Choosing an Agent Persona.

Short answer: the first agent persona you deploy should satisfy five constraints: a narrow scope, a clear escalation path, an obvious owner, a measurable success metric, and a rollback that takes less than one hour. If your candidate use-case fails any of these, pick a different use-case before you pick a vendor.

Why first-persona choice matters more than vendor choice

Most enterprise AI deployments stall not because the vendor was wrong but because the first persona was the wrong shape. Picking an unbounded use-case ("an agent that helps sales") guarantees scope creep, unclear ownership, and a public failure mode. Picking a narrow, well-owned use-case ("an agent that drafts quarterly customer health summaries for the CS team") gives you a measurable win in four weeks and a template for the next persona.

The five questions, in order

1. Can a human do this job today in under 30 minutes?

If yes, you have a bounded task with known inputs and an obvious quality bar. If no — either the task is poorly scoped or it requires judgement an agent shouldn't own yet. Pick something a human currently does.

2. Who escalates when the agent says "I don't know"?

Name the human, by name, before naming the persona. If no human can comfortably escalate to, the agent shouldn't be deployed yet — there are decisions about cases the agent can't handle that nobody has authority to make.

3. What system of record does the agent write to?

Read-only first agents are easier to deploy but produce less value. Write-capable agents need explicit per-action authorisation logic. Decide in advance which write classes are auto-approved, which require human sign-off, and which are always blocked. This is the audit boundary your DPO or auditor will care about most.

4. What does success look like, numerically?

"Customers are happier" is not a success metric. "Average ticket triage time drops from 12 minutes to 4 minutes within 30 days, with same or lower escalation rate" is. If you cannot write the success metric on a Post-it before deployment, you will never know if the agent worked.

5. How fast can we turn it off?

Define the rollback procedure as a sequence of named actions, each assigned to a named person. Target: cold rollback in under one hour. Anything longer means the agent is too tightly bound to a critical workflow for a first deployment.

The buyer's decision matrix

Score your candidate persona out of 5 against these questions. Each "yes / clearly answered" is one point. Useful interpretation:

• 5/5 — proceed; this is a good first persona.
• 4/5 — proceed, but fix the missing constraint before kickoff.
• 3/5 or below — rescope. The use-case is probably the second or third agent, not the first.

Where AITG fits

Scoping the first persona is where most procurement conversations should start — and where most vendor pitches skip past. AITG's scoping conversation runs this matrix against your candidate use-cases as a free two-conversation engagement before any contract. Tell us your top three candidate personas and we'll score them with you.