Arcas Case Study: an AI agent that handles 100,000 customer emails a year
Arcas built an AI agent that researches customer requests in the client’s own systems and prepares the reply, after first building the knowledge base the company never had.
Overview
An automotive service group runs several service centres and receives more than 100,000 customer emails a year, around 400 on a working day, in four languages, handled by six employees. Each email carries a research job. Identify the vehicle, verify its registration, find the customer record, work out what is missing, then answer under the company’s own service conditions. No inbox handles that.
The challenge
A windscreen replacement looks like one email but costs four. The advisor has to establish which vehicle the request concerns, check its registration against the customer record, and work out which documents and service conditions apply. Customers rarely send everything at once: the registration arrives today, the next detail tomorrow. The deeper problem was structural. The knowledge was there, in what the advisors knew and in the documents they worked from, but not in a form any system could reach. An agent without the client’s systems and documents put into an organised structure would produce answers the company could not stand behind. Personal data had to stay in the EU, and a reply attached to the wrong customer or vehicle is worse than no reply.
A knowledge base that did not exist before
Arcas built the missing foundation first. The client’s documentation (service descriptions, conditions, required documents, and years of answers the advisors had given) was brought together into a structured, scalable knowledge base, with the requirements of every service type written as explicit decision trees. The agent answers from the company’s own knowledge rather than general knowledge, reads registration documents sent as photos or PDFs, and flags one that points at a different vehicle. What the company did not have before is now an asset it owns: an information platform ready for whatever it automates next.
What the agent does with each customer request.
Hover or tap a box to see what it does.
Built inside the system the client already runs
The client works in its own service software and mail system rather than in a Microsoft environment. Arcas built the agent directly into what the company already had, rather than asking it to change anything: a panel inside the interface the advisors already use, bringing together customer and vehicle details, missing documents, the rules that apply to the request and a suggested reply. The agent answers a request directly, prepares a draft for an advisor, or passes the conversation on. No parallel application, no migration, no side tool.
Improved by the people who use it
The first working version reached the advisors very quickly, and delivery has stayed iterative since. They rate suggested replies, edit them and comment on them in the panel; each correction becomes a backlog item, is written back into the knowledge base and the decision rules, and ships in short cycles, all within the same week. The effect compounds: a knowledge base that grows more valuable the longer it is used, instead of drifting out of date.
The results
The agent assembles the information behind a request, checks documents against the customer record and produces a reply an advisor can use as it is. Because each service type’s requirements are known in advance, follow-up questions that used to take several emails come back in one response. Advisors spend that time with customers instead, and requests handed to a colleague arrive with everything available already gathered. What the agent handles on its own is the client’s decision, and it widens as the agent proves itself.
| Metric | Before | After |
|---|---|---|
| Emails handled without advisor involvement | 0 | 35% |
| Follow-up emails per request | 4 | 1-2 |
| Advisor search time per request | 7 minutes | 2 minutes |
| Advisor capacity released | – | 40h a week for 6 FTE |
Could this work for your business?
If your team spends hours answering similar emails, searching for information or chasing missing details, there is a way to handle that workload, and it usually begins with knowledge already available but never put to work. Arcas builds AI agents around the systems a business already runs, in any environment. The foundation built alongside the agent keeps paying long after it goes live.
Get in touch: contact@arcas-agentic.com

