Put AI where it moves the number

The problem it solves

The conversation about AI usually starts with the tool and ends in a pilot nobody uses. The right order is the reverse: first the process that hurts, then whether AI is the answer, and often it is not.

What I do

I look at where time goes in your operation and point to the place where AI genuinely moves the number. I work in two areas with real projects behind them:

• Documents and language. Rewriting complex texts in plain language, extracting data from documents, classifying incoming requests, drafting repeated replies or searching your own documentation. Example: legal documents you understand first time.
• Computer vision. Counting, detecting or measuring from video, continuously. Example: platform occupancy where there was no data at all.

We pick one case. What “working” means is defined before any code is written, and it goes into production with evaluation and monitoring in place, so you know whether it is still working three months later and what each operation costs.

If the case does not justify the cost, I tell you in the feasibility study and that is where it ends.

Scope

• Included: a feasibility study, one use case taken to production and its operational set-up: deployment, evaluation and monitoring.
• Not included: training models from scratch when an existing one solves the problem, or pilots with no intention of reaching production.

Deliverables

• A feasibility report with the success criterion in writing and the estimated cost per operation. If the answer is that it is not worth it, the report says so.
• An evaluation set built from real examples of your process, reviewed by someone on your team.
• The system in production, in your infrastructure or your cloud account.
• Quality and cost monitoring, with an alert if quality drops.
• Documentation and handover to your team.

How it is measured

The criterion is agreed and written down before starting: cycle time, percentage of cases resolved without intervention, accuracy against a manual review, or cost per operation. It is measured in production, with real data, not in a demo.

What I need from you

Access to a representative sample of the real data from the process, and to someone who can tell a good answer from a bad one.