Choose your first automation workflow

Start with an observable task

A useful first automation has a clear beginning, a result someone can check and a person responsible for exceptions. Begin with a small handoff between existing tools, rather than redesigning a whole department. This is planning guidance, not a claim about client results or a promise of savings.

Write one sentence: when this event happens, prepare this result for this owner. Name the source system, destination, required fields and evidence of completion. A record created in a CRM is different from an email sent, a customer contacted or a sale completed.

Compare a few candidates

For each repeated task, answer:

• How often does it happen, including quiet periods and peaks?
• Are the rules understood, and are exceptions visible today?
• Can you use an authorized API or export with the permissions available?
• What happens if the same item arrives twice or a system is unavailable?
• Can a person inspect, correct and reverse the proposed result?
• Who will own the process after the pilot?

Prefer a task with stable inputs and a recoverable mistake. A draft internal task is usually easier to contain than an external commitment. If authority, data access or the current process is unclear, resolve that before automating it. Payments and consequential decisions need their own approval design.

Use rules before adding interpretation

Use deterministic checks for required fields, identifiers and routing rules. Add an AI interpretation step only if variable messages or documents require it, and compare its output against reviewed examples. Keep the action it can trigger narrow. Do not give a pilot permission to send messages or change records merely because it can prepare a draft.

Write the pilot brief

Record the current volume, elapsed time, active staff effort and corrections over a stated period. Separate time spent working from time waiting. Keep a representative test set covering ordinary items, missing information, duplicates and failures; use approved test data and restrict access.

Agree acceptance and stop criteria before the trial. Include correct completion, review workload, oldest unresolved item and recurring cost. Set the limits with the owner, rather than borrowing a universal accuracy or savings target. Name who may pause the workflow and how staff continue manually.

Illustrative starting point

A shared inbox receives routine inquiries. A first pilot could prepare an internal CRM task and route uncertain messages for review. It would not automatically promise an appointment, send a quote or overwrite an existing customer. Compare prepared tasks with the team's current work, then decide whether the next step is justified.

Before launch, demonstrate a successful run, a rejected input, a duplicate, an unavailable destination and a pause followed by controlled recovery. Keep the manual route available until the team accepts the evidence.

Read next and sources

Continue with email-to-CRM recovery and ongoing costs.

Reviewed 21 October 2026. Microsoft's Operational Excellence checklist provides related guidance on operating and improving a workload. This checklist is a proposed working method; it does not require a particular vendor or establish its suitability for your data.