Most bad AI projects are visible in the proposal, if you know which sentences to read twice. Here are seven that should slow you down.
The seven
-
1. A day rate with no fixed outcome
You are carrying all the delivery risk and the supplier earns more the longer it takes. For a defined system, this is the wrong instrument.
-
2. Open-ended discovery
A discovery phase with no fixed fee and no defined output is you paying the supplier to learn your business, with no guarantee of a plan at the end.
-
3. No named integrations
If the proposal says integrates with your existing systems rather than naming Sage, or Xero, or the 2011 database with no API, nobody has looked yet. The awkward one is where the cost is.
-
4. The word roadmap doing heavy lifting
Phase one is a discovery. Phase two is a pilot. Phase three is the thing you actually wanted. Ask what you have if you stop after phase one, and watch the answer.
-
5. No running cost
A build price with no monthly figure is half a quote. Model calls, hosting and monitoring are real. Ask what year two costs.
-
6. No mention of failure
Every automation fails sometimes. If the proposal does not say who gets the alert and what queues for a human, nobody has designed for it, and you will find out live.
-
7. You cannot take the work with you
If the workflows live inside a platform only they can access, you are renting a capability. Ask directly: if we part company, what do we keep?
The question that cuts through all seven
Ask: what exactly do I own at the end, and what does it cost me to run it next year? A supplier who has thought properly about your problem answers that in a paragraph. One who has not will talk about partnership.
What good looks like instead
Every integration named. The payback arithmetic with your own numbers. A fixed build quote and a delivery date. The monthly running cost broken out. A named person who gets the alert. If those five are present, the rest is detail. See how we quote.
Questions we get asked
Is a phased approach always a red flag?
No, phasing genuinely risky work is sensible. The flag is when every phase is preparation and none of them deliver something that works on its own.
What if they will not give a fixed price?
Ask why. If the honest answer is that the scope is not clear enough yet, that is fair, and the fix is a paid scoping step with a fixed fee and a document you keep.
Should the proposal include the arithmetic?
Yes, with your figures in it. If a supplier cannot show why the build pays back, they are asking you to take the business case on faith.
How long should a proposal be?
Short. The good ones are a few pages of specifics. Length is usually a substitute for the specifics being missing.