Latest / Deep Tech Germany – AI, Robotics & Frontier Innovation by Startuprad.io™ / AI Agents Replaced an IT Department (Part 1 of 2)

Transcript

AI Agents Replaced an IT Department (Part 1 of 2) · Deep Tech Germany – AI, Robotics & Frontier Innovation by Startuprad.io™

• 0:00
  Jörn "Joe" Menninger: If custom software used to be too expensive for small companies,
• 0:04
  Jörn "Joe" Menninger: what happens when AI makes it cheap enough to compete with SaaS?
• 0:09
  Jörn "Joe" Menninger: Can AI really make small and medium-sized companies faster,
• 0:12
  Jörn "Joe" Menninger: or does it simply automate broken processes at higher speed?
• 0:16
  Jörn "Joe" Menninger: How should founders, SME owners, and investors think about a company like Amadeni AI,
• 0:22
  Jörn "Joe" Menninger: led by Delia Hallberg, which positions itself around AI-supported custom software
• 0:26
  Jörn "Joe" Menninger: for small and medium businesses, says typical projects range from €15,000 to €50,000,
• 0:31
  Jörn "Joe" Menninger: and argues that software should adapt to company workflows rather than forcing companies into rigid systems.
• 0:38
  Jörn "Joe" Menninger: Hello and welcome everybody.
• 0:39
  Jörn "Joe" Menninger: How do we separate real AI-enabled operational leverage from another wave of digital transformation theater?
• 0:45
  Jörn "Joe" Menninger: Delia, welcome to Startuprad.io.
• 0:47
  Jörn "Joe" Menninger: What is the operating problem in the Mittelstand that you think most people still describe incorrectly?
• 0:53
  Delia Hallberg: Hi Jörn, thank you for having me.
• 0:56
  Delia Hallberg: I would say, the Mittelstand is often misunderstood.
• 0:59
  Delia Hallberg: Let's talk about it in the interview.
• 1:02
  Jörn "Joe" Menninger: Great.
• 1:03
  Jörn "Joe" Menninger: Let's get started.
• 1:05
  Jörn "Joe" Menninger: When you look at a small or medium-sized business for the first time,
• 1:10
  Jörn "Joe" Menninger: what tells you within the first hour whether the real problem is software, process,
• 1:15
  Jörn "Joe" Menninger: or even the leadership?
• 1:17
  Delia Hallberg: I would say it's almost never just one.
• 1:20
  Delia Hallberg: And what makes this moment in time critical, when you introduce AI,
• 1:25
  Delia Hallberg: you stress test every process.
• 1:27
  Delia Hallberg: A bad process with AI on top of it breaks faster, louder, and more expensively.
• 1:33
  Delia Hallberg: And I would say, because you asked about software and process,
• 1:37
  Delia Hallberg: I would say the two are married.
• 1:40
  Delia Hallberg: You won't find a perfect process with a software problem or
• 1:45
  Delia Hallberg: terrible software with a great process.
• 1:47
  Delia Hallberg: And that's why we approach development in an extremely process-driven way.
• 1:52
  Delia Hallberg: There is usually a signal for each of them.
• 1:55
  Delia Hallberg: So on the software process side, the signal is, I always ask,
• 2:00
  Delia Hallberg: walk me through what happens when you receive, for example, an invoice.
• 2:05
  Delia Hallberg: And then I count systems, how many times people copy and paste information,
• 2:10
  Delia Hallberg: and how much inefficiency the process contains.
• 2:13
  Delia Hallberg: For example, we had a bakery with around 40 media discontinuities,
• 2:17
  Delia Hallberg: which cost 35 extra hours of work.
• 2:20
  Delia Hallberg: So if you would have thrown AI on top of that you would have automated chaos.
• 2:26
  Delia Hallberg: Then there is this leadership signal.
• 2:28
  Delia Hallberg: For example, when the managing director describes a workflow completely different
• 2:33
  Delia Hallberg: from what the team described 5 minutes earlier, that gap is a real risk.
• 2:38
  Delia Hallberg: So we have to be very cautious of it.
• 2:42
  Delia Hallberg: And that's why we run extensive workshops before writing a single line of code.
• 2:48
  Delia Hallberg: The diagnosis determines everything.
• 2:49
  Jörn "Joe" Menninger: Diagnosis determines everything.
• 2:51
  Jörn "Joe" Menninger: That is interesting.
• 2:52
  Jörn "Joe" Menninger: You have an interesting background because you actually studied sociology before
• 2:57
  Jörn "Joe" Menninger: moving into technology and product work.
• 2:59
  Jörn "Joe" Menninger: What did that teach you about organizations that are purely technical founder might miss?
• 3:05
  Delia Hallberg: I would say I don't take requirements at face value.
• 3:09
  Delia Hallberg: When a client says we need that tool that does X,
• 3:13
  Delia Hallberg: or even a lot of conversations with clients start with,
• 3:17
  Delia Hallberg: I need AI to do this and that for me.
• 3:21
  Delia Hallberg: Maybe many technical founders would hear a feature request.
• 3:25
  Delia Hallberg: So I come from systems thinking, and I think that's what sociology has taught me.
• 3:31
  Delia Hallberg: And I hear a symptom and I ask, why do you need that?
• 3:36
  Delia Hallberg: What's the process behind it?
• 3:38
  Delia Hallberg: And understanding the entire system is sociology at its core.
• 3:42
  Delia Hallberg: And for example, with one client, a youth welfare organization, the brief was,
• 3:47
  Delia Hallberg: we need a better documentation tool.
• 3:49
  Delia Hallberg: But when we start looking at the whole picture,
• 3:53
  Delia Hallberg: like how do you actually work and where inefficiency occurs in the entire process,
• 3:59
  Delia Hallberg: we very quickly understood that they don't need a better documentation tool.
• 4:03
  Delia Hallberg: They need an entirely different process of how they interact with
• 4:08
  Delia Hallberg: the software on a daily basis.
• 4:10
  Delia Hallberg: So in the end, we wrote an entire ERP system for them with voice-first interactions.
• 4:16
  Delia Hallberg: and semi-automated processes where documentation is only one smart part of the chain.
• 4:21
  Jörn "Joe" Menninger: You basically taught them how to use the tools.
• 4:25
  Jörn "Joe" Menninger: Okay, I see.
• 4:26
  Jörn "Joe" Menninger: ERP with voice-first interaction.
• 4:28
  Jörn "Joe" Menninger: That's also an interesting piece.
• 4:30
  Jörn "Joe" Menninger: Would love to see that one day.
• 4:33
  Jörn "Joe" Menninger: In the past, you have argued that AI made your own
• 4:37
  Jörn "Joe" Menninger: company possible because it lets you operate with a much smaller team.
• 4:42
  Jörn "Joe" Menninger: What exactly would have required 4 people before and what is now handled by agents?
• 4:48
  Delia Hallberg: Exactly.
• 4:50
  Delia Hallberg: So our situation is that we started with 2 people, with 2 founders,
• 4:55
  Delia Hallberg: and were actually able to go a very long way with only 2 people.
• 5:00
  Delia Hallberg: But I think we as you already said, we really did
• 5:05
  Delia Hallberg: the work of 4 or even more people.
• 5:08
  Delia Hallberg: The first area in which we work closely with agents is the most obvious:
• 5:13
  Delia Hallberg: marketing and content.
• 5:15
  Delia Hallberg: I would say 80% are outsourced to AI.
• 5:18
  Delia Hallberg: Agents hold our strategies, our content plans, our pipeline.
• 5:22
  Delia Hallberg: And I do the Q&A and finalization.
• 5:25
  Delia Hallberg: So then there is this research because we work across industries
• 5:29
  Delia Hallberg: and sometimes we start a project in a new industry that
• 5:33
  Delia Hallberg: we are not yet familiar with.
• 5:36
  Delia Hallberg: So we do deep dives into sectors and client processes before
• 5:40
  Delia Hallberg: we start a first conversation, which AI is great at helping us with.
• 5:45
  Delia Hallberg: And then there is the sales part, obviously, that were able
• 5:50
  Delia Hallberg: to automate quite a big workflow that basically makes sure that
• 5:54
  Delia Hallberg: during the sales process we have quality assurance, every piece of
• 5:59
  Delia Hallberg: data we can gather enters the process so that when this
• 6:03
  Delia Hallberg: lead converts into a project, we do not start the project from zero,
• 6:08
  Delia Hallberg: but we already have, for example, a project outline created by the previous sales agent.
• 6:14
  Delia Hallberg: And then the fourth and most significant area is coding.
• 6:18
  Delia Hallberg: This is where we really have the biggest efficiency gains.
• 6:22
  Delia Hallberg: We have multiple coding agents.
• 6:24
  Delia Hallberg: We have, I would say by now we are replacing an
• 6:29
  Delia Hallberg: entire IT department with a very coherent system of agents and it's not vibe coding.
• 6:35
  Delia Hallberg: Please don't misunderstand me.
• 6:36
  Delia Hallberg: It is an elaborate process designed to ensure consistent, coherent code quality.
• 6:41
  Delia Hallberg: Previously, that would have required maybe one additional person that would somehow do sales,
• 6:47
  Delia Hallberg: research, marketing combined, and at least three additional developers.
• 6:50
  Jörn "Joe" Menninger: When you talked about preparing for a new engagement,
• 6:54
  Jörn "Joe" Menninger: you were talking about all the data you add before the client actually says yes.
• 7:00
  Jörn "Joe" Menninger: I was wondering how, because it's from the outside,
• 7:03
  Jörn "Joe" Menninger: how do you ensure sufficient quality that you don't have the typical system problem,
• 7:09
  Jörn "Joe" Menninger: garbage in, garbage out?
• 7:11
  Delia Hallberg: That's a very good point.
• 7:13
  Delia Hallberg: So first of all, there is always a human in the
• 7:17
  Delia Hallberg: loop and I always have to be able to understand the results.
• 7:22
  Delia Hallberg: Otherwise, yeah, that's a risk.
• 7:24
  Delia Hallberg: And then again, it's about setting up a system where you
• 7:28
  Delia Hallberg: lower the risks of the AI producing low-value results or wrong results,
• 7:33
  Delia Hallberg: which you can do by certain steps that you do while
• 7:37
  Delia Hallberg: creating your agent system and not by simply throwing a prompt
• 7:42
  Delia Hallberg: in ChatGPT and hoping for the best.
• 7:45
  Jörn "Joe" Menninger: Hoping for the best.
• 7:46
  Jörn "Joe" Menninger: Yeah, I think for some people out there it's still strategy.
• 7:51
  Jörn "Joe" Menninger: Where is the boundary between a lean AI-enabled company and a
• 7:55
  Jörn "Joe" Menninger: founder simply hiding the operational fragility behind some automation?
• 7:59
  Delia Hallberg: That's a very relevant question, and, especially at this time.
• 8:03
  Delia Hallberg: So I would say the boundary is where you understand what
• 8:07
  Delia Hallberg: your agents are doing well enough to do it yourself if they stop tomorrow.
• 8:13
  Delia Hallberg: Agents don't replace competence, they replace capacity.
• 8:15
  Delia Hallberg: The judgment, the quality standard, the strategic thinking, this always has
• 8:20
  Delia Hallberg: to stay with the humans, in that case with me.
• 8:24
  Delia Hallberg: And I always think about it it's very tricky because AI
• 8:28
  Delia Hallberg: is so good at producing plausible output that you can fool
• 8:33
  Delia Hallberg: yourself for much longer before you realize that you've actually lost the plot.
• 8:38
  Delia Hallberg: That's also something that we really take care of with our clients,
• 8:43
  Delia Hallberg: with the tools that we create for them, that they always stay in the driver's seat.
• 8:49
  Delia Hallberg: We do a few things to manage that.
• 8:52
  Delia Hallberg: First, as I already said, there's always a human in the loop.
• 8:57
  Delia Hallberg: Second, we document agent architectures just like software architectures.
• 9:00
  Delia Hallberg: It needs to be scalable and not just living inside my head.
• 9:05
  Delia Hallberg: And third, I'm really honest about what AI can't do.
• 9:09
  Delia Hallberg: For example, client relationships, reading the room, judgment calls in whether
• 9:13
  Delia Hallberg: and how a project needs to shift.
• 9:16
  Delia Hallberg: And I would say an indicator of operational fragility is when
• 9:21
  Delia Hallberg: founders or people talk about AI replacing this and that department,
• 9:25
  Delia Hallberg: AI replacing this and that person because they are thinking about AI in terms of replacement.
• 9:31
  Delia Hallberg: I think the discussion that we should be having is AI in the sense of amplification.
• 9:37
  Delia Hallberg: The competence was there first and then AI makes it scale.
• 9:42
  Jörn "Joe" Menninger: Not only that I have learned that AI agents drift over time.
• 9:46
  Jörn "Joe" Menninger: So maybe at the start you get a marketing department and
• 9:51
  Jörn "Joe" Menninger: when you are towards the end of the year, the tool only does social media anymore.
• 9:57
  Jörn "Joe" Menninger: So I don't think there's right now a case for complete AI replacement for humans.
• 10:03
  Jörn "Joe" Menninger: So as you said, you always have to have a human in the loop.
• 10:08
  Jörn "Joe" Menninger: Your public positioning says Amadeni, your company, starts with a workshop, process analysis,
• 10:14
  Jörn "Joe" Menninger: and then decides whether custom software is actually needed.
• 10:17
  Jörn "Joe" Menninger: What is your decision rule for saying nope, you should not build software here?
• 10:23
  Delia Hallberg: So it's important to say we always look for a standard solution first,
• 10:28
  Delia Hallberg: because if you have a standard business case, and then there
• 10:32
  Delia Hallberg: is a SaaS product that genuinely fits, then I would say
• 10:37
  Delia Hallberg: that this is the way to go.
• 10:40
  Delia Hallberg: That's part of being an honest IT consultancy.
• 10:43
  Delia Hallberg: But what I've observed in the Mittelstand, especially in Eastern Germany,
• 10:47
  Delia Hallberg: is that businesses that have been established for 20,
• 10:51
  Delia Hallberg: maybe 30 years are the ones that are sitting in a niche.
• 10:56
  Delia Hallberg: And they're at the intersection of 2 industries.
• 10:59
  Delia Hallberg: So what happens quite frequently is that there is no proper software solution for them.
• 11:05
  Delia Hallberg: What they have done until now is they have tried to
• 11:09
  Delia Hallberg: use software maybe from an industry that they're slightly part of.
• 11:14
  Delia Hallberg: But this is creating so much problems than in daily processes.
• 11:18
  Delia Hallberg: That for example, a straightforward case where we go, okay, custom development really is the answer.
• 11:24
  Delia Hallberg: And one signal that is growing stronger and stronger, for example,
• 11:28
  Delia Hallberg: with an engineering firm in Berlin that we are currently working with,
• 11:33
  Delia Hallberg: they were very happy with their SaaS solution, with their sort
• 11:38
  Delia Hallberg: of project management solution, but they had been asking for a
• 11:42
  Delia Hallberg: long time for a very specific feature.
• 11:45
  Delia Hallberg: They didn't get it.
• 11:46
  Delia Hallberg: They didn't get an API.
• 11:48
  Delia Hallberg: So this, I would say, minor issue is really what caused them to say,
• 11:54
  Delia Hallberg: okay, we have to say goodbye to our solution.
• 11:58
  Delia Hallberg: And they proactively decided to go for a custom development.
• 12:02
  Delia Hallberg: Because if a SaaS product doesn't have an API, it means that nobody can add automation.
• 12:08
  Delia Hallberg: There is no flexibility.
• 12:09
  Delia Hallberg: And this is something that surprised even me.
• 12:13
  Delia Hallberg: This is really coming from our clients.
• 12:15
  Delia Hallberg: So the decision tree, I would say, is simple.
• 12:19
  Delia Hallberg: Is there a good standard solution that exists?
• 12:22
  Delia Hallberg: If yes, go for it.
• 12:24
  Delia Hallberg: If no, then custom software is really becoming a very viable alternative.
• 12:29
  Delia Hallberg: One point is especially important for clients.
• 12:32
  Delia Hallberg: If they don't have an API or some way to interact with their data,
• 12:37
  Delia Hallberg: then they're also going for the alternatives.
• 12:40
  Jörn "Joe" Menninger: Yes, we should always keep in mind it is not only
• 12:45
  Jörn "Joe" Menninger: a question of standard software or not, but also is the
• 12:49
  Jörn "Joe" Menninger: standard software breaking the bank or not?
• 12:52
  Jörn "Joe" Menninger: I think you know what I'm talking about here.
• 12:56
  Jörn "Joe" Menninger: I've been going through a little bit of your data,
• 13:00
  Jörn "Joe" Menninger: and in your Impuls KBB case, the reported gain was from
• 13:04
  Jörn "Joe" Menninger: 8 hours to 2 hours for development reports and a 4-week return on investment.
• 13:09
  Jörn "Joe" Menninger: What had to be true operationally for that result to be
• 13:14
  Jörn "Joe" Menninger: real rather than just a software demo?
• 13:17
  Delia Hallberg: I really like it that you're looking at the numbers.
• 13:21
  Delia Hallberg: So let's look at the whole picture.
• 13:24
  Delia Hallberg: The four-week ROI and calculated savings of €21,000 per month did
• 13:28
  Delia Hallberg: not come from one process, this efficiency increase from 8 to 2 hours.
• 13:33
  Delia Hallberg: As I mentioned a few questions earlier, went through the entire
• 13:37
  Delia Hallberg: process chain starting with case clarification from the youth welfare office,
• 13:42
  Delia Hallberg: really to the very end of invoicing.
• 13:45
  Delia Hallberg: And we really looked at every step where we can increase efficiency.
• 13:50
  Delia Hallberg: Across the entire process, there was so much room for improvement
• 13:54
  Delia Hallberg: that in that case were really able to achieve this, I would say, impressive ROI.
• 14:00
  Delia Hallberg: So first the client made a very, I would say,
• 14:04
  Delia Hallberg: bold move at the time because we had already started the
• 14:08
  Delia Hallberg: project and soon after the project began they decided to replace their old software.
• 14:14
  Delia Hallberg: They saw that it did not have an API.
• 14:17
  Delia Hallberg: Certain features in their previous software were causing a lot of
• 14:22
  Delia Hallberg: problems and they decided not to supplement it but replace it.
• 14:26
  Delia Hallberg: And that unlocked the enormous automation potential.
• 14:29
  Delia Hallberg: For example, the voice-first interactions, because these are social welfare workers
• 14:33
  Delia Hallberg: that go to the homes of children or youth.
• 14:37
  Delia Hallberg: And then afterwards they have to sit in their car,
• 14:41
  Delia Hallberg: open the laptop and type up documentation about what they discussed Now,
• 14:46
  Delia Hallberg: on their way back to the car, they can simply make
• 14:50
  Delia Hallberg: a voice message which then gets transcribed and then translated into
• 14:55
  Delia Hallberg: official bureaucratic language and then placed into an automated workflow.
• 14:59
  Delia Hallberg: Second, we built the system for how the social workers actually work.
• 15:03
  Delia Hallberg: As I said in the example of the voice notes,
• 15:07
  Delia Hallberg: They also had a lot of unofficial communication about what was
• 15:12
  Delia Hallberg: the status quo of this document and have you thought about this and that?
• 15:17
  Delia Hallberg: So we took a lot from that informal communication that created
• 15:22
  Delia Hallberg: really a lot of inefficiency and put it into a process within the software,
• 15:27
  Delia Hallberg: which eliminates a lot of wasted time.
• 15:30
  Delia Hallberg: And then third, we integrated it with the rest of the
• 15:35
  Delia Hallberg: system because they had a separate HR system.
• 15:38
  Delia Hallberg: So they had to do 2 types of time tracking.
• 15:42
  Delia Hallberg: And we said, no, the system has to work for the social workers.
• 15:47
  Delia Hallberg: They need only one interface.
• 15:49
  Delia Hallberg: And in the background, we will create a flow of data
• 15:53
  Delia Hallberg: so that they have only one touchpoint.
• 15:56
  Delia Hallberg: And I would say this entire process optimization built or visualized in the tool,
• 16:02
  Delia Hallberg: that's what really created this four-week ROI.
• 16:05
  Jörn "Joe" Menninger: I had to smile when you said that you make a
• 16:09
  Jörn "Joe" Menninger: voice memo and then it gets translated into bureaucratic language.
• 16:13
  Jörn "Joe" Menninger: It's— that's really a good point, a translator between bureaucrats and normal human beings.
• 16:19
  Jörn "Joe" Menninger: But let's go a little bit further.
• 16:21
  Jörn "Joe" Menninger: When a client has 20 systems, 40 media discontinuities,
• 16:25
  Jörn "Joe" Menninger: and 35 hours of avoidable duplicate work per week.
• 16:29
  Jörn "Joe" Menninger: How do you decide what you will fix first?
• 16:32
  Delia Hallberg: Yeah, so the numbers that you quote come from a case,
• 16:37
  Delia Hallberg: from a bakery that produces its own sourdough goods.
• 16:40
  Delia Hallberg: That's actually important for the process because sourdough makes things very complicated.
• 16:45
  Delia Hallberg: But the CEO called us because they had a bad gut feeling.
• 16:50
  Delia Hallberg: So we did an analysis in a 2-day workshop and this
• 16:54
  Delia Hallberg: bad gut feeling translated into these numbers that are quite shocking or quite surprising,
• 17:00
  Delia Hallberg: I would say.
• 17:01
  Delia Hallberg: And I'm very open about it.
• 17:04
  Delia Hallberg: We were a little bit overwhelmed at first, but what is really important,
• 17:09
  Delia Hallberg: and I think that's great that the CEO actually took that
• 17:13
  Delia Hallberg: step because you don't have to automate everything.
• 17:16
  Delia Hallberg: You don't have to optimize for everything, but you need to
• 17:21
  Delia Hallberg: prioritize and have a plan for how you want to deal with the other things.
• 17:27
  Delia Hallberg: First, we looked for ticking time bombs.
• 17:30
  Delia Hallberg: That's how I call them.
• 17:32
  Delia Hallberg: So something that could potentially go off any day and threaten the business.
• 17:37
  Delia Hallberg: So if you find a ticking time bomb, then that's obviously
• 17:41
  Delia Hallberg: at the very top of your roadmap.
• 17:44
  Delia Hallberg: And then you deal with it regardless of everything else.
• 17:48
  Delia Hallberg: In that case, we did not find any ticking time bombs.
• 17:52
  Delia Hallberg: But then you look at the fundamental structural things.
• 17:56
  Delia Hallberg: In that case, it was actually that they didn't have a leading system.
• 18:01
  Delia Hallberg: which led to a lot of distributed data flows.
• 18:05
  Delia Hallberg: And so it was very clear that we first had to
• 18:09
  Delia Hallberg: build a very sound foundation, like a building, and you don't
• 18:14
  Delia Hallberg: want to build it on sand.
• 18:16
  Delia Hallberg: And then we looked at impact.
• 18:19
  Delia Hallberg: So where did we have the lowest investment or where could
• 18:23
  Delia Hallberg: we create the greatest business impact with the lowest effort?
• 18:27
  Delia Hallberg: Then you have a prioritization, a roadmap.
• 18:30
  Delia Hallberg: And as I said initially, it's totally okay if you don't tackle every process,
• 18:35
  Delia Hallberg: the important difference is that you are aware of them,
• 18:39
  Delia Hallberg: have quantified them and you have a plan.
• 18:43
  Delia Hallberg: And I think that makes a difference between a vague gut
• 18:47
  Delia Hallberg: feeling and having worked through it.
• 18:49
  Jörn "Joe" Menninger: Yeah, you just have to have a framework built on numbers that you can add to.
• 18:55
  Jörn "Joe" Menninger: I was wondering, talking about numbers and metrics here,
• 18:59
  Jörn "Joe" Menninger: what is the metric you trust most when evaluating whether an
• 19:03
  Jörn "Joe" Menninger: AI automation project is worth building?
• 19:06
  Jörn "Joe" Menninger: Do you think about hours saved, reduction in errors, adoption rate, revenue impact,
• 19:11
  Jörn "Joe" Menninger: or maybe management relief?
• 19:13
  Delia Hallberg: I would say I try to collapse all these things that
• 19:17
  Delia Hallberg: you just said into a single number because at the end
• 19:22
  Delia Hallberg: you can put a price tag on nearly all those things.
• 19:26
  Delia Hallberg: What is it costing you right now in terms of time, errors, management firefighting,
• 19:31
  Delia Hallberg: but also what would the solution cost?
• 19:34
  Delia Hallberg: So if the solution has an ROI of less than six months,
• 19:39
  Delia Hallberg: then it's a very straightforward decision to go for it.
• 19:43
  Delia Hallberg: If it is 12 or 18 months, we have a more nuanced conversation,
• 19:48
  Delia Hallberg: and it's also more about strategy, about potential risks also of not acting, right?
• 19:54
  Delia Hallberg: But I would say there is a very big misconception that SMBs frequently have.
• 19:59
  Delia Hallberg: It's a conversation that I have with a lot of especially
• 20:04
  Delia Hallberg: potential clients at the beginning is they treat IT as its own budget.
• 20:09
  Delia Hallberg: So for example, a €25,000 project looks like it would blow the IT budget,
• 20:15
  Delia Hallberg: but and that's the difference from my perspective.
• 20:18
  Delia Hallberg: You have to calculate the IT budget in terms of personnel
• 20:22
  Delia Hallberg: costs of the inefficient process, and then when you compare the
• 20:27
  Delia Hallberg: personnel costs versus the investment into optimizing these processes and the
• 20:31
  Delia Hallberg: potential personnel costs that you save, not because you're letting go of people, but because.
• 20:37
  Delia Hallberg: Suddenly you can come from a very reactive business to a more strategic one,
• 20:42
  Delia Hallberg: or your people can actually do more meaningful things.
• 20:46
  Delia Hallberg: That reframing is one of the most important conversations that we have,
• 20:51
  Delia Hallberg: and then suddenly the decision-making becomes much easier.
• 20:54
  Jörn "Joe" Menninger: The decision-making.
• 20:56
  Jörn "Joe" Menninger: making becomes easier.
• 20:57
  Jörn "Joe" Menninger: Talk about processes here.
• 20:59
  Jörn "Joe" Menninger: You argue that a a poor process plus AI becomes a poor process on steroids.
• 21:05
  Jörn "Joe" Menninger: What is the clearest sign that process must be redesigned before it's even automated?
• 21:10
  Delia Hallberg: There is a very clear sign.
• 21:13
  Delia Hallberg: And it is when nobody can explain the process without naming a person.
• 21:18
  Delia Hallberg: So if people start saying, so I get the invoice and
• 21:22
  Delia Hallberg: then Sabine does this and that and then Thomas does that for example,
• 21:27
  Delia Hallberg: that's not a process, those are personal habits.
• 21:31
  Delia Hallberg: And we don't want to automate people's personal habits because if
• 21:35
  Delia Hallberg: you automate that and then the moment Sabine goes on holiday, the system doesn't work anymore.
• 21:41
  Delia Hallberg: So especially in the Mittelstand, that is a process that we
• 21:45
  Delia Hallberg: go through to get away from this person name-driven thinking to, okay,
• 21:50
  Delia Hallberg: what are the actual steps?
• 21:52
  Delia Hallberg: A second straightforward sign is if information gets typed more than twice.
• 21:57
  Delia Hallberg: A third is, shadow systems.
• 21:59
  Delia Hallberg: Shadow systems—shadow IT—are a very strong indicator that the process isn't working.
• 22:04
  Delia Hallberg: And if you automate that process without integrating the shadow systems
• 22:08
  Delia Hallberg: or without looking at why people had to create shadow systems,
• 22:13
  Delia Hallberg: then you'll create a tool that nobody's going to use.
• 22:17
  Delia Hallberg: I'm talking about private Excel files where the real numbers are,
• 22:21
  Delia Hallberg: or the WhatsApp group that's the actual project management tool.
• 22:25
  Delia Hallberg: So these are indicators where it's very clear that you have
• 22:30
  Delia Hallberg: to redesign first and automate second.
• 22:32
  Jörn "Joe" Menninger: Yes, I was just smiling talking about shadow IT and WhatsApp
• 22:36
  Jörn "Joe" Menninger: groups that actually are used for project management.
• 22:40
  Jörn "Joe" Menninger: I have seen that and in my experience, startups tend to do this more frequently.
• 22:46
  Jörn "Joe" Menninger: There was a heyday for Slack, but I think it's partly gone.
• 22:50
  Delia Hallberg: WhatsApp has remained popular in the Mittelstand.
• 22:53
  Jörn "Joe" Menninger: Yeah, maybe it just, it just moved there.
• 22:56
  Jörn "Joe" Menninger: That's also fine.
• 22:58
  Jörn "Joe" Menninger: Talk a little bit about more cases that I've seen from you for example,
• 23:03
  Jörn "Joe" Menninger: the Ostdeutscher Bankenverband case.
• 23:05
  Jörn "Joe" Menninger: The promise was source traceability and zero hallucination for an AI.
• 23:09
  Jörn "Joe" Menninger: What architectural choice prevents hallucination by design rather than by, yeah, you know, just hope?
• 23:15
  Delia Hallberg: Yeah, hope and prayers are not something that you want,
• 23:19
  Delia Hallberg: in your system design or in your IT.
• 23:23
  Delia Hallberg: so what we did is, I think the key is obviously
• 23:27
  Delia Hallberg: the key is not literally zero hallucinations, but lowering the risk
• 23:31
  Delia Hallberg: and having a system that can catch hallucinations.
• 23:35
  Delia Hallberg: Because in that case with the Ostdeutscher Bankenverband, they provide information
• 23:39
  Delia Hallberg: to their members that is crucial that their members base decisions on.
• 23:44
  Delia Hallberg: And if you have a hallucinated piece of information, that's reputation damaging, right?
• 23:49
  Delia Hallberg: So what we did is obviously human in the loop at every step.
• 23:54
  Delia Hallberg: Then second, we gave different, unconnected LLMs small tasks, because the smaller the task,
• 24:00
  Delia Hallberg: the more concrete the information, the less the AI is forced to hallucinate.
• 24:05
  Delia Hallberg: When does AI hallucinate?
• 24:07
  Delia Hallberg: When models lack information but still have to come up with an answer, right?
• 24:12
  Delia Hallberg: And so the answer was to cut the task very small
• 24:17
  Delia Hallberg: and to include in the system a traceability of information.
• 24:21
  Delia Hallberg: So every piece of information that I had, I could trace
• 24:25
  Delia Hallberg: back to the original website, to the original source of information.
• 24:29
  Delia Hallberg: And this was especially important for the humans.
• 24:33
  Delia Hallberg: who were in the loop to be able to prove every piece of information.
• 24:38
  Delia Hallberg: I would say that these two or three design decisions lowered
• 24:42
  Delia Hallberg: the risk of hallucinations very much.
• 24:45
  Jörn "Joe" Menninger: So you give the AI a very small area of responsibility
• 24:49
  Jörn "Joe" Menninger: at each step and enough information to fulfill it in order
• 24:54
  Jörn "Joe" Menninger: that they Don't need to hallucinate.
• 24:56
  Jörn "Joe" Menninger: Great.
• 24:58
  Jörn "Joe" Menninger: Delia, it was a pleasure talking to you.
• 25:01
  Jörn "Joe" Menninger: We agreed that we're going to cut this interview into 2 pieces.
• 25:06
  Jörn "Joe" Menninger: We'll be back with number 2 for you in a few days,
• 25:10
  Jörn "Joe" Menninger: for us in a few minutes.
• 25:13
  Jörn "Joe" Menninger: Talk to you soon.
• 25:14
  Jörn "Joe" Menninger: Thank you.
• 25:16
  Jörn "Joe" Menninger: Bye.
• 25:17
  Delia Hallberg: Bye.