Many technology purchases that disappoint were decided before the first demo. Not because the tool was bad, but because nobody had written down what it was for. When the problem is vague, the vendor defines it for you, and their definition tends to look a lot like their product.
The fix costs almost nothing: answer five questions in writing before you book a demo. If you can't, you're not ready to buy yet.
1. What problem are you solving, and how will you know it's solved?
Write the problem in one sentence, in business terms. "We need a new CRM" is not a problem. "Sales follow-ups depend on each rep's memory, and we lose quotes we should have won" is.
Then name the sign that it's fixed: a measure you can check, a task that disappears, a delay that shrinks. Write down where that measure stands today. Athletes test before a training block for the same reason: without a starting point, every tool looks like a success after six months, and none can prove it.
2. What changes in the way you work, and who owns that change?
A tool doesn't change a process. People do, and the tool makes the new way easier or harder. Before buying, describe the process as it will run once the tool is in place: who enters what, who approves what, what stops being done.
Then name one person on the business side who owns that change. Not IT, not the vendor. If no manager is willing to own it, the project has already told you something.
3. What data does it need, and is that data good enough today?
Every system runs on data you already have, in a state you may not like: customer records with duplicates, item codes that mean different things in different plants, spreadsheets that live on one person's laptop.
For AI tools, this question matters even more. A model fed messy data produces confident, wrong answers. Ask where the data comes from, who keeps it clean and how much work it will take to get it into shape. That work is part of the project, whether or not it shows up in the vendor's quote.
4. What will it really cost over three years?
The licence is the visible part. The rest includes integration with your other systems, data cleanup, configuration, training, support, price increases at renewal and the time your own team will spend. That time isn't invoiced, but it's real.
Build a three-year total cost, even a rough one, before comparing options. Two tools with similar licence prices can end up far apart once integration and internal time are counted. And put a figure on doing nothing: sometimes it's cheaper than it looks, sometimes much more expensive.
5. How do you get out?
This is the question vendors are least eager to answer, and the one buyers ask least. Can you export all your data, in a usable format, at any time? What does the contract say about term, renewal and termination? What happens to your configuration and your history if you leave?
For AI tools, add two more: what happens when it's wrong, and who checks? Decide in advance which outputs a person must review before they're used, and what you'll do if quality drifts. A tool you can neither leave nor verify is a risk, not an asset.
The rule
If you can't answer at least four of these five questions in writing, don't book the demo yet. Spend a week on the answers instead. The demo will be shorter, sharper and much harder for a vendor to steer.
How to use the answers
The answers aren't just homework. They become the brief you send each vendor before the demo. Share the problem statement and the future process, and ask every vendor to show how their tool handles that process, with an example of your own data if possible. Put the cost and exit questions in writing, and ask each vendor to answer them in writing too.
Then compare the answers side by side, not the demos. Demos are built to impress; written answers to the same questions are much easier to compare, and much harder to blur.
Checklist before the first demo
- The problem, in one sentence, in business terms
- The measure that will show it's solved, and its current value
- The future process, described step by step
- A named business owner for the change
- The data required, its source and its current state
- A three-year total cost estimate, including internal time
- The cost of doing nothing
- Exit terms: data export, contract term, renewal, termination
- For AI: which outputs need human review, and who does it
Where to start
The first question is almost always the hardest. Once the problem is clear, the other four go faster, and the list of candidate tools often shrinks on its own. If you're facing a decision like this and want a second pair of eyes on your answers before the demos start, a 30-minute call is a good place to begin.