$28k a year saved by answering the same question once
Past tickets + docs → searchable answers, draft ready → agent approves and sends
This support team spent its day answering questions it had already answered before. The answers existed, but they were buried in old tickets and docs, so each one took about eight minutes of searching and rewriting. We organized that history into something searchable, then added a layer that drafts a reply the moment a question arrives. An agent still approves every send. Eight minutes became under one.
What it's worth
Agent time back
- Questions per day
- ~40
- Time per ticket, before
- 8 min
- Time per ticket, after
- <1 min
- Agent hours a day, before
- 5.3 hrs
- Agent hours a day, after
- 0.7 hrs
- Hours back per year (240 days)
- ~1,120 hrs
- Support agent cost
- $25 / hr
- per year
- $28,000
How this was calculated — every assumption
- About 40 questions a day at roughly 8 minutes each before, dropping to under a minute after. The remaining minute is the agent reading and approving a draft that is already written.
- Support agent time is costed at $25/hour, at the low end of a loaded agent cost.
- 240 working days a year, allowing for holiday.
- Only recovered hours are counted. Faster replies probably help retention on a CRM product, but that is not measured and is not claimed.
Results at a glance
- ~$28,000 a year in agent time back
- 1,120 hours a year returned
- Each question answered in under a minute, down from about eight
- ~40 questions a day handled with on-brand answers
- A human approves every reply before it sends
The problem
About 40 questions came in a day, most of them variations on things the team had already answered. But those answers sat in buried threads and individual memory rather than anywhere organized. So agents spent roughly eight minutes a ticket searching for and rewriting replies the company had already written once.
What was built
Organized answers
Past tickets and documentation pulled into one structured place, with a dashboard that keeps it current as the product changes.
Draft on arrival
A new question triggers a search of the relevant history and a first draft, based on the company's own answers rather than a generic guess.
Human approval
Agents review, edit if needed, and approve before sending. Nothing unchecked reaches a customer. That is what the remaining minute buys.
What keeps it from breaking
Answers come from your own history
Drafts are built from real tickets and docs, not the model's general knowledge. A confident wrong answer about your own product is worse than a slow one.
A person approves every send
The system drafts, an agent approves. Full automation would have saved another 40 minutes a day and been the wrong trade.
Someone owns the answers
A dashboard keeps the knowledge current as the product changes. Unmaintained answers turn confidently out of date.
How the approach worked
The win was not a smarter model. It was putting the company's own answers where the model could reach them, and keeping a person in the loop so quality held. Get the knowledge out of where it is stuck and put it where someone can use it.
The results
Replies went from about eight minutes a ticket to under one. Agents clear the day's volume quickly because the draft is already there and based on the company's own history. That is roughly 1,120 hours a year, about $28,000 of support capacity. A year of scattered knowledge became something the team can reuse.
Tools & technologies
Have a process like this slowing your business down? Tell me your stack and goals and I'll send back a build plan.
Take the stack assessment →