Every pitch sounds the same now
You will hear a version of the same pitch several times this quarter. Sometimes from a vendor, sometimes from a consultancy, sometimes from an ambitious team inside your own organisation. The words change. The shape does not: a striking demo, a promise of transformation, a sense that everyone else is already doing it, and a deadline.
None of that is evidence. A demo shows the best case on prepared data. "Transformation" describes an aspiration, not a result. And the feeling that competitors are ahead is exactly the pressure that stops leaders asking dull questions, which are the ones that matter.
This lesson gives you those dull questions. Use them on outside vendors and internal proposals alike. A good idea survives them easily.
The six questions
1. What problem, exactly? Ask for the specific process and the specific pain, described without mentioning AI at all. If the answer only makes sense with the technology in it, the proposal has started from the tool, not the problem.
2. Measured how? What is the current performance of that process, and what number will change? Time to complete, error rate, backlog, customer wait. If nobody has measured the current way, there is nothing to compare against, and any later claim of success is a guess.
3. What happens when it is wrong? Not if. When. Ask for the failure case specifically: what does a bad output look like, how often might it happen, and what does it cost when it reaches a customer, a regulator or a court?
4. Who checks, and when? A named role, at a named point in the process. "The system flags low-confidence results" is not an answer on its own, because the dangerous errors are the confident ones.
5. What data leaves the building? Where does your information go, who can see it, is it kept, and is it used to train anything? Get the answer in writing, in the contract, not in a slide. The Safe AI for Teams course covers the staff-level rules that follow from it.
6. What does it cost to leave? If this does not work, or the supplier changes terms, how do you get your data and your process back? Exit costs are the ones nobody mentions in the first meeting.
Ask the questions in writing and read the answers side by side. Vague answers look far more vague on paper than they sound across a table.
Checkpoint
Six questions cut through any pitch: what problem, measured how, what happens when it is wrong, who checks, what data leaves the building, and what it costs to leave.
Reading the answers
The answers matter, but so does the reaction. Serious suppliers and serious internal teams answer these questions plainly, including "we do not know yet", because they have been asked them before. Weak pitches respond with a new demo, a customer logo, a reference to how fast the field is moving, or an offer that expires at the end of the month.
| What you hear | What to ask next |
|---|---|
| "It saves hours every week" | Measured against what, on whose process, including checking time? |
| "It is highly accurate" | On which cases, and what happens with the ones it gets wrong? |
| "Your data is secure" | Where exactly is it stored, who can access it, and is it used for training? |
| "Everyone in your sector is adopting this" | Which problem did it solve for them, and how do you know? |
| "The price goes up next quarter" | What is the cost to leave, and what does the contract say? |
Internal proposals deserve the same treatment
It is tempting to be gentler with your own people. Resist that, kindly. An internal team that has not answered these questions is walking into the same risks as a vendor that has not, and they will thank you later for being asked early rather than after launch.
Make the six questions the standard template for any AI proposal that reaches you. Teams stop bringing you demos and start bringing you problems, which is exactly the shift you want.
โ Weak prompt
Prompt
Is this AI product worth buying? [pastes the vendor brochure]
Output
This product offers strong capabilities in automation and analytics, and could deliver significant efficiency gains for your organisation.
It summarised the marketing back to you. A brochure contains no failure cases, no exit costs and no data terms, so neither does the answer.
โ Good prompt
Prompt
Here is a vendor proposal. Do not tell me whether to buy it. List what it says, and what it does not say, under six headings: the specific problem, how success is measured, what happens when it is wrong, who checks the output, what data leaves our organisation, and the cost of leaving. Where a heading has no answer in the text, write NOT ADDRESSED.
Output
Problem: described only as 'operational efficiency'. Measurement: NOT ADDRESSED. Failure case: NOT ADDRESSED. Data: states encryption but not storage location or training use. Exit: NOT ADDRESSED.
It shows you the gaps rather than the gloss. Those gaps become your questions for the next meeting.
Below is an AI proposal from [a vendor / an internal team].
Do not recommend for or against it.
Under each heading, quote what the proposal says, or write NOT ADDRESSED:
- The specific problem and process it targets
- How the current process performs today, and what number will change
- What happens when the output is wrong, and what that costs
- Who checks the output, and at what point
- What data leaves our organisation, where it goes, and whether it is kept
- The cost and process of leaving
Then list the five questions I should ask next, most important first.
PROPOSAL:
[paste it]
Do not paste confidential contract terms or commercially sensitive figures into a tool your organisation has not approved for that kind of material. Remove names and numbers first if you are unsure.
๐ Quiz
Question 1 of 4Why should a proposal be able to describe its problem without mentioning AI?