The demo is not the product
Demos show the happy path on clean data with a practised operator. Everything that decides whether a tool works for you — your data, your volume, your integrations, your exceptions — is outside the demo by construction.
The useful counter-move is cheap: ask to drive it yourself, on a trial account, using your own data. What takes the salesperson four clicks frequently takes a new user considerably longer, and that gap is what your team will live with daily.
1. How do I get my data out?
Full export, on demand, in a usable format, including history and attachments. A vendor who makes leaving hard is charging a switching cost that never appears on the invoice, and it decides whether you can build on that data later.
Ask them to demonstrate it rather than confirm it. “Yes, there’s an export” covers everything from a complete API to a partial spreadsheet missing attachments, comments and history — and you find out which during the migration you were forced into.
2. What does it cost at three times this volume?
Per-seat, per-record, and per-transaction pricing behave very differently as you grow. Ask for the number at the size you expect to be, not the size you are.
Also ask what happens at renewal, and whether there is a cap on increases. A tool that becomes structural to your operations has enormous pricing leverage over you at renewal, and that leverage is worth negotiating before you have handed it over.
Ask what it costs when you succeed. That is the price you will actually pay.
3. Which parts are on the roadmap rather than shipped?
Anything demonstrated that does not yet exist should be identified explicitly and, if you are relying on it, written into the contract with a date. This has become sharper as products acquire AI features quickly — the label carries little information now.
4. What happens to my exceptions?
Every business has cases that don’t fit the standard model. Ask how the product handles yours; the answer “you can use a custom field” usually means you will run the real process in a spreadsheet beside the tool — which is itself the specification for what you should have built.
Bring three real exceptions to the demo and make them work through each one. It is the single most informative twenty minutes of any evaluation, and the response tells you whether you are the awkward shape this product was not built for.
5–8. The ones people forget
Who owns the integration work and how long has it taken others; what the support response time is contractually rather than aspirationally; where data is stored and under whose jurisdiction; and what the last two years of incident history look like. All four are answerable, and reluctance is itself an answer.
On support, ask what a real ticket looked like recently — response time, resolution time, and whether it needed escalating. Contractual targets describe the worst they are permitted to do, not the experience you should expect.
Two questions about the ending
What is the notice period, and what happens to your data after termination — how long they keep it, and whether you can still export during the wind-down. Both are trivial to agree at signing and impossible to change once the relationship has soured.
Ask a current customer, not a reference
References are selected. Find a user of comparable size through your own network and ask what they would do differently, what took longer than expected, and whether they would buy it again.
That conversation is worth more than the entire evaluation process around it, and the third question is the one that produces the honest answer — people who would not buy again rarely volunteer it, and almost always say so when asked directly.