:37:47.829 --> 00:37:54.129 <v Speaker1>And this really is helpful for your most engaged users to give them a clear path to get more value. 00:37:54.429 --> 00:37:58.789 <v Speaker1>And you also get to charge a premium for it, just like we discussed in the Canva example. 00:37:59.529 --> 00:38:02.449 <v Speaker1>Another possibility is Gen AI as a value booster. 00:38:02.709 --> 00:38:06.449 <v Speaker1>So in this model, the AI functionality enhances an existing product, 00:38:06.629 --> 00:38:08.549 <v Speaker1>but it's not the central feature. 00:38:09.029 --> 00:38:13.729 <v Speaker1>So the AI adds value to the core service, which makes it more efficient or powerful, 00:38:13.969 --> 00:38:18.729 <v Speaker1>which can justify a higher over price point. A good example which many startups 00:38:18.729 --> 00:38:20.269 <v Speaker1>are using is Slack, for instance. 00:38:21.289 --> 00:38:26.549 <v Speaker1>And lastly, Gen AI as an add-on means that the AI feature is an optional component 00:38:26.549 --> 00:38:28.449 <v Speaker1>that customers can pay for separately. 00:38:28.729 --> 00:38:33.709 <v Speaker1>So this lowers the barrier to entry for your core product while giving you still 00:38:33.709 --> 00:38:39.269 <v Speaker1>an opportunity to upsell and monetize those who want to use your AI capabilities. 00:38:40.544 --> 00:38:43.924 <v Speaker1>But for me, it's also important, apart from the monetization strategy, 00:38:43.924 --> 00:38:46.224 <v Speaker1>to look at the infrastructure supporting it. 00:38:46.544 --> 00:38:50.164 <v Speaker1>So really, it's about managing the variable cost of AI workloads. 00:38:50.624 --> 00:38:56.244 <v Speaker1>So when a founder is building their platform, they have to choose a lot of different things. 00:38:56.424 --> 00:39:01.324 <v Speaker1>So, for instance, the right inference consumption strategy, which directly impacts 00:39:01.324 --> 00:39:03.144 <v Speaker1>their ability to scale profitability. 00:39:03.864 --> 00:39:08.244 <v Speaker1>And there are many options here. So, for instance, there's the pay-as-you-go 00:39:08.244 --> 00:39:09.424 <v Speaker1>model with no commitment. 00:39:09.424 --> 00:39:13.604 <v Speaker1>It's great for prototyping proof of concepts and small workloads because it's 00:39:13.604 --> 00:39:19.624 <v Speaker1>so flexible, but it can mean that you have higher latency because you're using shared resources. 00:39:20.224 --> 00:39:25.104 <v Speaker1>If you want to go for consistent production workloads with a predictable need 00:39:25.104 --> 00:39:28.624 <v Speaker1>for large scale processing, go for provision throughput. 00:39:28.864 --> 00:39:33.564 <v Speaker1>It's redesigned to process big workloads at scale, but it can also be expensive 00:39:33.564 --> 00:39:37.404 <v Speaker1>if your usage has low utilization during off peak hours. 00:39:38.164 --> 00:39:42.784 <v Speaker1>And then another one that I see a lot of startups use is batch inference. 00:39:43.064 --> 00:39:47.424 <v Speaker1>So it's really great for asynchronous workloads, like imagine processing large 00:39:47.424 --> 00:39:49.924 <v Speaker1>documents or conducting offline experiments. 00:39:50.444 --> 00:39:54.864 <v Speaker1>It's a lot more cost effective, up to 50 percent cheaper than on demand. 00:39:55.044 --> 00:39:59.124 <v Speaker1>But it's not, of course, suitable for every application that need real time responses. 00:39:59.844 --> 00:40:05.824 <v Speaker1>So really, it depends, as always, for me, the successful founders are really 00:40:05.824 --> 00:40:09.404 <v Speaker1>the ones who think about this from both a business and a technical perspective. 00:40:09.704 --> 00:40:14.724 <v Speaker1>So they don't just pick a pricing model. So they align it with a smart, 00:40:15.144 --> 00:40:19.004 <v Speaker1>cost-effective infrastructure strategy, which allows their business to scale 00:40:19.004 --> 00:40:22.664 <v Speaker1>profitably as their customers find more value in their product. 00:40:23.764 --> 00:40:32.064 <v Speaker0>We know startups work a lot with examples, with tests, with their own set-up, 00:40:32.104 --> 00:40:36.284 <v Speaker0>so how important is pricing experimentation? 00:40:36.704 --> 00:40:45.084 <v Speaker0>Can Gen.ai itself help optimize pricing points for AI products based on promotion data? 00:40:46.171 --> 00:40:51.551 <v Speaker1>Yeah, you hit it here on a really great concept that I quite often discuss now with founders. 00:40:51.791 --> 00:40:56.631 <v Speaker1>So really, as a founder, your job is never truly done when it comes to pricing. 00:40:56.851 --> 00:41:01.471 <v Speaker1>Though there's always a need for iterative experimentation, how we call it. 00:41:01.651 --> 00:41:07.051 <v Speaker1>So it's about being outcome oriented and not being afraid to act and test potential 00:41:07.051 --> 00:41:09.111 <v Speaker1>solutions, even if they are not perfect. 00:41:09.771 --> 00:41:13.951 <v Speaker1>So this is particularly true for Generative AI because the value proposition 00:41:13.951 --> 00:41:17.611 <v Speaker1>and the customer willingness to pay can be unclear at the beginning. 00:41:17.911 --> 00:41:22.291 <v Speaker1>So you have to constantly learn and adapt based on how people actually use your 00:41:22.291 --> 00:41:27.691 <v Speaker1>product, what your customers truly value and how they're willing to pay for it. 00:41:28.031 --> 00:41:32.191 <v Speaker1>And for me, this is where Gen.EI can be a good sparing partner, 00:41:32.331 --> 00:41:34.031 <v Speaker1>the game changer for pricing itself. 00:41:34.731 --> 00:41:39.631 <v Speaker1>So we just discussed how to price an AI product. But Gen AI can also be used 00:41:39.631 --> 00:41:42.691 <v Speaker1>as a powerful tool to determine the optimal pricing. 00:41:43.051 --> 00:41:47.931 <v Speaker1>So imagine if you have Gen AI models that can process and analyze vast amount 00:41:47.931 --> 00:41:52.051 <v Speaker1>of data from your promotions, user interactions, and even market trends. 00:41:52.171 --> 00:41:57.651 <v Speaker1>So it can define and identify patterns that a human analyst might miss. 00:41:57.651 --> 00:42:03.751 <v Speaker1>So let's think about a specific type of user response to a discount or how a 00:42:03.751 --> 00:42:07.051 <v Speaker1>certain feature usage correlates with a particular pricing tier. 00:42:07.551 --> 00:42:13.731 <v Speaker1>So if you learn from past promotion data and user behavior, DEI can build predictive models. 00:42:13.911 --> 00:42:19.151 <v Speaker1>So you can ask it, for instance, to forecast if we want a 20 percent off promotion 00:42:19.151 --> 00:42:24.651 <v Speaker1>for our super tier, what will be the impact on our monthly recurring revenue and customer churn? 00:42:24.651 --> 00:42:30.151 <v Speaker1>It can then provide data-driven insights to help you make a more informed decision. 00:42:31.182 --> 00:42:36.262 <v Speaker1>Lastly, Gen AI can enable really a dynamic pricing model that automatically 00:42:36.262 --> 00:42:41.922 <v Speaker1>adjusts in real time based on demand, usage, and even individual user profiles. 00:42:42.362 --> 00:42:48.302 <v Speaker1>So, for example, if a user is a heavy power user, which is who is perhaps on 00:42:48.302 --> 00:42:53.282 <v Speaker1>the bridge of exceeding the included credits, the AI could serve them a personalized 00:42:53.282 --> 00:42:55.282 <v Speaker1>offer to upgrade to the next tier. 00:42:55.522 --> 00:43:00.322 <v Speaker1>And this means that you can capture more value while providing the user with a predictable bill. 00:43:00.882 --> 00:43:05.362 <v Speaker1>So really, to sum this up, instead of just using spreadsheets and intuition, 00:43:05.982 --> 00:43:07.442 <v Speaker1>founders can leverage Gen.E.I. 00:43:07.622 --> 00:43:10.242 <v Speaker1>To become more scientific and automated in their pricing. 00:43:10.482 --> 00:43:14.162 <v Speaker1>So it's really a shift from a static pricing page to a living, 00:43:14.622 --> 00:43:18.522 <v Speaker1>breathing pricing engine that's constantly learning, adapting and optimizing 00:43:18.522 --> 00:43:21.202 <v Speaker1>for both customer value and business profitability. 00:43:23.982 --> 00:43:29.562 <v Speaker0>Now that we talked so much about monetizing, let's go a little bit into risk and barriers. 00:43:29.562 --> 00:43:37.502 <v Speaker0>In the past, we had a lot of startups and subject matter experts talk about 00:43:37.502 --> 00:43:40.602 <v Speaker0>the data quality that needs to go into models. 00:43:40.942 --> 00:43:48.182 <v Speaker0>What hurdles around data quality and use case alignment most often block the monetization? 00:43:49.022 --> 00:43:51.962 <v Speaker1>This is a problem I observe a lot of the times. 00:43:52.502 --> 00:43:57.802 <v Speaker1>And it also ties back to what we discussed earlier, why promising proof of concepts fail to scale. 00:43:58.522 --> 00:44:02.982 <v Speaker1>So a POC is often built for curated success, how we call it. 00:44:03.142 --> 00:44:07.742 <v Speaker1>So it really operating on a small, clean data set in an isolated environment. 00:44:08.582 --> 00:44:12.342 <v Speaker1>But when you really try to bring it to a production-ready state, 00:44:12.582 --> 00:44:16.822 <v Speaker1>you, of course, also encounter the messy, fragmented data of the real world. 00:44:17.162 --> 00:44:21.382 <v Speaker1>And the challenge here is that Gen AI models require high-quality, 00:44:21.542 --> 00:44:23.682 <v Speaker1>up-to-date data to perform effectively. 00:44:24.302 --> 00:44:29.242 <v Speaker1>So if your data is incomplete, outdated, or unstructured, the garbage in, 00:44:29.342 --> 00:44:30.802 <v Speaker1>garbage out principle applies. 00:44:31.082 --> 00:44:35.762 <v Speaker1>So an AI solution built on a weak data foundation will produce inaccurate, 00:44:36.242 --> 00:44:38.522 <v Speaker1>unreliable, or nonsensical outputs. 00:44:38.722 --> 00:44:41.382 <v Speaker1>So this really directly kills monetization. 00:44:41.862 --> 00:44:46.902 <v Speaker1>So as we discussed, a customer will only pay for a product that delivers a tangible return. 00:44:47.042 --> 00:44:51.002 <v Speaker1>So if your AI isn't producing consistent, high-quality results, 00:44:51.182 --> 00:44:55.342 <v Speaker1>it can't deliver on its promises. So no matter how elegant your pricing model 00:44:55.342 --> 00:44:58.882 <v Speaker1>is, a product that doesn't work is impossible to monetize. 00:44:59.584 --> 00:45:05.044 <v Speaker1>And secondly, and perhaps the most interesting hurdle, is the lack of clear use case. 00:45:05.244 --> 00:45:10.464 <v Speaker1>So many startups fall into the trap of being a solution in search of a problem. 00:45:10.584 --> 00:45:15.024 <v Speaker1>So they develop a technically impressive AI model, but fail to connect it to 00:45:15.024 --> 00:45:17.064 <v Speaker1>real quantifiable business need. 00:45:17.324 --> 00:45:22.444 <v Speaker1>This is why, together with a group of colleagues, we launched an EMEA-wide Gen 00:45:22.444 --> 00:45:26.484 <v Speaker1>AI launchpad for about 200 customers in five cities in Europe, 00:45:26.484 --> 00:45:31.964 <v Speaker1>where we worked with each startup prior to the event to refine their use case, 00:45:32.444 --> 00:45:34.724 <v Speaker1>align architecture, and ensure data readiness. 00:45:35.024 --> 00:45:39.524 <v Speaker1>So as part of the event series, we also worked on the AI Canvas I created, 00:45:39.764 --> 00:45:45.344 <v Speaker1>which gets customers to define measurable KPIs to track Gen AI and data-driven impact. 00:45:45.724 --> 00:45:49.864 <v Speaker1>So the program's success came from pushing customers to dive deeper into the 00:45:49.864 --> 00:45:53.644 <v Speaker1>business metrics, which helped them select the right use case from the start. 00:45:54.484 --> 00:45:59.044 <v Speaker1>So really, the AI canvas to elaborate starts by asking questions around the 00:45:59.044 --> 00:46:03.384 <v Speaker1>customer's pain point and the negative consequence on the productivity or revenue. 00:46:03.404 --> 00:46:07.964 <v Speaker1>So the value added by your project and what's the impact of not doing it? 00:46:08.164 --> 00:46:11.844 <v Speaker1>And lastly, very important, how do you measure success? 00:46:12.224 --> 00:46:16.784 <v Speaker1>And when a founder fails to clearly define these elements, their project becomes 00:46:16.784 --> 00:46:19.404 <v Speaker1>a costly experiment with no clear path to revenue. 00:46:19.984 --> 00:46:25.444 <v Speaker1>So really, this is a job number zero without a tangible business problem to solve. 00:46:25.584 --> 00:46:30.184 <v Speaker1>The solution can't be tied to a measurable outcome and therefore it can't be monetized. 00:46:30.664 --> 00:46:34.584 <v Speaker1>And in the end, these hurdles that I just mentioned are very linked. 00:46:34.804 --> 00:46:39.584 <v Speaker1>So a great use case on its own is just an idea. So it really needs a foundation 00:46:39.584 --> 00:46:41.484 <v Speaker1>of good data to become a reality. 00:46:41.924 --> 00:46:46.504 <v Speaker1>And good data on its own is also just a resource. So it needs a clear use case 00:46:46.504 --> 00:46:51.844 <v Speaker1>to unlock its value. So for an AI solution to be monetized, it must consistently 00:46:51.844 --> 00:46:54.544 <v Speaker1>deliver on a clear, compelling value proposition. 00:46:54.604 --> 00:47:00.324 <v Speaker1>And it really can only do so if it's built on the right data and solves a real business problem. 00:47:02.384 --> 00:47:08.864 <v Speaker0>You've warned about risks like cross-tenant attacks in SaaS AI products. 00:47:09.124 --> 00:47:15.664 <v Speaker0>How should founders communicate risks transparently without scaring away their investors? 00:47:16.444 --> 00:47:20.564 <v Speaker1>That's a really interesting one because, first of all, when I talk to investors 00:47:20.564 --> 00:47:23.364 <v Speaker1>as well, they are not really looking for a risk-free company. 00:47:23.604 --> 00:47:26.184 <v Speaker1>They know also that those don't really exist. 00:47:26.604 --> 00:47:30.424 <v Speaker1>They're really looking for founders who are aware of the risks and have a clear, 00:47:30.564 --> 00:47:32.384 <v Speaker1>well-thought-out plan to manage them. 00:47:32.624 --> 00:47:37.204 <v Speaker1>So by presenting your security strategy with confidence and specific details, 00:47:37.444 --> 00:47:38.744 <v Speaker1>you're not scaring them off. 00:47:39.024 --> 00:47:42.884 <v Speaker1>Instead, you're really building trust and showing them that you have thought 00:47:42.884 --> 00:47:48.004 <v Speaker1>about this And also you are thinking about building a successful, resilient business. 00:47:49.016 --> 00:47:52.816 <v Speaker1>So the first step is to communicate that security isn't an afterthought. 00:47:52.816 --> 00:47:55.376 <v Speaker1>It's really a foundational principle to your business. 00:47:55.596 --> 00:47:59.756 <v Speaker1>So this means that you've baked in security measures from the very beginning, 00:47:59.756 --> 00:48:02.656 <v Speaker1>rather than patching them on at a later stage. 00:48:02.776 --> 00:48:05.496 <v Speaker1>And this is really key for gaining investor trust. 00:48:05.776 --> 00:48:11.456 <v Speaker1>So instead of talking about the risk of a cross-tenant attack as a scary possibility, 00:48:12.056 --> 00:48:15.496 <v Speaker1>frame it as a known challenge that you are actively solving. 00:48:16.336 --> 00:48:20.656 <v Speaker1>So to also recap for people that don't know about cross-tenant attacks, 00:48:20.896 --> 00:48:25.696 <v Speaker1>that means that there's a malicious actor gaining unauthorized access to a single 00:48:25.696 --> 00:48:28.936 <v Speaker1>instance that it uses to breach other customers' data. 00:48:29.016 --> 00:48:33.296 <v Speaker1>And it is a well-known vulnerability in a multi-tenant SaaS system. 00:48:34.016 --> 00:48:39.476 <v Speaker1>And this is just my kind of recommendation, how you can communicate your solution 00:48:39.476 --> 00:48:41.216 <v Speaker1>in a confident, transparent way. 00:48:41.856 --> 00:48:46.096 <v Speaker1>So this could be explaining how you've implemented strong guardrails to ensure 00:48:46.096 --> 00:48:51.756 <v Speaker1>that customer data is secure and private, or how each customer's data is stored 00:48:51.756 --> 00:48:54.716 <v Speaker1>in its own dedicated encrypted database or schema. 00:48:54.916 --> 00:48:59.336 <v Speaker1>So this is a crucial technical defense against cross-tenant attacks. 00:48:59.556 --> 00:49:04.256 <v Speaker1>So mention that you've implemented robust identity and access management policies 00:49:04.256 --> 00:49:08.276 <v Speaker1>to ensure that one customer's credentials can't be used to access another's. 00:49:08.276 --> 00:49:12.696 <v Speaker1>Talk, for instance, also about your robust monitoring and alerting systems that 00:49:12.696 --> 00:49:17.996 <v Speaker1>don't just tell you when something has failed, but also provide early warnings of potential issues. 00:49:18.256 --> 00:49:23.696 <v Speaker1>So this really demonstrates to your investor a productive and proactive approach 00:49:23.696 --> 00:49:27.716 <v Speaker1>rather than a reactive way and thinking towards security. 00:49:28.376 --> 00:49:32.896 <v Speaker1>So you can show also that your applications are designed to gracefully degrade 00:49:32.896 --> 00:49:34.416 <v Speaker1>rather than completely fail. 00:49:34.416 --> 00:49:40.016 <v Speaker1>So if one service becomes unavailable, your system will continuously operate 00:49:40.016 --> 00:49:43.696 <v Speaker1>with reduced functionality instead of shutting down entirely. 00:49:44.116 --> 00:49:49.236 <v Speaker1>So by presenting your security strategy in this way, you're really not scaring investors. 00:49:49.496 --> 00:49:54.216 <v Speaker1>Instead, you're building trust and showing that you have the foresight and technical 00:49:54.216 --> 00:49:57.416 <v Speaker1>capabilities to build a successful, resilient business. 00:49:57.416 --> 00:50:02.676 <v Speaker1>So this moves the conversation from what if this happens to what makes your 00:50:02.676 --> 00:50:04.576 <v Speaker1>security plan better than the competition. 00:50:04.796 --> 00:50:09.516 <v Speaker1>So this is a lot more powerful when you are in the fundraising discussions, for instance. 00:50:11.796 --> 00:50:17.196 <v Speaker0>Let's talk after so much input and we may add for our audience that we are already 00:50:17.196 --> 00:50:19.436 <v Speaker0>recording here for more than 45 minutes. 00:50:20.796 --> 00:50:27.076 <v Speaker0>Let's go a little bit into the outlook. which industry verticals do you expect 00:50:27.076 --> 00:50:34.956 <v Speaker0>to see the fastest return on investment from Gen.AI, SaaS, FinTech Healthcare, or actually others? 00:50:36.251 --> 00:50:41.571 <v Speaker1>So based on what I've observed, so the quickest ROI tends to come from operational 00:50:41.571 --> 00:50:44.611 <v Speaker1>improvements rather than really high risk product innovations. 00:50:44.931 --> 00:50:50.291 <v Speaker1>So this is where companies can automate internal processes, reduce manual labor 00:50:50.291 --> 00:50:52.551 <v Speaker1>and drive immediate efficiency gains. 00:50:52.971 --> 00:50:57.811 <v Speaker1>So industries that have a high density of these types of tasks are more likely to win early. 00:50:58.091 --> 00:51:03.351 <v Speaker1>So SaaS and fintech are at the top of the list for rapid GNI return on investment. 00:51:03.351 --> 00:51:08.471 <v Speaker1>So the primary reason for this is that these industries have a massive volume 00:51:08.471 --> 00:51:12.851 <v Speaker1>of customer interactions and data, much of which that can't be automated. 00:51:13.351 --> 00:51:19.011 <v Speaker1>So, for instance, startups can deploy AI agents to handle common customer inquiries. 00:51:19.031 --> 00:51:22.471 <v Speaker1>So this reduces the need for human agents on the front line. 00:51:22.711 --> 00:51:27.531 <v Speaker1>And this directly translates into lower operational cost and faster response 00:51:27.531 --> 00:51:31.451 <v Speaker1>time, which provides ultimately a clear and quantifiable ROI. 00:51:31.451 --> 00:51:37.131 <v Speaker1>And FinTech in particular has many high-volume repetitive tasks in thinking 00:51:37.131 --> 00:51:41.211 <v Speaker1>about verification, data mapping for B2B clients. 00:51:41.571 --> 00:51:45.751 <v Speaker1>And Gen AI can automate these workflows, really improving the time to value 00:51:45.751 --> 00:51:49.891 <v Speaker1>for customers and freeing up human employees for more strategic work. 00:51:50.431 --> 00:51:55.951 <v Speaker1>SaaS companies can also use Gen AI to create personalized brand-compliant content at scale. 00:51:56.311 --> 00:52:00.551 <v Speaker1>Think about marketing emails to sales collateral. So this really speeds up go-to-market 00:52:00.551 --> 00:52:03.411 <v Speaker1>efforts and reduce the cost of content creation. 00:52:04.249 --> 00:52:09.709 <v Speaker1>I also am a strong believer of healthcare and life science industry because 00:52:09.709 --> 00:52:14.649 <v Speaker1>they have a compelling case for rapid ROI, as we just discussed in the Pfizer example, 00:52:14.869 --> 00:52:20.309 <v Speaker1>because they have the potential to automate the administrative and research-related tasks. 00:52:20.849 --> 00:52:26.349 <v Speaker1>So, GNI can automate the generation of medical reports, summarize patient records 00:52:26.349 --> 00:52:30.629 <v Speaker1>for doctors, and even streamline billing and coding processes. 00:52:30.629 --> 00:52:34.609 <v Speaker1>So this frees up doctors and nurses to focus on patient care, 00:52:34.789 --> 00:52:36.909 <v Speaker1>which is a significant value driver. 00:52:37.249 --> 00:52:41.409 <v Speaker1>And also, as we've seen before, in drug discovery and clinical trials, 00:52:41.849 --> 00:52:47.349 <v Speaker1>Gen AI can analyze vast amounts of data to identify new compounds or predict 00:52:47.349 --> 00:52:48.729 <v Speaker1>the yield of new medicine. 00:52:49.109 --> 00:52:53.769 <v Speaker1>So as we've seen in the Pfizer example, they have focused their Gen AI efforts 00:52:53.769 --> 00:52:59.309 <v Speaker1>on these scientific applications and predicted annual cost savings of hundreds of millions of dollars. 00:52:59.309 --> 00:53:04.429 <v Speaker1>So the overall takeaway here is that while the product innovation and customer 00:53:04.429 --> 00:53:09.449 <v Speaker1>phasing features offer the promise of high returns, they often come with higher risk. 00:53:09.669 --> 00:53:14.309 <v Speaker1>So if we contrast this with internal operational improvements, 00:53:14.669 --> 00:53:21.889 <v Speaker1>they offer a lower risk path to a medium risk path, but they often also have a faster return. 00:53:22.549 --> 00:53:26.809 <v Speaker1>So for founders, ultimately, the key is to prioritize use cases with a clear 00:53:26.809 --> 00:53:30.909 <v Speaker1>link between a Gen AI solution and a measurable business outcome, as we said earlier. 00:53:31.109 --> 00:53:35.529 <v Speaker1>So this could be cost reduction or efficiency gains. So you can really demonstrate 00:53:35.529 --> 00:53:41.109 <v Speaker1>a clear ROI to stakeholders and build momentum for even more ambitious projects down the road. 00:53:44.689 --> 00:53:52.829 <v Speaker0>Do you believe in the next five years, we'll see AI-first companies valued differently 00:53:52.829 --> 00:53:55.809 <v Speaker0>by investors from traditional SaaS companies? 00:53:56.029 --> 00:54:06.289 <v Speaker0>We've seen this, for example, with remote-first, with cloud-first, and so on and so forth. 00:54:06.449 --> 00:54:08.689 <v Speaker0>Do you see that also playing out in the future? 00:54:09.549 --> 00:54:14.549 <v Speaker1>So absolutely, really, when looking at the current state of the market. 00:54:14.809 --> 00:54:20.069 <v Speaker1>So I believe that over the next five years, we are very certain to see AI-first 00:54:20.069 --> 00:54:24.369 <v Speaker1>companies that are valued differently by investors than traditional SaaS firms, 00:54:24.489 --> 00:54:27.009 <v Speaker1>for instance. So let's break that down a bit more. 00:54:27.369 --> 00:54:32.329 <v Speaker1>So for me, an AI-first company is a business whose core product, 00:54:32.489 --> 00:54:37.249 <v Speaker1>internal operations and competitive advantage really depend on artificial intelligence. 00:54:37.249 --> 00:54:41.809 <v Speaker1>So this is often agentic AI, which means software not just performing really 00:54:41.809 --> 00:54:45.449 <v Speaker1>narrow tasks, but autonomously making decisions, for instance, 00:54:45.589 --> 00:54:49.869 <v Speaker1>setting goals and refining actions across complex multi-step processes. 00:54:50.269 --> 00:54:55.069 <v Speaker1>So agentic AI is a step beyond classic automation. We're talking about systems 00:54:55.069 --> 00:54:59.709 <v Speaker1>that can analyze problems, coordinate with other AIs and adapt over time. 00:54:59.849 --> 00:55:03.849 <v Speaker1>So essentially managing entire workflows without constant human input. 00:55:04.636 --> 00:55:07.596 <v Speaker1>So this is also what the research shows around this. 00:55:07.816 --> 00:55:13.136 <v Speaker1>So if you think about investors, major VCs, and enterprise-focused funds, 00:55:13.356 --> 00:55:17.756 <v Speaker1>they are already allocating significant capital to agentic AI startups. 00:55:18.136 --> 00:55:24.776 <v Speaker1>So according to DROOM, there are more than 755 agentic AI startups as of mid 00:55:24.776 --> 00:55:29.556 <v Speaker1>of this year, with combined funding that is topping $10 billion for categories 00:55:29.556 --> 00:55:32.056 <v Speaker1>like coding agents, business process automation, 00:55:32.456 --> 00:55:34.756 <v Speaker1>health agents, and agent building frameworks. 00:55:35.496 --> 00:55:41.716 <v Speaker1>Gartner predicts that while the current environment is experimental with limited production examples, 00:55:42.256 --> 00:55:47.256 <v Speaker1>agentic AI projects will both attract acquisitions from legacy automation manufacturers 00:55:47.256 --> 00:55:51.916 <v Speaker1>and also face challenges, including also high cancellation rates for projects 00:55:51.916 --> 00:55:54.836 <v Speaker1>that don't meet real autonomy needs. 00:55:55.516 --> 00:55:59.936 <v Speaker1>And we're really seeing a pivot here. So if we think about these deterministic 00:55:59.936 --> 00:56:01.796 <v Speaker1>domains like finance or logistics, 00:56:01.976 --> 00:56:08.856 <v Speaker1>enterprises chase a lot of these proven automation solutions and may also acquire 00:56:08.856 --> 00:56:11.196 <v Speaker1>Agentec AI startups as an addition. 00:56:11.556 --> 00:56:17.596 <v Speaker1>And we see this also or what I see here happening is to have this for creative 00:56:17.596 --> 00:56:19.116 <v Speaker1>or knowledge based workflows. 00:56:19.476 --> 00:56:24.136 <v Speaker1>So Agentec AI is opening doors for brand new players. So these are the companies 00:56:24.136 --> 00:56:28.296 <v Speaker1>really investors are keeping a close look on because they're not just placing 00:56:28.296 --> 00:56:30.896 <v Speaker1>old tech, they're really redefining what's possible. 00:56:31.316 --> 00:56:35.796 <v Speaker1>And I'm working actually with different startups that are using Gentic AI for 00:56:35.796 --> 00:56:41.636 <v Speaker1>workflow automation, data science and cybersecurity, where nearly half of them are AI first. 00:56:41.836 --> 00:56:45.336 <v Speaker1>So they're building solutions like autonomous AI, employees, 00:56:45.536 --> 00:56:49.896 <v Speaker1>decision making, executives and multi-agent platforms that handle everything 00:56:49.896 --> 00:56:52.356 <v Speaker1>from debt collection to healthcare automation. 00:56:52.596 --> 00:56:59.096 <v Speaker1>And if we think about SaaS businesses, they still deliver value by digitizing 00:56:59.096 --> 00:57:02.176 <v Speaker1>and streamlining manual processes. 00:57:02.256 --> 00:57:04.776 <v Speaker1>So they're offering software plus a service. 00:57:05.446 --> 00:57:09.766 <v Speaker1>But agentic AI firms really promise scalable autonomy. 00:57:10.066 --> 00:57:15.346 <v Speaker1>So the ability to grow and adapt without requiring a customer to upsize their human team. 00:57:15.746 --> 00:57:20.306 <v Speaker1>So investors see a lot of growth potential here. So imagine a startup that is 00:57:20.306 --> 00:57:21.946 <v Speaker1>serving thousands of customers with 00:57:21.946 --> 00:57:26.786 <v Speaker1>only a handful of staff and a suite of autonomous AIs running the show. 00:57:26.986 --> 00:57:30.126 <v Speaker1>So it really changes a lot how value is created. 00:57:30.386 --> 00:57:34.746 <v Speaker1>But also I want to outline a little bit the challenges. So reliability, 00:57:35.186 --> 00:57:39.266 <v Speaker1>hallucinations, context limitations, as we also mentioned earlier, 00:57:39.986 --> 00:57:42.646 <v Speaker1>These all pose a big hurdle. 00:57:42.646 --> 00:57:47.426 <v Speaker1>So it is really important to have evaluation and oversized frameworks in place, 00:57:47.466 --> 00:57:51.266 <v Speaker1>which are really, for me, timeless principles that are still high in demand 00:57:51.266 --> 00:57:52.726 <v Speaker1>and will remain that way. 00:57:53.086 --> 00:57:57.906 <v Speaker1>So let's also acknowledge this. So I see a lot of agent washing also happening. 00:57:57.906 --> 00:58:02.386 <v Speaker1>So labeling really simple automations as agentic AI for hype, 00:58:02.686 --> 00:58:09.226 <v Speaker1>really savvy investors are probing deeper here to really find true autonomy and scalability. 00:58:09.686 --> 00:58:14.306 <v Speaker1>So if we think also about numbers that Gardner is publishing here, 00:58:14.486 --> 00:58:19.826 <v Speaker1>more than 40% of agentic projects will be canceled because many use cases don't 00:58:19.826 --> 00:58:22.006 <v Speaker1>actually need genuine creative autonomy. 00:58:22.006 --> 00:58:27.966 <v Speaker1>So often partial autonomy, not really full independence, is actually the way to go. 00:58:28.226 --> 00:58:32.806 <v Speaker1>So on the flip side, those who get the tech at this stage, and that's mostly 00:58:32.806 --> 00:58:39.746 <v Speaker1>startups with really true agentic capabilities, will actually receive good valuations 00:58:39.746 --> 00:58:41.306 <v Speaker1>and get a lot of interest. 00:58:41.546 --> 00:58:46.246 <v Speaker1>So there's a high probability that EI first companies, especially those that 00:58:46.246 --> 00:58:51.226 <v Speaker1>are generally agentic, will be valued differently by investors than traditional SaaS firms. 00:58:52.006 --> 00:58:56.546 <v Speaker1>So the premium really will come from this autonomous scaling, new margin structures. 00:58:57.426 --> 00:59:01.286 <v Speaker1>And yeah, investors will really look for those where AI is not just a feature, 00:59:01.486 --> 00:59:03.726 <v Speaker1>but the engine powering growth and innovation. 00:59:04.206 --> 00:59:07.806 <v Speaker1>But also there will be experimentation insurance. So as mentioned, 00:59:07.986 --> 00:59:11.246 <v Speaker1>some projects and even startups will not survive this hype cycle. 00:59:11.466 --> 00:59:15.766 <v Speaker1>So for me, the winners will be those who combine deep agentic autonomy with 00:59:15.766 --> 00:59:19.486 <v Speaker1>robust evaluation security and operational infrastructure. 00:59:21.906 --> 00:59:28.586 <v Speaker0>I have to admit, I have to smile when you talk about mislabeling of agentic AI. 00:59:29.226 --> 00:59:33.466 <v Speaker0>We've seen this with AI in general. We've seen this with Greentech, 00:59:33.546 --> 00:59:35.966 <v Speaker0>so it's kind of a repeating pattern. 00:59:36.526 --> 00:59:40.606 <v Speaker0>Jennifer, thank you very much for being a guest. We will have you back for a 00:59:40.606 --> 00:59:43.806 <v Speaker0>second episode together with AWS. Thank you very much. 00:59:48.560 --> 01:00:14.116 <v Music>