If you're running an AI services firm, you're probably not short of demand. The harder question is whether you're building something that compounds or something that just keeps you busy.

Over the last six months, we spoke with more than 100 founders of AI startups and established AI services companies. Some are happy running lean, highly profitable operations in a steady state. Others want to build something larger. Among those who are succeeding at the latter, a few clear patterns are emerging.

Services, by nature, are bespoke. That's fine. But doing everything from scratch for every client is a recipe for unstable margins and a GTM motion that sprawls in every direction. When you serve everyone, you don't just complicate delivery. You also multiply the events you need to attend, the thought leaders you need to track, and the subject-matter depth you need to maintain.

The founders we’ve seen break through typically narrowed harder than felt comfortable.

A non-obvious example: focusing on private equity funds as a niche market. Small teams of highly skilled and demanding people who negotiate like their life depends on it are hardly anyone’s dream customer. But when you treat them not as the end customer, but as a vehicle into their portfolio companies, a different picture emerges. PE firms are under pressure to generate returns in a high(er) interest rate environment, which brings “operational improvements” to the top of their agenda. They move faster than corporates, are less encumbered by internal politics, and see cost reduction as a core competency. One AI services firm we spoke with reached 10+ highly qualified prospects through a single GP relationship. The economics of that motion are hard to replicate by going direct.

We've also seen impressive results in industries that aren't traditionally associated with technology adoption. Manufacturing is one. These clients aren't trying to keep up with Silicon Valley. They're trying to skip several rungs on a digitisation ladder they know they're behind on. As one founder put it: "We're taking them from fax machines to LLMs." The competitive intensity in these pockets is a fraction of what it is in technology or financial services.

Horizontal firms exist and some will succeed. But at earlier stages, the math favours finding a a structural vehicle or an underserved pocket, rather than trying to win in the most crowded part of the market.

Finding repeatability in services is an art. The firms that have done it best treat it as a strategic priority from early on, not something they'll get to once they're bigger.

The clearest historical case is Bain’s Net Promoter Score (NPS) framework. A single question (“How likely are you to recommend us to a friend or colleague?”), released in 2003 as an open methodology with no licensing, spread quickly because it gave companies a simple way to measure customer loyalty. But the metric wasn’t the product. As adoption scaled, execution was inconsistent. That gap became consulting revenue: Bain helped enterprises  implement NPS properly, standardise data, and translate results into decisions. Consulting drove implementation, implementation generated structured data, the data powered a product, and the product drove more consulting.McKinsey's Organizational Health Index works on the same principle: a proprietary diagnostic that benchmarks a company against 2,600+ organisations on factors that predict future performance. Less visible externally, but the same structural idea.

Palantir took it further still, turning Foundry, Gotham, and Apollo from internal tools into delivery platforms that let the firm standardise parts of delivery and implement faster without reducing what it charges.

The fastest growing AI services companies are already experimenting with the same logic earlier. We’re seeing founders build small, self-contained “products” at $1–2m in revenue. Client-facing products help with lead generation. But internal products matter just as much. Streamlining delivery and simplifying operations can be the difference between growing 25% year-on-year and growing 50%.

But here's the thing: productisation only becomes tractable once you've picked your vertical. When you're trying to serve everyone, there's nothing repeatable enough to productise.

History is clear on how successful software vendors handle implementation: most of them outsource it.

Salesforce's partner ecosystem is estimated at 5–6x the size of Salesforce itself and generates $6.19 in revenue for every dollar of Salesforce revenue. SAP’s multiplier is 5x. Microsoft's is the largest of them all: 400,000+ organisations generating $8.45 for every $1 Microsoft earns.

These aren't niche arrangements. They are the dominant go-to-market model for enterprise software, and AI vendors are building the same structures.

Anthropic committed $100 million to its Claude Partner Network in 2026, with the explicit goal of making it straightforward for consulting firms and professional services organisations to build Claude practices. OpenAI is moving in the same direction, though hasn't yet publicised a specific figure.

If OpenAI and Anthropic feel out of reach, the same logic applies at a smaller scale. Clay runs a structured partner programme across two tracks: Solutions Partners (agencies and RevOps consultancies) and Integrations Partners (technical builders). Their ecosystem already counts 125+ agencies and 150+ integration partners, with a dedicated Clay Partner Growth Fund for co-investment. Glean, Intercom, n8n, and Make run less formalised versions of the same model.

What a vendor partnership gives you isn't just a lead channel. It's credibility that can't otherwise be bought. For buyers trying to evaluate a field of AI services providers where a significant proportion of firms are less than two years old, a formalised vendor relationship functions as a trust signal.

If a category has established incumbents, trying to beat them at their own game is rarely the right first move. The better approach is to find the pockets they've neglected.

When the dominant model for enterprise AI training in a market is top-down (large vendors selling to L&D directors and CHROs) going bottom-up is a legitimate wedge. The question to ask isn't "will this motion get me to $20m revenue?" It's "will this get me to the next stage, where I have the proof points and credibility to open more doors?"

The same instinct shows up inside the firm. Some founders have rebuilt common SaaS applications from scratch using low-code tools, configured precisely for their own workflow at a fraction of the off-the-shelf cost. When your margins are what they are, operational efficiency compounds quickly.

Others are rethinking org design. We have written before about about Forward Deployed Engineers: client-facing technical people who combine the ability to implement and troubleshoot AI tools with a commercial instinct for identifying where expansion and upsell opportunities exist. Palantir popularised the role. Smaller AI services firms are now building a lighter version, the Forward Deployed Empath: less technically brilliant perhaps, better at  building trust quickly, and able to work in a scrappy way.

It’s not a widely used model yet, which is partly why it's interesting. In a market where the line between technical delivery and account development is often blurry, a few strong Forward Deployed Empaths can quickly become a significant driver of growth.

SaaS companies scale by moving from founder-led sales to a repeatable sales team. Services companies scale by moving from founder-led sales to a founder-led brand. The mechanism is different, and it demands a different kind of investment. We covered this in depth while looking at founder-led sales in AI services.

Trust is the friction point in AI adoption right now. It keeps coming up in our conversations with founders and emerged as the dominant theme in the first ever survey of AI enablement services buyers. Buyers short-list vendors based on credentials they can verify, and eliminate vendors who lack them. The firms growing fastest are treating credibility-building as a first-class activity, not something that happens passively.

There's no shortcut. But the structure of it is more mechanical than it sounds. A successful POC, documented carefully and turned into a one-page case study (even anonymised) becomes the asset that earns the next client. That case study, packaged into a point-of-view piece, becomes thought leadership. That thought leadership, shared consistently, generates inbound: speaking slots, newsletter mentions, reposts from clients who want to be associated with the thinking. The firms that are building fastest aren't waiting for a marquee logo before they start publishing. They're making their second client feel like they're buying a proven system.

– Daria