What an AI Workflow Review Should Deliver

Choose one process you can describe

“Help us use AI” leaves too many decisions open. A more useful starting point is “Understand how a new job inquiry reaches the person preparing the estimate.” That gives the review a beginning, an ending, and people who can explain the work.

Before paying, agree which workflow is included and where it stops. Decide who needs to attend, what examples are appropriate, and what you will receive. Name the business question you want answered: Where does ownership become unclear? Which missing details cause a request to return? Is the current approach already adequate?

The reviewer should be able to explain how those questions fit the promised service.

A map that someone can use

A current-state map should show the trigger, required inputs, main steps, decisions, handoffs, and finish. Add responsibility at each step. Include the route taken when information is missing, a request is unusual, or the next person is unavailable.

Read the map with someone who knows the process. Can they point to where they take over? Can they explain how they know a request is ready? If an arrow says “follow up,” ask who does it, where the status is recorded, and what returns the request to the main path.

Keep the map readable enough to use in a conversation. A diagram crowded with every possible tool feature makes it harder to check the actual work.

Make the evidence and unknowns visible

Ask how each finding was established. A step described during a kickoff, an example walked through together, and a live system test provide different kinds of evidence. The review should identify which were used and what remains unverified.

Suppose someone reports that estimates often stall because job details are missing. A map can show where that might happen. Establishing how often it happens, how much work it adds, or whether a change helps requires appropriate observations and a separately agreed method.

Watch for precise savings attached to vague evidence. Ask what baseline supports the number, what period it covers, and whether review and correction time were included. An honest “we have not measured that” helps you make a better decision.

One recommendation with a reason

A useful recommendation states what to do next, why it fits the finding, and what could change the advice. It should be small enough that you can understand the decision without becoming a software specialist.

Here is a fictional example. A service business agrees that an estimate request needs a job description and location before the estimator takes over. The next step could be a shared intake checklist and a named person responsible for completeness. Before claiming that it improves anything, the business would need to check whether missing details are a recurring problem and assess any agreed trial.

Other reasonable outcomes include keeping the existing process, clarifying a responsibility, or considering a separately scoped improvement. AI belongs in the discussion when a defined task and suitable inputs justify evaluating it. Anthropic similarly recommends starting with the simplest workable approach and adding complexity when it is warranted in its guide to building effective agents.

A clear boundary around the next decision

You should be able to finish the review and keep its deliverables without buying a build. If implementation is proposed, ask for its own scope: changes, access, data handling, testing, acceptance conditions, maintenance responsibility, and price.

For a possible AI assistant, also define what it may read, draft, change, or send; which actions need approval; and when it must stop. A promising demonstration leaves those operating questions open until they are explicitly addressed.

If the recommendation is to gather more evidence, name the question that evidence should answer. That gives you a useful next decision instead of an open-ended invitation to buy more work.

What AMH’s $500 review includes

The One-Workflow Review covers one agreed process: a 45-minute kickoff, a visual workflow map, one practical recommendation with its reasoning and limits, and a 30-minute findings review with Mike Bradway.

Delivery is within five business days after the kickoff and receipt of all agreed inputs. Your agreement confirms the dates. Fit and scope come first; work begins after the agreement is signed, payment clears, and kickoff is scheduled.

The review excludes implementation, integrations, system access, customer or employee-data review, and ongoing support. Examples must be blank, fictional, or appropriately de-identified and agreed in the scope. Any later build has a separate scope and decision. Business savings are not promised.

Ask these questions before you buy

• What is the exact workflow boundary?
• What do I receive, in a form I can keep and use?
• How will described steps and verified findings be distinguished?
• What information can I safely bring?
• Which activities and costs are outside the review?
• Can the recommendation be to keep the current approach?
• When does the delivery clock start?

If those answers fit your needs, ask Mike about your workflow. A general process description is enough for the inquiry. Leave out private client and employee information.