Who builds what they recommend?
Some consultancies advise and hand the build to another firm. Others build what they design. When the people who recommend a design are not the people who build it, the problems in the plan are found late, by someone else.
Ask to meet the people who will write and run the software, not only the people who will present the proposal.
What will be working, and where?
A good proposal names one process, the system that will run it and where that system will run. An AI strategy is not something your people can use or test.
Ask what your team will be able to use at the end, on which data and inside which of your systems.
How is the result measured?
Agree a measure before work starts, such as time per case, cases per person or requests completed, and write down where it stands today. Without a baseline nobody can say whether the work helped.
Ask whether the fee depends on that measure. A supplier that is paid for the result has the same interest you do.
Where does your data go?
Ask which cloud and which region the system runs in, who can see your data, and whether it is ever used to train a model. Ask for the answers in writing.
For a UK organisation it is reasonable to expect data kept in-country, a clear list of the models used and no training on your data.
Who stays in charge?
An AI agent that acts in your systems should have a person approving what matters, and a record of every action it takes.
Ask how approvals work, what is logged, who can see the log and how the system is switched off.
What are you left with?
At the end you should control the system, its documentation and its data, and your own team should be able to run it.
Ask about handover and training, and what happens if you stop working with the supplier.
What can they show you?
Ask to see systems the consultancy has built and runs itself, and the certifications that show how it protects data, such as ISO/IEC 27001 and Cyber Essentials Plus.
A supplier that can only show slides and logos is asking you to take the delivery on trust.
Can you start with one process?
The lowest risk way to test a consultancy is one real process, measured against a baseline you agree first. If it works you have evidence for the next step. If it does not, you have found out at small scale.
How Sendient approaches this
We build what we recommend. The first thing we deliver is one workflow working on your own data, in Microsoft Azure in your tenancy or ours, with your data kept in-country and never used to train models. A person signs off what matters and every action is logged.
We agree the measure and the baseline with you first, and our fee is tied to at least 3x improvement on that measure. See First Workflow Live, the AI readiness review and our certifications.
