How do you approach it?
Ask every candidate four questions:
- where do you start?
- how small can the first step be?
- how will we know in a month whether it works?
- what can we do ourselves once you are done?
When you listen to the answers, pay attention to who asks follow-up questions. A partner worth having asks a lot about your work. A salesperson starts talking.
Where do you start?
The only good answer: on the floor. An AI partner who spends a morning watching your customer service team knows more about your process by lunch than a steering committee learns in three months. The wrong answer is easy to spot: a demo of their own platform without a single substantive question about your organisation.
How small can the first step be?
A good partner wants to start on one process quickly, preferably low effort and high value, and wants it working fast. Think sorting and answering incoming customer email, or preparing sales conversations. They do that to get to know you, so that later in the relationship they can judge far better how to help you further. After a first assignment like that, a partner can tell whether your organisation is ready to change and whether the team is ready to adopt it. What you do not want to discuss at the start: an eighteen-month programme with a seven-phase roadmap.
How will we know in a month whether it works?
A partner who genuinely understands that AI is about organisational goals rather than technology will set KPIs with you for every project. Experimenting without KPIs is nothing more than very expensive innovation theatre. Put more sharply: if you do not know what you want to achieve, every experiment is pointless before it starts. Show the agency that wants to define project KPIs in a later phase to the door.
What can we do ourselves once you are done?
Handover is routinely underrated and matters enormously. A good partner moves you forward and works towards making itself unnecessary. Once your AI application is running, your team works in a new way, the KPIs are being met and the documentation is complete, all maintenance should be transferable. Lock-in at the start of your AI adoption is unwise, because you do not yet know what you will need. So: your partner should have to earn the next assignment. A licence, a retainer or a prepaid block of hours up front is a red flag, because that agency and that account manager will not be leaving any time soon. In short, the right partner makes itself redundant.