Prepare reply templates for routine questions
The templates were written a year ago, and operators long since answer their own way. The agent rebuilds them from situations that actually repeat.
Medium · 30 min · once a quarter
What you get
Template "Order delayed" · built from 37 real cases
Hi {name}. Order {number} was held up at the warehouse.
New dispatch date: {date}. Sorry for the wait.
Use when: the delay is no more than {N} days
Do not use: the customer is already asking for a refund
→ needs a person, the template promises no compensationA sample on made-up data — your numbers will be your own.
Who it fits
- There are new operators on the team, and the same situation sounds different from each of them.
- Templates exist, but they are a year old and have drifted from what operators actually write.
When it won't work
Requests are almost always non-standard — the template gets rewritten in full and only slows people down.
How the agent does it
Give threads and limits
Pick a period when the current team was in place and attach replies you like. List separately what an operator must never promise.
Run the build
The agent groups situations by frequency and writes templates only for the recurring ones. Variable parts stay as empty fields, not borrowed data.
Proofread before rollout
A template spreads across the whole team and goes out unchecked. A mistake in it is multiplied by the number of operators.
What you set
Connect Telegram and email. By hand: the period of correspondence to use, your tone rules, and the limits of what an operator may promise alone.
What you'll need
Starter prompt
Copy the prompt or open it straight in a chat with the agent.
Build reply templates from the Telegram and email correspondence for [period]. Identify the situations that occurred at least [N] times, and write a template for each in the tone from [the attached examples]. Everything that changes case by case — name, order number, deadline, amount — stays as a field [in square brackets]; do not fill in a typical value, or the operator will send someone else's deadline. Do not write deadlines, refunds, discounts and goodwill compensation into the text: put a field [decision per the rules] and note who is entitled to fill it in. For each template: when to use it, when not to use it, what to fill in. Output a table of templates in descending order of situation frequency.
Open in chatCheck the set of templates. List: — templates built on fewer than [N] cases; — places where a value from a specific conversation is filled in instead of a field; — texts promising a deadline, refund or goodwill compensation without a note on who approves it; — situations you merged into one template, and how they actually differ; — frequent situations from the correspondence left without a template; — which templates are ready for use and which the support lead must proofread.
A second prompt — the agent uses it to review its own work and show what's left for you.