Short answer: Hire an AI automation specialist when repeatable work is getting lost between tools or people. Evaluate candidates against one real workflow and pay for a small pilot that includes failure tests and a handover.

What a specialist does and when to hire one

A specialist maps the current process, connects applications, tests exception paths and helps staff operate the system. Hire when the manual steps are understood but the team cannot connect them reliably. If nobody owns the process or nobody can agree on the desired output, start with AI Consulting before commissioning a build.

Freelancer, agency or in-house?

  • Freelancer: suits one clearly bounded integration when you have a staff owner and can arrange backup support.
  • Agency: useful when discovery, several integrations, training and ongoing support need coordination. Ask who will actually build and maintain the system.
  • In-house: makes sense with a steady backlog and daily ownership needs. Budget for documentation, backup coverage and ongoing training.

Skills and evidence to look for

Ask for a workflow diagram, sample input and resulting record. Have the candidate explain credentials, field mapping, retries and how staff recover a failed run. Ask what they personally built and which parts were run in production. For an illustrative feedback workflow, ask to see what triggers the request, where the reply is recorded and how staff receive the handoff.

How to evaluate custom AI agent experience

Ask for a demonstration with ambiguous input, conflicting sources and an action that requires approval. A capable candidate can show what the agent reads, which tools it can use, when it asks a person, and how a reviewer traces an outcome. Ask them when they would use rules instead. See Custom AI Agents for the difference between the two approaches.

Interview scorecard and questions

  • Process judgement: Which system owns the record? Which decisions stay with staff?
  • Integration design: What happens if an app is unavailable? What prevents a retry from sending a second message?
  • Evidence: Can you show a relevant build, your contribution and the failure cases tested?
  • Ownership: Who controls billing, credentials, exports and documentation when the engagement ends?

Watch for guarantees of hands-free operation, a demo that only shows the successful path, or a quote that groups software subscriptions and ongoing support into a single unexplained fee.

Who should own accounts, data and automations?

Agree in writing that the business controls its production accounts and customer records. Give the provider only the access needed and agree how it will be removed. Ask for workflow exports where supported, documentation, a recovery runbook and a handover session. Clarify any licensed components you cannot transfer.

Structure a paid pilot and compare proposals

Pick one outcome: a website form creates or updates a CRM record and assigns a follow-up. Test duplicate submissions, incomplete requests and an unavailable CRM. Agree that staff can see the failure, recover it and avoid duplicate messages. Use synthetic records until access and handling rules are agreed.

Give each provider the same pilot brief. Compare acceptance tests, deliverables, exclusions, one-time and recurring costs, support response expectations and the cost of changes. For cross-border work, confirm the billing currency and the hours when both teams can respond.

Take the checklist to your interviews

Download the hiring checklist (plain text). Share it with the person who will own the workflow.

Start working with Lucy Flow Lab

Bring one repetitive process, the applications involved and examples of what goes wrong. We can help you decide whether a workflow automation pilot or an agent is a sensible first step.