Why I Didn't Make It Fully Autonomous

Two days ago I shared the Collection Engine I built.
The most common question I got back was not about the €336K it modelled. It was: why didn't you make it fully autonomous?

Fair question. Here is the framework behind my answer: what is real in AI collections, what is hype, and what I would tell any finance leader to do now.

First, the pressure is real. Allianz Trade's 2026 Insolvency Report puts global business insolvencies up another 6% this year. Their 2026 Collection Complexity Score sits at 47.2/100, a "High" reading. So the temptation to throw AI at it is understandable. But AI in collections is not one thing. The question is not should we use it? It is where does it actually help the collector decide better, faster?

Three categories.

1. AI that works today
Prioritisation. Risk scoring. Payment behaviour analysis. Next-best-action.
Real value: removing the first layer of noise.
Not theory: Coface already runs an AI-built score that calculates a company's probability of default from financial and non-financial data, and moves credit limits dynamically as payment incidents and sector stress shift. The largest credit insurers are already doing the "real" part.
It answers the three questions every team faces every morning: Who needs attention first? Why is this customer risky? What happens next?
This is where finance leaders should start.

2. AI that is overhyped
Fully autonomous collections. (The answer to the question I was asked two days ago.)
The pitch sounds clean: detect the risk, contact the customer, negotiate, resolve the dispute, update the ERP, close the case.
B2B collections is not that clean. Disputes. Pricing gaps. Credit notes. Relationship risk. Legal sensitivity.
AI can prepare the case, suggest the action, trigger the workflow. Giving it full ownership today is not maturity. It is risk dressed as progress.

3. AI that is coming, but NOT ready
Real-time dispute resolution: the system instantly checking contract, PO, delivery proof, pricing, credit-note history, approval flow.
A genuine breakthrough, when it works. Most companies are not there. The blocker is NOT the AI. It is Data Quality, Process Ownership, Integration, Accountability. AI cannot resolve a dispute in real time if the business cannot say where The Truth sits.

So what should finance leaders do now?
Not all-in. Not ignore it. Start selectively, where the data is good enough, the decision repetitive enough, the impact clear enough.
Use AI to prioritise, explain, recommend, route. Keep humans on judgment, negotiation, escalation, relationships.

The winner will not be the company with the most AI. It will be the one that knows exactly where AI belongs.
Not autonomous collections at any cost. Intelligent Collections, designed around real business decisions.

Where do you see AI creating real value in collections today, and where do you still see more hype than substance?