Free self-check
Ready to build AI into your business?
Ten questions for companies that want AI inside their product or operations: assistants that answer from your own data, automation, scoring. Not just a coding tool. One minute. You get a score, the stage that is holding you back, and what to do about it.
- A scoreOut of 100, with what it means for you.
- Your checklistWhat is in place, what is partly there, what is missing.
- What to doA fix for each gap, with steps to check it yourself.
- A report to shareBy email if you want it, ready to forward to your team.
Your result
0/100
Ready to build
Your foundations are in place: data, trust and production thinking. The work now is building the feature well, fast, and measuring it against the outcome you named.
Build the foundations first
There is a real opportunity, but the gaps below tend to surface halfway through an AI project, when they are expensive. They follow the order we build in: data first, then trust, then the feature.
Start with the data
The AI is the easy part; the foundations under it are missing. Start with the first item below: until the data is reachable and consistent, any model will give unreliable answers.
Where to start
- Do you know which systems hold the data your AI would need?
Map your sources first. An AI feature is only as good as the data it can reach, and most projects discover a missing source halfway through.
- Can that data be reached through APIs or exports, with permission to use it?
Secure access before design: APIs or exports, credentials, and the right to use the data. A blocked source late in a project costs weeks.
Read: Ingestion, normalization, trust: feeding practitioner data to an AI assistant
- Is the same customer, product or case recorded the same way across your systems?
Normalize before you build. When the same customer exists under five IDs, the AI will treat them as five people and answer accordingly.
- Is there one current version of each document or record that counts as the truth?
Decide the source of truth. An assistant that retrieves an outdated document will quote it, fluently and confidently.
- Have you defined what a correct answer looks like, with real examples to test against?
Write the test before the feature: real questions with known correct answers, and a pass mark agreed in advance. Without it, “the AI works” is an opinion.
- Will people be able to see where each AI answer came from before acting on it?
Build provenance in: every answer should show what it was based on. People trust what they can check, and catch what they can see.
Read: How to trust an AI score
- Have you decided what the AI must never do, or never decide alone?
Set the guardrails early: what the AI may answer, what it must refuse, and which actions need a person’s approval.
- Is the AI tied to a specific process and a number it should move?
Pick the outcome before the model. A feature with no number to move can’t be judged, funded or improved.
- Have you compared models on your own examples, by quality and cost?
Choose the model by measured quality on your own data, not by benchmark or price. The cheapest model is often the first one replaced.
Read: The right model, not the cheapest: what a few cents per lesson buys
- Will it run on your own accounts, with monitoring, tests and someone responsible after launch?
Plan production from the start: your accounts, tests, monitoring, failure handling and an owner. That is the difference between a demo and a system.
Read: Production: what a prototype skips and a real system can’t
No gaps on this list. You are ready to build.
Want an engineer to look at your plan?
Send your result to an engineer and get a reply with how we would approach your first AI feature: which data, which checks, which outcome. Free, with no obligation.
Built from real projects
Every question comes from what we have seen go right, and wrong, on real work. See it in practice:
The Data Layer Behind a Practitioner AI Platform →Questions
- Is it really free?
- Yes. The result appears on the page, and nothing is asked of you to see it.
- What happens to my answers?
- They are scored in your browser. We only receive them if you choose to send them to an engineer or ask for the report by email. See our privacy policy.
- Will you add me to a newsletter?
- No. If you ask for the report, we send that one email. We only write again if you reply.
- Who is behind it?
- EVDevs: experienced engineers working with AI on revenue, data and AI systems. About us.