Plan your first small-business automation

← All guides and worksheets

For an ordinary working week, note the tasks that require someone to copy information, chase a missing item, prepare the same report, or check the same conditions. Capture a concrete example and the person doing the work. Avoid estimating the whole list from memory at the end of the week.

Separate task volume from frustration. An annoying task that happens once a year may be a weaker first project than a small handoff repeated every day. Also record whether the underlying process is stable: if nobody agrees on the next step, define that step before automating it.

• Task name, trigger, frequency, and the person currently responsible.
• Tools and records used, including where information is copied.
• Time spent on an ordinary case and on a difficult case.
• Typical mistakes, missing facts, and the manual recovery process.
• What a correct finished result looks like to the person receiving it.

Compare candidates by observed repetition, clarity of rules, availability of inputs, and ease of checking the result. Prefer a first project with a named owner and a reversible output, such as a draft, an internal task, or a record prepared for review.

Keep uncertain judgments visible. Extracting an invoice number from a document and deciding whether to pay a disputed invoice are different jobs. The first may have a checkable answer; the second needs the right person's authority and context. A useful automation can prepare the evidence without taking over the decision.

Record the observation period, number of cases, active handling time, elapsed wait time, and corrections. Active work and waiting answer different questions: reducing five minutes of copying does not prove that a three-day approval delay disappeared.

Use a simple capacity estimate: monthly cases multiplied by active minutes removed, minus the review and exception minutes added. Keep the inputs visible. For a purely hypothetical example, 40 cases with six minutes removed and two minutes of review added would release 160 minutes a month. That is an illustration of the calculation, not a result achieved by Agent Run Apps.

Decide what the team would actually do with that time. If the main value is consistency, fewer missing fields, or a clearer customer handoff, measure that directly instead of forcing every benefit into a revenue estimate.

Define one trigger, one finished output, the tools involved, and what remains with a person. Check built-in settings before proposing a connection or custom build. Make the first phase useful enough to keep even if the business decides against further work.

Include the operating details in the brief: who receives a failure alert, how to pause the workflow, how a retry avoids duplicates, and where records live. Decide the success criteria before the test starts so a polished demonstration does not replace the result you needed.

• Trigger, input sources, required output, and explicit scope exclusions.
• Human review points and actions that require approval.
• Accounts and access needed, with a named business owner.
• Ordinary, incomplete, duplicate, and failed-connection test cases.
• Completion checks, cost boundary, manual fallback, and handoff instructions.
• Pilot owner, review date, and evidence required to expand, revise, or stop.

Start with fictional or appropriately prepared examples, then a controlled set of real work once the necessary access and action boundaries are agreed. Compare the same measures with the baseline. Note changes in the case mix, interruptions, and any extra checking the pilot required.

Count runs requiring correction and work that never completed, not just successful runs. Ask the person doing the job whether the new process removed effort or moved it elsewhere. Keep the manual route available while issues are resolved.

At the review date, choose one action: keep the current scope, repair a specific failure, expand one proven step, or stop. A pilot that reveals the existing process is already sufficient can still prevent an unnecessary larger project.

A working first phase should have a clean stopping point: useful output, visible records, a named owner, and instructions the team can use without the builder present.

Download the plain-text template and paste it into a shared document, or copy the version below. Fill it in with your team. No email address or account is required.

Download: First automation planning worksheet (.txt) ↓

FIRST AUTOMATION PLANNING WORKSHEET

Business / workflow owner:
Planning date / review date:

1. TASK SHORTLIST (COPY FOR EACH)
Task and trigger:
Recent example:
Observed frequency:
Person responsible:
Tools / input sources:
Ordinary handling time / difficult-case handling time:
Missing facts and common mistakes:
Correct finished result:

2. FIRST-TASK DECISION
Selected task:
Why this repeats enough to investigate:
Rules the team agrees on:
Inputs we can access:
How the result can be checked:
How to pause or recover:
Current tool settings already checked:

3. BASELINE
Observation period:
Cases observed:
Active handling minutes:
Elapsed wait time:
Corrections / incomplete cases:
Data limitations or unusual cases:

4. PILOT BRIEF
When [trigger]:
Use [inputs]:
Produce [output]:
Done means [completion checks]:
Outside this phase:
Human approval points:
Account / access owner:
Cost boundary and recurring costs:
Failure alert recipient:
Pause / manual fallback:
Duplicate and retry behavior:
Handoff instructions required:

5. ACCEPTANCE CASES
For each: input / expected output / observed output / pass or fix.
[ ] Ordinary case
[ ] Missing or conflicting information
[ ] Duplicate input
[ ] Connection failure and retry
[ ] Stop condition or human override

6. REVIEW
Cases completed / cases attempted:
Handling, review, and exception minutes:
Corrections / incomplete cases:
Difference from baseline and possible other causes:
What the team would do with reclaimed time:
Decision: keep / repair / expand one step / stop
Reason and next action:
Owner / next review date:

Bring one recurring task and a rough version of this worksheet. Adam can help choose a useful first step and define how you would know it worked.

A one-on-one AI setup session is $295; team working sessions start at $950. Full custom automations start at $2,500. Scope, fixed phase pricing, and completion checks are agreed before custom work starts; software subscriptions and usage charges are separate.

Talk through a first automation ↗

See scope and pricing · See setup sessions

Related capabilities

• How workflow automation gets built
• Start with an AI setup session

Keep planning

• The inquiry-to-estimate handoff checklist
• Use your existing tools or build custom automation?