Latest / B2B SaaS Talks with Fexingo: Enterprise Software, Sales Cycles, and Procurement / Why Enterprise Buyers Now Demand a Vendor AI Maturity Model
Transcript
- Lucas: So you're a VP of Engineering at a Series B cybersecurity startup. You've got a $4.7 million enterprise deal on the line, and the prospect's procurement team sends over their vendor assessment packet. It's 80 pages. But there's a new section you haven't seen before: AI Maturity Model. They want you to self-assess and provide evidence for each level. Luna: And this is not just a checkbox. They're asking for artifacts — board minutes, model cards, incident logs. Lucas: Exactly. This is Episode 118, and we're talking about why enterprise buyers are now demanding vendors submit an AI Maturity Model as part of the procurement process. I'm Lucas. Luna: I'm Luna. And Lucas, I've seen this pop up in a few RFPs recently. Where did this come from? Lucas: It really gained traction after a Gartner survey in early 2025. They asked 300 enterprise procurement leaders: would you only sign with vendors at a certain AI maturity level? 68% said they'd require at least Level 3 on a 5-level scale. That's a huge shift. Luna: Level 3 being… what? Defined? Managed? Lucas: Right. The most common frameworks have five levels: Initial, Repeatable, Defined, Managed, Optimized. Level 3, Defined, means you have documented AI governance policies, a model inventory, and regular bias audits. That's the floor now for many enterprise deals. Luna: So if you're a startup and you haven't formalized any of that, you're already out of the running before they even look at your product. Lucas: That's exactly what happened to that cybersecurity firm I mentioned. They had a great product — really innovative adversarial ML detection. But their AI governance was ad hoc. The CTO was also the AI ethics officer. No formal board oversight. They scored Level 2. Luna: And the buyer walked? Lucas: Walked. The procurement lead told me — off the record — that the board had mandated that all AI vendors must be at Level 3 or above. No exceptions. The startup lost the deal. They're now scrambling to set up an AI ethics committee and document their model life cycle. Luna: I wonder if this is just a larger-company phenomenon though. Are mid-market buyers doing this too? Lucas: It's spreading. Gartner's survey included companies with revenue over $100 million, but I've seen RFPs from firms as small as $50 million that now include an AI maturity questionnaire. It's becoming table stakes. Luna: Let's break down the five levels. What does each actually require in practice? Lucas: Level 1 is Initial. Basically, AI is experimental. No governance. No documentation. It's fine for internal prototypes, but not for selling to enterprises. Luna: Level 2 is Repeatable. That's where you have some processes but they're not standardized. Maybe one team does bias testing, but another doesn't. Lucas: Right. Level 3, Defined, is the big one. You need a formal AI governance policy, a model registry, documented training data provenance, and regular bias audits. You also need a designated AI ethics officer or committee. Luna: That's a lot of overhead for a 50-person company. Lucas: It is. But consider the alternative: losing a $5 million deal. For many startups, that one contract could fund the entire governance program several times over. Luna: True. Level 4 is Managed. That means you have quantitative metrics for AI performance, drift detection, and automated monitoring in production. Lucas: And Level 5 is Optimized. Continuous improvement, automated retraining, full lifecycle traceability. Very few vendors are there. Maybe some of the big cloud providers and a handful of ai native firms. Luna: So the question becomes: how do you assess your own maturity honestly? And how do buyers verify it? Lucas: That's the tricky part. Some buyers ask for a self-assessment with evidence — like a link to your model card repository, or a signed policy document. Others use third-party auditors. There are now consulting firms that specialize in AI maturity assessments. Luna: It feels like SOC 2 all over again. At first, it was a differentiator. Now it's a gating item. Lucas: Exactly. And like SOC 2, it creates a barrier to entry for smaller vendors. But it also raises the bar for everyone. If you're a buyer, you want to know that the AI you're buying won't blow up in your face. Luna: Speaking of raising the bar — there's a real cost here. For a startup, building a Level 3 program could take six months and cost a couple hundred thousand dollars in legal, engineering, and audit time. Lucas: But here's the thing: the frameworks themselves are becoming more standardized. There's the NIST AI Risk Management Framework, the ISO/IEC 42001 standard for AI management systems, and now these maturity models. They all overlap significantly. Luna: So if you build for one, you're most of the way toward the others. Lucas: Right. And some VCs are starting to fund this work preemptively. They see it as a competitive advantage. If your portfolio company can walk into a procurement meeting with a Level 3 certification, that's a huge leg up. Luna: I want to get into a specific example. You mentioned a cybersecurity firm. Can you give us more detail without naming names? Lucas: Sure. This company — let's call them ShieldSec — had about 120 employees, $15 million in ARR. They had a really strong product for detecting ai generated phishing emails. Their technical demo blew the prospect away. Luna: But the procurement team asked for their AI maturity model self-assessment. And they scored Level 2. Lucas: Specifically, they scored Level 2 on governance. They had no board-level AI oversight committee, no formal AI ethics policy, and their model documentation was spread across wikis and Slack threads. The procurement team flagged it as a critical gap. Luna: And they couldn't close? Lucas: Couldn't close. The buyer's AI governance policy required that all vendors have a board-level AI committee. ShieldSec's board was just the two co-founders and one outside investor. They didn't have the bandwidth to stand up a separate committee quickly. Luna: So they lost the deal. And now they're playing catch-up. Lucas: They are. But here's the silver lining: they're using that loss as a case study to raise their next round. They're telling investors, 'We need $500K to build out AI governance so we don't lose the next $5M deal.' And it's working. Luna: That's a smart framing. Turn a failure into a fundraising narrative. Lucas: Exactly. And honestly, if this conversation is useful for what you're building or running, we're glad. We keep the show ad-free, and listener support is what makes that possible. If today's discussion was worth a coffee to you, that's the link — buy me a coffee dot com slash fexingo. Luna: Yeah, it's a small way to keep these deep dives coming without any sponsor interruptions. Lucas: So back to the maturity model. One criticism I hear is that it's just bureaucratic theater — that a vendor can check boxes without actually being safe. Luna: I've thought that too. Just because you have a policy doesn't mean you follow it. Or that your models are actually robust. Lucas: That's why some buyers are moving beyond self-assessment to independent audits. They want a third party to verify the maturity level. But that's expensive and still nascent. Luna: Is there any data showing that higher maturity correlates with fewer incidents? Lucas: Anecdotally, yes. There was a Stanford study last year that looked at 50 AI vendors. Those with formal governance processes had significantly fewer model failures in production. But it's not a huge dataset yet. Luna: So we're in the early days of this becoming standard. But the trend is clear. Lucas: Very clear. I'd expect within two years, any vendor selling AI to enterprises will need at least a Level 3 certification, just like SOC 2 is now expected for any SaaS handling customer data. Luna: For startups listening, what's the one thing they should do this week? Lucas: Start a model registry. Even a spreadsheet. Document every AI model you have in production or development, its training data, its performance metrics, and who's responsible. That's the foundation for everything else. Luna: And if you don't have an AI ethics policy, write a one-pager. Borrow from NIST's framework. It doesn't have to be perfect, but it has to exist. Lucas: Because the enterprise buyer isn't just buying your product. They're buying your AI governance. And maturity models make that explicit. Luna: Good place to leave it. Thanks Lucas. Lucas: Thanks Luna. Talk to everyone next time.