ops that run.All field guides ↗
← Back to the field guides

Getting started / Workflow design

Choose your first AI workflow by the decision it improves

A practical way to choose one recurring task, test it on past work and decide whether automation earns its upkeep.

A consultancy owner can spend Friday collecting updates and still finish without a clear answer to Monday's priorities. Automating the collection helps only if the resulting report changes a decision. Start by naming that decision: which client needs attention, which invoice needs checking, or which delivery commitment is at risk.

Choose a task you can inspect

Write down three tasks that recur in your business. For each, record who does it, the input, the output and the person who acts on the result. Choose one with an identifiable source, a consistent format and a reviewer who can tell whether the answer is correct. A weekly project exceptions list is easier to evaluate than an assistant asked to run the business.

Check what you already have. Anthropic's September 2026 release describes 43 packaged small business workflows. That makes reviewing existing options a sensible first step before commissioning a custom integration. Availability and fit still need checking in your account. Read Anthropic's release.

Run a small comparison

  1. Keep a baseline. Save one completed report and record the preparation, checking and correction effort separately.
  2. Write the output rule. Specify the fields, reporting period, source links and treatment of missing information. Require unresolved items to stay visible.
  3. Use past inputs. Recreate the report from an approved historical sample. Compare it with the source records, including an awkward exception.
  4. Keep the action with a person. Let the trial propose priorities. Do not let it email clients or change financial records.
  5. Decide after observing. Count corrections and missed issues alongside preparation effort. Include the work of maintaining access and fixing failures.

A fictional example

Cedar Studio manages eight client projects. Its owner initially asks for an AI dashboard covering everything. The first trial instead lists milestones due next week, unresolved blockers and the person responsible. One project has no update. The report marks its status unknown rather than declaring it healthy. That missing update becomes the owner's next action.

When to pause

Pause if nobody owns the source data, if the process changes every time, or if reviewers cannot distinguish a correct answer from a plausible one. Fix the handoff or reporting definition first. A modest workflow with clear ownership is more useful than a broad system nobody trusts.

Start here: Bring one recurring report and the decision it is meant to support. We can discuss whether a native feature, a simpler process or a small automation is the appropriate next step.

These are practical workflow guides, not claims of a tested client implementation. Examples are fictional. Features, plans and availability can change; check the linked documentation before setup.

Choose a useful starting point.

Bring one recurring task. We can discuss its inputs, owner and the decision it should improve.

Explore all seven guides →Try the company dashboard ↗