Find where AI actually fits in your business — before you pay to build anything.
A short conversation, then a written assessment: which processes are worth automating, what data is needed, where the risks are, and a small pilot with clear success measures. If AI is not a good fit for your task, I will say so.
What the assessment covers
Six parts, written for decision makers as well as engineers.
Use-case list
The time-consuming processes, an estimate of impact (volume, time, error rate) and a ranking by value and difficulty.
Data and documents
What data exists, where it lives, its quality and sensitivity, and what must still be collected to measure results.
Pilot design
A small scope, acceptance criteria measured on held-out real examples, and a human-review step where it is risky.
Risk and privacy
Where data is sent, who can access it, which model provider is used, and what happens when the model is wrong.
Running costs
An estimate of model costs (API or self-hosted) at your real volume, alongside the build effort.
Roadmap
Next steps at fixed scope: build the pilot, scale it, or stop if the measurements do not justify it.
Common use cases
The requests small and mid-sized businesses ask about most.
- A chatbot that answers customer questions from the business own documents and policies — chatbot & agent page
- Search and question answering over internal documents (procedures, contracts, handbooks)
- Reading invoices, forms and emails into a spreadsheet — custom AI tools
- Sorting tickets, customer feedback and reviews
- Recurring reports compiled automatically from several sources
- Sensor/IoT monitoring and anomaly detection on device data — technical example on the blog
Process
Four steps, each with a concrete output.
1. Short conversation
You describe the business, the process that hurts and what you hope for. I ask about anything unclear.
2. Collect examples
You send a few dozen real examples (sensitive data removed) and describe what a correct result looks like.
3. Written assessment
You receive the six-part assessment above with a pilot proposal at fixed scope and price.
Optional next step: build the pilot, measure it on your data, then hand over source code, documentation and a walkthrough. Payment by PayPal invoice.
Frequently asked questions
Can a small business with little data do this?
Many tasks (extraction, classification, question answering over documents) use existing models rather than training a custom one. What is needed is a few dozen real examples to measure results.
Will my data leave my premises?
It depends on the design: a model provider API or a model running on your own machine. I state where data goes before starting; see the Privacy Policy.
Does AI replace staff?
I design for assistance, with a human-review step where it is risky. Not a fit for automated decisions on health, legal, credit or employment matters without human oversight.
How much and how long?
It depends on scope, so I give a fixed price and timeline once I understand the request. Two or three sentences are enough to start.
Tell me about the process you want to improve
I will reply with clarifying questions and a proposed next step.