A system that speaks services: why understanding beats horsepower
In the last article we drew the line between AI that routes and AI that reasons. But reasoning does not happen in a vacuum. A system can only evaluate a decision if it understands what the decision is about. That is the first thing that breaks when you drop general-purpose AI into a services firm. Not accuracy. Meaning.
A services firm sells time, but the client buys judgment, relationships and trust. The economics that sit behind that are specific and unforgiving, and they are invisible to a model trained on the average of the internet.
AI for Professional Services: Part 3 of 6
A system that speaks services: why understanding beats horsepower
In the last article we drew the line between AI that routes and AI that reasons. But reasoning does not happen in a vacuum. A system can only evaluate a decision if it understands what the decision is about. That is the first thing that breaks when you drop general-purpose AI into a services firm. Not accuracy. Meaning.
A services firm sells time, but the client buys judgment, relationships and trust. The economics that sit behind that are specific and unforgiving, and they are invisible to a model trained on the average of the internet.
"Built for everyone" fits no one
Most AI reaching the market is one of two things: a general-purpose model trained on everything, or an AI layer bolted onto a transactional system that only ever recorded what happened, never what it meant. Both are blind to the concepts that define services work.
A billable hour. A write-off. Revenue recognition across milestones. Utilisation across geographies. A resource allocation. To a generic model these are just words. It does not know that a write-off is a direct hit to margin, that revenue recognition is bound to delivery, or that a resource allocation made today is a delivery risk three weeks from now. It will produce fluent, plausible output that is quietly wrong in ways only a practitioner would catch, and in a services firm the person who catches it is often the client, or the finance team at quarter-end.
Jepsen is precise about where real capability comes from:
"An ontology is the foundation of it, the structured account of what the objects are and how they relate. But the real value lies in the meaning layered on top of it." — Claus Jepsen, Chief Technology Officer, Unit4 (Diginomica)"²
And he is equally clear that this kind of depth cannot be improvised:
"Vertical knowledge is something you build over years, through deep engagement with the customers who operate in those verticals."²
Understanding is now the differentiator, not a nice-to-have
This is not a philosophical point. SPI Research's 2026 benchmark of 509 firms found AI moving from experiment to operation, with between 26% and 38% of projects now incorporating it, and high-performing firms distinguished by markedly stronger AI fluency, at a time when utilisation had slipped to a record-low 66.4% and margins stayed under real pressure.⁶ The firms pulling ahead are not the ones with the most AI. They are the ones whose AI understands the economics it is working inside. Source Global's client-side research adds the demand signal: buyers now rank effective use of AI as the single most sought-after consulting service, up from eleventh a year earlier.⁷
How Unit4 approaches it
Unit4 has spent forty years inside organisations like yours: consultancies, agencies, engineering and advisory firms, the mid-market of professional services. That is not a marketing line. It is the reason ERPx knows what a write-off does to a margin.
ERPx is built around services concepts rather than a generic ledger: multi-entity engagements, revenue recognition across milestones, utilisation across geographies, and the dynamics of project economics as an engagement evolves. The system understands how those concepts relate and what rules govern them. So when it reasons over your data, a suggestion arrives already aware of the margin at stake and the delivery milestone in play. In practice that means a margin drift is flagged two months into delivery, because the system tracks actuals against forecast continuously, before it ever lands in AR.
The payoff is subtle but decisive. When the AI understands the domain, its output starts inside the economics rather than needing to be dragged back inside them. Your experts stop catching errors and start making judgements, which is exactly where scarce expertise should be spent.
The takeaway
Before you ask whether an AI is powerful, ask whether it understands your world. A model that does not know what a write-off is will happily help you book one. In professional services, domain understanding is not a layer on top of the AI. It is the foundation that makes everything above it worth trusting.
The real test is whether it speaks your firm's language. If you are already a Unit4 customer, you can try our AI for free on your own engagements and prove the value before you expand:
If you are new to Unit4, find out more about AI at Unit4 at:
Next in the series
Once the AI understands your world, the question becomes what it is allowed to do. We turn to control, and why in services it is the difference between a margin that holds and one that leaks.
Sources
2. Claus Jepsen, Chief Technology Officer, Unit4, "From routing to reasoning," Diginomica.
6. SPI Research, 2026 Professional Services Maturity Benchmark (Fall 2025 survey of 509 firms).
7. Source Global Research, the market for AI-related consulting serviceshttps://sourceglobalresearch.com/blog-post/understanding-the-market-for-ai-related-consulting-services, 2025.