Request arrives
Every incoming email lands in the team's existing Outlook mailbox and is picked up automatically through the Microsoft Graph API, in French or English, with its history attached.
Automated
All case studies
Case study · Support automation
Safari Condo A support desk that answers before anyone types
Safari Condo, a Québec maker of motorhomes and travel trailers, was receiving more support requests than its team could read. We automated the desk so every request is categorized and prioritized on arrival with an email already drafted from the company's own documentation, while a person still decides what gets sent.
Engagement
From scratch
Where it runs
Their Outlook inbox
Auto-send
Never
Languages
EN / FR
Urgency levels
4
Priority, urgent, moderate and low, assigned on arrival.
Request types
5 inboxes
Technical support, appointments, dealers, parts and federal recalls.
Every draft
Cites its sources
The agent can open each document the answer came from.
Auto-send
Never
Nothing reaches a customer without someone sending it.
The brief
The bottleneck was the inbox
Safari Condo builds motorhomes and travel trailers in Québec. Its customers use those vehicles on the road, and they write to support when they need an answer about a system in the vehicle, a part, a service appointment or a warranty question.
The answers almost always existed already: in the internal service documentation the team maintains, or in the public documentation owners have access to. What the team lacked was the time to find them, request by request, while new requests kept arriving.
- Request triage
- Priority ranking
- Documentation retrieval
- Pre-composed replies
- Bilingual EN/FR
- Outlook via Microsoft Graph API
What was at risk
Support automation usually fails by answering confidently and wrongly
A vehicle maker cannot afford a plausible-sounding but wrong answer about a propane system or a warranty term, so we designed against a system that invents answers before we designed for speed.
The usual support bot
How this one runs
Talks straight to the customer, unsupervised.
Drafts for the agent, who replies to the customer.
Answers from whatever the model happens to know.
Answers only from the company's own documentation, quoted back.
Treats every message as equally urgent, first in and first out.
Ranks by category and urgency, so the road-side case surfaces first.
Sends the underlying documents straight to the customer.
Sources are named for the agent only. The customer sees the reply, never the documentation behind it.
Silently drifts as products, parts and policies change.
Re-indexes the documentation, and every draft names the source it used.
What we built
What happens before an agent reads the request
The team kept the desk it already used. We added a layer underneath it that reads, sorts and drafts each request, so the first human action is a judgement rather than a search.
Categorized
It lands in the inbox it belongs to: technical support, appointments, dealers, parts or federal recalls.
AutomatedPrioritized
Urgency is scored, so someone stuck on the road doesn't sit behind a brochure request.
AutomatedReply pre-composed
The relevant passages are pulled from the company's documentation, and a reply is drafted with the sources named for the agent to check.
AutomatedA person sends it
The agent reads, edits if needed, and sends. Nothing goes to a customer on the system's own authority.
HumanWhere the answers come from
One document set, cited for the agent, never seen by the customer
Service procedures, parts and warranty detail, manuals and published FAQs are ingested as a single corpus. That corpus shapes every draft, but a customer receives only the agent's reply, never the documents behind it.
Ingested
The company's documentation
- Service procedures and known cases
- Parts, configuration and warranty detail
- Manuals, guides and published FAQs
Drafted
A grounded reply
- Passages named so the agent can verify them
- Editable before anything is sent
- Never shown to the customer directly
The guardrails
What the system is not allowed to do
We closed off the failure modes before automating anything, so these constraints were written into the spec from the start.
No answer without a source
Every draft is grounded in the company's documentation and names what it used. If the documentation doesn't cover it, the draft says so and routes it to a person instead of inventing an answer.
No automatic sending
The system prepares each reply and the team decides what to do with it. Every customer-facing message is read and sent by a person, which makes the added speed safe to use.
Sources stay with the agent
The documentation shapes the draft, but it is never sent to a customer. Citations exist so the agent can verify an answer before sending it.
The path
Built against the real inbox
We started from the requests the team had received and measured every step against what a good agent would have replied, so the team's intent drove each technical choice.
-
Read the desk as it is
Before anything was built- Real request history reviewed
- The categories the team works in
- What makes a request urgent, in their words
- Which answers already existed in writing
-
Make the documentation retrievable
Foundation- The company's documentation ingested as one corpus
- Citations attached to every passage, visible only to the agent
- Bilingual content handled as one corpus
- Re-indexing when documentation changes
-
Triage before drafting
First working slice- Categorization shipped and checked against real mail
- Priority scoring tuned with the team
- Misroutes reviewed weekly, rules adjusted
- Queue order visible to the agents
-
Drafts the team would send
Iteration- Drafts compared against replies agents had written
- Tone and structure set by the team, not the model
- Source citations required in every draft
- “I don't know” treated as a valid output
-
Handed over, still watched
In operation- Edits agents make tracked as the quality signal
- Categories and priorities revisited as volume shifts
- Documentation gaps reported back to the business
- Owned as a running system, not a delivered file
What the desk can now see
An operation the business can measure
A shared mailbox produces no numbers. Because every request now flows through a database, the team can look at the backlog by status and urgency, and at how long replies and resolutions are taking.
The result
The same team and inbox, with the groundwork done
Each request now reaches the desk already sorted, ranked and drafted, and every answer a customer receives is still chosen by a person
For the customers
Faster first replies, and answers that match the documentation for their vehicle instead of a general guess.
For the team
Agents no longer search 2 sets of documentation for each request, so their time goes to judging and editing the reply.
For the business
Volume absorbed without adding headcount, and a live map of what owners keep asking, which is a product signal as much as a support one.
Your turn
If your inbox is the bottleneck, this is the shape of the fix
Safari Condo is one case, and the method is how we run every AI build: start from the real work, ground every answer, and keep a human on the send button.
Automate the work before the reply
Reading, sorting, ranking and retrieving are where the hours go. Automating those steps is safe to measure and check, and it is usually enough to clear the backlog.
Your documentation is the model's source
Answers come from what your business has written down, cited so your team can check them before anything is sent, which makes every output defensible.
Coherence, then speed
We check that the system does what you meant and keeps doing it as your products, parts and policies change.
Automate the sorting and drafting on your support desk
Tell us what lands in your inbox every day. We'll show you which part of it can be automated safely, and which part should stay human.