Choose a request your team can finish
“Automate our support” is too broad to test. “Help a verified member find the replay included in their plan” is a workflow with a recognisable beginning and end.
The best first workflow is one your team already understands, with a clear source of truth, a useful outcome and a safe way to handle exceptions. It might be a direct answer, a verified lookup or a narrowly permitted action. It does not have to be the most impressive demonstration.
Start with a manageable sample of recent tickets. Record what the member wanted, what the agent checked, what finally resolved the request and whether the conversation reopened. Use your own queue to establish repetition; a common FAQ elsewhere does not prove that it consumes your team’s time.
Choose a workflow your members recognize
Education communities may begin with replay discovery, session guidance, or eligible access restoration. A startup ecosystem may start with program onboarding and mentor-request intake. For a neobank or fintech service, approved feature guidance and secure onboarding orientation are different from making a financial or verification decision.
Look for these five starting points
- Finding a resource: locate the correct guide, replay or event page and explain who can use it.
- Understanding a rule: answer from the current community policy and show the relevant source.
- Getting started: guide a member through the first steps appropriate to their goal.
- Checking access: identify whether a verified member’s entitlement matches their account or role.
- Tracking an existing request: report the status available in the authoritative system without inventing a completion date.
The first three can often start with approved public or member-facing knowledge. The last two require a working connection to the right account data and appropriate permissions. Treat that dependency as part of the workflow, not something to fill in with a confident-sounding answer.
Score each candidate before building
For each request, ask four questions. How often did it occur in your sample? How much active work did it take? Can the necessary evidence be accessed reliably? What happens if the answer or action is wrong?
A frequent, predictable question with a maintained source may be a better first choice than a complicated request with a larger apparent time saving. A workflow that changes access, money or account ownership needs stronger identity checks and explicit action rules.
Write a one-sentence scope: “For verified members with an active standard membership, check the expected Discord role and use the approved sync process when it is missing.” Then list the exceptions: unclear identity, conflicting records, a different tier, missing permissions or an unsupported billing state.
Define done before you define the automation
Illustrative request: “Where is yesterday’s workshop replay?”
Done: the member receives the correct replay link, any access requirement is explained and the link works for their account.
Not enough: the system sends a general link to the resource library and the member opens another ticket.
Use that distinction to write a small set of realistic test cases. Include the normal request, an ambiguous version, an outdated source, a member without the required access and a request outside the workflow’s scope.
When the system cannot finish, it should transfer what it already knows: the member’s request, verified facts, checks performed and the reason it stopped. A handoff should shorten the next person’s investigation.
Run a limited pilot and measure real work
Choose one workflow and a defined group of tickets. Keep a person responsible for reviewing failures and correcting the procedure. Expand after the workflow handles both the normal case and its exceptions consistently.
- Completed outcomes: how many eligible requests actually reached the intended result?
- Human effort: how much active staff time was needed, including review and follow-up?
- Reopens: how often did a supposedly resolved request return?
- Wrong answers or actions: what failed, and did the control catch it?
- Handoff quality: could the next agent continue without restarting?
Do not count every automated message as saved work. Review the result from the member’s perspective. The first successful workflow gives you a repeatable pattern for choosing the next one.