Hiring a workflow automation consultant: questions that show how the work will be done

Ask whose accounts the workflows will run under, and who will hold the credentials for your CRM, inbox and accounting system. The answer you want is that your business owns the automation platform account and the credentials, and the consultant works with access you grant and can revoke.

Be cautious if the automation will run inside the consultant's own account or if credentials will sit in their personal workspace. That arrangement can make it hard to leave, hard to audit, and hard to rotate keys when someone departs.

A useful proposal names the workflow, the systems it touches, the cases it handles and the cases it deliberately does not. It also states acceptance conditions: what has to be true for the work to be considered complete. "Lead follow-up automation" is a topic. "New web form leads create a CRM contact, assign an owner by territory and send the approved first reply, with duplicates routed to a review list" is a scope.

Ask how they will test it. Testing against a few clean examples is not enough. Ask whether they will use real, messy cases from your business, including duplicates, missing fields and the odd requests your team handles every week.

Every automation fails sometimes. Ask how failures will be detected, who will be told, and what information the alert will contain. Ask what happens to a request that was half-processed when a step failed, and whether retrying could send a duplicate message or create a duplicate record.

A consultant who has run automations in production will have a concrete answer. One who has mostly built demonstrations may not have thought about it yet.

If the proposal includes AI steps, ask what each one decides. Classifying an email, extracting fields from a document or drafting a reply are reasonable uses when a person reviews the result or when a wrong answer is cheap to fix. Approving payments, changing customer records or sending client communications without review deserve more caution.

The NIST AI Risk Management Framework core includes a Govern category for risks arising from third-party software and data, including contingency processes for failures in third-party AI systems. You do not need a formal framework for a small workflow, but the underlying question is fair: what is the plan if the model or the service behind it behaves differently next month?

Pricing models differ. Fixed scope gives you a known cost for a defined result. Retainers give you ongoing capacity but can blur what you are paying for. Free first builds lower the barrier to starting, and it is worth asking what the ongoing arrangement will be after the first build. None of these is wrong, but you should understand what you are buying.

Ask what is excluded. Maintenance, new workflows, changes when a connected system updates its API, and credential renewals are common sources of surprise if they were not discussed.

Ask for a description of the handoff. At a minimum, I would expect a list of every workflow and what it does, where it runs, the credentials it uses and who owns them, how to pause it, what to do for the most likely failures, and evidence from testing. Ask whether your own team could operate it if the consultant became unavailable.

If the consultant's answer is that you will always need them to operate it, that may be acceptable for some businesses, but it should be a choice you make knowingly.

For some teams, meeting in person helps, especially at the start when someone needs to watch how work actually moves through the office. We work in person across the Columbus, Ohio metro and remotely across the United States, Canada and the European Union. What matters more than location is whether the consultant will look at your real process and your real data before proposing a solution.

Ask the consultant to describe, in writing, how they would handle one specific messy case from your business. The answer will tell you more than a portfolio. Look for specifics about data, ownership, failure and review, and for honest statements about what they would not automate.

If you want our answers to these questions for your own workflow, describe the process, the systems involved and what goes wrong today.