Every AI vendor demos well. Demos are built to. The questions that predict whether a tool works for you are mostly not about capability.
- ›The five questions that matter most
- ›Why the data questions come first
- ›Evaluating without a long process
- ›Signals to walk away
The five questions
1. Where does our data go, and what happens to it?
Ask explicitly: is our data used to train models that serve other customers? Where is it stored? What is the retention? Get it in writing, because verbal assurances during a sales process are not a contract.
2. Can it see enough to be useful?
AI quality is bounded by context. A tool with access to one system will answer questions about that system. If your questions span departments, evaluate whether the tool can reach across them. See AI agents for business.
3. What can it do without asking, and can we see it?
If the tool takes actions, ask for the audit view during the evaluation, not after. "Show me everything the AI did yesterday, and undo one of them." Vendors who built this will demonstrate it immediately.
4. What does it cost as we grow and as we use it more?
Two separate questions. Per-seat pricing scales with hiring; usage pricing scales with use. Model both against your actual expectation. See usage-based AI pricing.
5. Can we leave?
Full export, in a usable format, without asking permission. This single question makes every other decision reversible.
Why the data questions come first
Because they are the ones you cannot fix later. A capability gap is an annoyance; a data-use term you agreed to and cannot unwind is permanent. Ask before the demo, not after the contract.
Evaluating without a long process
For most companies under a few hundred people, a two-week evaluation with real data and real users beats a formal scorecard process.
Week one: load your actual data, configure the one workflow you most want it to handle, and have two or three people use it for real work.
Week two: try the things that go wrong. A bulk action, a permissions question, an export, an undo. How a tool behaves when things go sideways is more informative than the happy path.
Then answer one question: did the people using it ask to keep it?
Tip: Evaluate with your own messy data, never with the vendor's demo data. Demo data is arranged to look good and hides exactly the problems you will hit.
Signals to walk away
- ›Cannot or will not answer the data-use question directly.
- ›No audit trail for actions the AI takes.
- ›Export is restricted, gated, or a paid feature.
- ›Pricing is not published and the sales process requires a scoping call before you see the product.
- ›The demo cannot be reproduced with your data.
FAQ
Should we buy point AI tools or a platform?
Depends on whether your questions cross departments. See business OS vs point solutions.
How many tools should we run?
Fewer than vendors suggest. Each one is context the AI does not have and a bill you pay.
What if we pick wrong?
If you can export everything, you can leave. That is why question five matters more than it looks.
Brainis is one data layer with AI across every module, full audit and undo, complete export, and usage-based pricing. See pricing.
Sharing insights on business operations, AI, and modern team management.
Run your company on Brainis
All 11 Operating Systems on every plan, from $29 a month. No per-seat pricing — you pay for AI capacity, not headcount.
See pricing