AI changed the buying committee inside the firm, not only the buyer outside it

External buyers changed first in the slides. Internal buyers changed first in the meeting that never quite ends. A vendor demonstrates a model. Marketing sees throughput. A partner asks what happens if a client complains. IT asks where the prompt goes. Everyone leaves agreeing AI matters. Nobody leaves with a decision. Six months later staff are using public tools unofficially, and the partnership still has no trail.

AI did not only change how clients shortlist firms. It changed the committee that has to approve anything that touches client data, public claims or measurement. SME professional services firms feel this acutely: one bad tool choice can freeze the whole programme because there is no enterprise architecture team to absorb the mess. The adjacent public-site argument is Which AI tools a regulated practice can defend. This piece is about the people in the room.

What does each seat actually need answered?

Partners need defensibility. What can we say we used, on which class of work, with what human review, if a regulator or client asks later. Marketing needs measurement that survives assistants and zero-click behaviour, not a dashboard that pretends the blind spot is not there. IT and risk need destination, retention, training use, contractual cover and an evidence trail. When those three questions stay in three rooms, demos multiply and decisions die.

Two follow-ups people usually ask next:

Can we separate “AI for marketing” from “AI for the practice”?

You can separate use cases. You cannot separate governance. Public models do not care which budget code pasted the text. One policy, many cleared workflows.

Who chairs the committee?

Someone who can refuse. In many SMEs that is a managing partner with IT in the room and marketing as sponsor. A vendor should not chair by charm.

Map the questions onto real workstreams

Website and content work answer how the firm presents itself to humans and assistants. Claims must be signed. Proof must not be invented. That is Content and Editorial and, where structure blocks quotation, Website Development. Tooling work answers where data goes for internal workflows. That is AI Governance and Assurance and AI Operations once something is cleared. Measurement work answers what the partnership can still see. That is Analytics and Measurement, with the blind-spot context in Your analytics cannot see AI assistants.

Foundations is the stage name for a reason (Foundations). Clear what you can clear. Decline what you cannot defend. Write the trail as you go. Waiting for a perfect policy while unofficial use spreads is the higher risk.

How programmes stall after the demo

The demo answers none of the three seat questions. It shows capability. Partners hear risk. Marketing hears delay. IT hears another shadow IT story. Someone proposes a working group. The working group collects screenshots. Staff keep pasting. The next vendor arrives with a brighter deck.

Break the loop with a written map: use case, data class, destination, retention, human review, owner, evidence location. If a row cannot be filled, the workflow is not cleared. That feels slow beside a demo. It is how you avoid becoming the firm that learns governance from an incident.

Visibility and conversion spend still depend on this spine. Search and AI Visibility on a site full of unsigned claims is reputational risk dressed as marketing. Price bands sit on Pricing.

If your firm stalls after every AI demo, start a conversation. We will say whether the next job is governance, measurement, public pages, or a refusal.

Key takeaways

• Internal AI buying now needs partner, marketing and IT questions answered together.
• Demos show capability. They do not clear defensibility, measurement or data paths.
• Map use cases to evidence before unofficial volume grows.
• One uncleared tool can block an SME programme. Decline early when you must.

The practical move is a single written page per use case that names the data path, the person who signs it off and the evidence that would satisfy the regulator. Three people reading the same page beats three separate conversations.