Unit4 ERPxUnit4 Enterprise Resource Planning (ERP)Professional Services
Author: Unit4 Communications September 8, 2026 3 min read

Routing, not reasoning: why most "AI" in your firm just moves work around

Person working at a computer workstation in a modern office environment, with multiple monitors and a tablet on the desk.

In our first article we asked why so much AI in professional services changes nothing underneath. The answer fits in a single line from Unit4's CTO, and it is worth sitting with before you buy another tool. 

Most AI added to enterprise software today is not reasoning. It is routing with better language. — Claus Jepsen, Chief Technology Officer, Unit4 (Diginomica)²

Routing feels like intelligence because the language is fluent. But moving a request to the right place is not the same as weighing whether it is the right thing to do. In a services firm that difference is the whole ball game, because a margin is not a task to be routed. It is a decision to be judged. 

Keep reading:

A ladder, not a label 

Jepsen frames a progression that cuts through most AI marketing: 

Workflow systems route. Semantic systems interpret. Reasoning systems evaluate.² 

Each level contains the one before it. A workflow system moves work along a predefined path. A semantic system understands enough about the content to interpret it. Only a reasoning system evaluates: it weighs options, holds uncertainty, and reaches a judgement it can defend. Evaluation is the rung that finally lets software act with something like judgement, rather than simply passing work from one inbox to the next. 

Most of what reaches the market stops on the first rung and dresses it up as the third. It answers quickly and confidently, and it leaves every decision that matters exactly where it was. 

Why this matters more in services than almost anywhere 

Consider a routine question: should this piece of work be written off, or billed? A routing system can find the timesheet, format it, and drop it in front of a partner. A reasoning system can weigh the engagement economics, the client relationship, the precedent it sets, and the margin impact, then recommend a call and show its working. One saves a few minutes. The other protects the number the firm is judged on. 

This is why the "how many agents does it have?" conversation misleads buyers. Gartner expects task-specific agents in 40% of enterprise applications by the end of 2026, yet forecasts that more than 40% of agentic AI projects will be cancelled by the end of 2027, and warns explicitly about "agent-washing."³ A huge number of agents that move work around does not create any increased efficiency. The count is not the capability. 

+40%

of agentic AI projects

will be cancelled by the end of 2027

Independent analysts describe the same split. Forrester, in its Predictions 2026 research, separates narrowly scoped "agentish" helpers that sit inside existing workflows from systems that "bring reasoning to the center," and expects fewer than 15% of firms to switch on the agentic features they have already bought.⁹ In plain terms, most of what is badged as AI still routes. Reasoning is the exception, not the norm. And the buy side is beginning to price that in: Source Global Research finds that help making effective use of AI has climbed from the eleventh most in-demand consulting service to the first in a single year.⁷ Fluent output is not enough when the client is betting on your judgement. 

How Unit4 approaches it 

ERPx is built to sit on the top rung. It reasons over your data with your firm's context already in place, and it reasons probabilistically rather than by rigid rules: it weighs likelihoods, holds uncertainty, and calibrates with every cycle. When confidence is high, it suggests or acts. When confidence is low, it routes the decision to the right owner with its reasoning visible, so a person applies judgement at exactly the moment judgement is needed. 

That is the opposite of a language layer bolted onto a system that only ever recorded what happened. It is evaluation built on meaning, which is the subject of the next article. 

The takeaway 

Before you ask a vendor how powerful its AI is, or how many agents it ships, ask where it sits on the ladder. Does it route, interpret, or evaluate? In professional services, only the system that evaluates can protect a margin, and only the system that shows its reasoning earns the right to be trusted with one. 

The fastest way to tell routing from reasoning is to watch it work on your own data. If you are already a Unit4 customer, you can try our AI for free and judge the difference yourself before you pay for anything: https://www.unit4.com/try-our-ai. If you are new to Unit4, find out more about AI at Unit4 at https://www.unit4.com/ai-native-erp. 

Next in the series: reasoning is only as good as the meaning underneath it. We look at what it takes for AI to actually speak the language of a services business. 

Sources 

2. Claus Jepsen, Chief Technology Officer, Unit4, "From routing to reasoning," Diginomica. 
3. Gartner, agentic AI adoption and cancellation forecasts, 2025. 
7. Source Global Research, analysis of the market for AI-related consulting services, 2025. 
9. Forrester, Leslie Joseph, Predictions 2026: Automation And Robotics, forrester.com, 2025. 

Sign up to see more like this

Recommended blogs

September 3, 2026 5 min read

The leverage problem: why most AI in professional services changes nothing 

Read more
A professional works on a laptop while speaking on a phone in a café or coworking environment.

Popular blogs

Subscribe to our blog

Don't miss the latest Unit4 blogs

Sign up for industry insights & exclusive content