In 30 seconds

A good demo tells you the vendor can make the product look good. It does not tell you what happens to your data, what the price becomes at your volume, or who answers when it breaks. Choosing a vendor is mostly about the boring parts the demo skips. Judge those deliberately.

Look past the demo

Every AI demo is run on a friendly example. Ask to see it handle a messy real case from your own work, including one where the right answer is "I do not know." How a tool fails tells you more than how it succeeds.

The five things that actually matter

1. What happens to your data

The most important question. Does the vendor use your data to train their models? Where is it stored? Can you delete it? Start with does the vendor train on your data and do not accept a vague answer.

2. Reliability and support

When the agent gets something wrong at 4pm on a Friday, who do you contact and how fast do they respond? Ask for their support terms in writing, not just a reassurance.

3. The real price at your volume

The advertised price is often a starting tier. Ask what it costs at the volume you actually expect, and what happens when you exceed a plan. Surprise usage bills are a common regret.

4. Lock-in and exit

If you leave in a year, can you take your data and your configuration with you? A vendor who makes leaving hard has less reason to keep earning your business.

5. Track record

How long have they run this, and can they name a business like yours that uses it? You do not need a giant reference, just evidence they have done it before.

Write the risks down

Whichever vendor you pick, you now depend on their uptime, their pricing, and their data practices. Record that dependency so a change on their side, such as a model update or a price rise, is something you monitor rather than discover the hard way.

What this does not solve

Choosing well does not remove your responsibility for the workflow. You still own the boundaries, the review, and the outcome. The vendor supplies the engine; the specification is yours.

Good enough to move on when

  • You have seen the tool handle a real, messy case from your own work.
  • You have clear written answers on data use, support, price at volume, and exit.
  • You have recorded the vendor as a dependency to keep an eye on.