Case study · Customer service
A small support team now has a full-time AI agent working the queue every day. A person approves every reply.
An AI agent works the help desk queue every working day, preparing replies and keeping the help centre current. A named person approves everything.
- Who
- A software company with a three-person support team, one help desk and a help centre.
- The problem
- Tickets needed attention every working day, customers kept asking what the help centre missed, and nobody owned the next change to the chat assistant or the policies.
- What we did
- A full-time AI agent took on the support role for one help desk: reply drafts, tags, handovers, help centre upkeep and a weekly written summary.
- The result
- The team approves prepared work instead of starting from nothing, and the help centre and the assistant's tests are kept current.
- Time
- Ongoing. Set up the day after the request.
What was going on.
The support team of three spent their days in the help desk queue. Facts changed: a price, a policy, a product setting. Unanswered questions piled up. The chat assistant had been set up and then left alone, and after an update from its vendor its answers shifted and nobody noticed for weeks.
Nobody owned the work in between: updating the help centre when a fact changed, checking the assistant, writing up bugs with the steps to reproduce them. It fell to whoever had a spare hour, which was nobody.
What we found.
The company did not need another tool. It needed a role: someone to work the queue and the upkeep every working day. So we set up a full-time AI agent for that role, for one help desk and its channels. It takes the next item as soon as it closes the last, and the team can talk to it in chat as they would a colleague.
What it produces is prepared work: reply drafts from the company's approved articles, tags and routing suggestions, handovers with the whole conversation attached, refund and cancellation requests, bug reports with steps to reproduce, and fixes to the help centre. A separate automated check reviews each piece before it reaches the team, and a short weekly summary shows what was done, what is open and what needs a decision.
The technical detail
The agent works from read-only exports and a test area, not from the live help desk. It is set up with the company's models, safety limits and its own notes about the business, and it keeps saved test sets for the chat assistant that are re-run after any change to content, policy, settings or a vendor update, including at least ten attempts to trick it. Replies, policies, prices, refunds, credits and cancellations all need the named owner's approval, and the team sends or publishes everything.
What we did.
How we checked it.
These had to pass before we asked for sign-off.
- Each piece of work meets its agreed checks and is accepted, or returned for revision, by the named owner.
- The report on the chat assistant states the sample size, the errors by type, the saved test results after each change and the results for at least ten attempts to trick it.
- A seeded change to an agreed fact finds every help centre article that contains it.
- The agent has made no live change, reply or post, and has decided no refund, credit or cancellation.
What they have now.
The team received a steady flow of prepared work: drafts with their source articles, tags, handovers, bug reports, help centre fixes, and the assistant's test results with undo steps for any proposed change. Each week the support lead read a short summary of what was done, what was open and what needed a decision.
The team spends its time approving, correcting and talking to customers. The help centre and the assistant's test sets are now kept current as facts change.
What we did not do.
- The agent changes nothing live. The named owner approves and applies every change to the assistant, the help desk and the help centre.
- It sends or posts nothing to customers or the public. The team sends.
- It decides no refund, credit, cancellation, discount, policy or price. It prepares the request for the owner.
- There is no out-of-hours cover and no guaranteed response time. Working hours and the route to a person are agreed in writing.
- We promised no result for resolution time, customer satisfaction or cost.
The service behind this.
If support work needs doing every day and nobody owns the upkeep, an AI agent can hold the role while your people keep every decision.
Full-time AI agent
AI customer support agent
Full-time AI customer support agent for one help desk and its channels: queue, reply drafts, bot checks and help centre upkeep.
$3,500/month. Payment, billing, cancellation and renewal are agreed in the written proposal before the engagement starts. A message creates no charge, reservation or automatic subscription.
More case studies.
One-off job
Checkout stopped taking orders after an update
Customers could fill a basket but not pay. We found the cause on a copy of the shop, fixed it and proved it with test orders.
Online store · Two days
One-off job
Genuine payment messages were being rejected
Paid customers still showed as unpaid. The company's code was checking the wrong thing. We fixed it without switching the check off.
Software team · Three working days
Project
Hundreds of products flagged in Google's shopping feed
Several problems at once, and every earlier fix had only revealed the next one. We grouped them by cause and worked through them in order.
Online store · Three weeks, mostly waiting for Google to reprocess
Get a price for something like this
You are asking about: AI customer support agent.
Sending this is not an order. We check we can help, then agree what is included, safe access and the price in writing before any work. Prefer email? hello@syntheticindustry.ai.