Inbox Triage | Real Minds AI
Professional Services /Customer-service agent live field guide · 9 min

Inbox Triage

Turns a chaotic shared inbox into a sorted, urgency-ranked queue with drafted replies — so the morning starts with the work, not the sorting.

theater/demos/profsvc_inbox-triage.html · sandbox · read-only
Open
FIG. 1

The live demo, running on fabricated data. Open it to step through the full flow — every output is shown for a person to approve before anything happens.

How it would work

Reads each inbound message to the firm's shared inbox, tags it by intent and urgency, re-ranks the queue, drafts a cited reply for routine traffic, and routes anything sensitive to a named solicitor to approve before anything sends.

Input 01
The shared inbox

Inbound mail arriving at the firm's reception or general inbox — sender, subject, received-time and body. New enquiries, existing-matter correspondence, billing and costs queries, scheduling.

Agent 02
Tags, ranks, drafts

Classifies each message by intent and urgency, suggests a route, writes a short rationale, and drafts a reply for routine items citing the firm's response library. Holds anything that reads as sensitive.

Output 03
A ranked queue, drafts shown

A queue sorted urgent-first, each routine item carrying a draft reply with its source shown, for a person to approve, edit, reassign or escalate. Sensitive items carry no draft and route to a named solicitor. Nothing sends on its own.

Where it works well

It does the daily first-pass triage of the shared inbox in full, every morning, and shows its working.

  • Best for the person who owns the firm's shared inbox — reception, an office manager, a sole practitioner triaging at 8am.
  • It earns its keep where inbound volume is high and repetitive — booking changes, fee and costs queries, document requests, general enquiries.
  • At dozens of messages a day, the recaptured hour goes back into the matter work and the client conversation, not the sorting.

The invisible cost in a law practice's front door is the sorting — someone reading every inbound message once just to decide which is a new enquiry, which is an existing matter, which is a costs query, and which can't wait. That triage produces nothing but an ordered list, and it gets repeated daily.

Where it works badly

It is confidently wrong on the message that doesn't say what it means — and a mis-tagged complaint or limitation deadline is the one that hurts.

  • Weak on long forwarded chains, sarcasm, and messages that bury the real request — it should flag low confidence, not guess.
  • Poor value where inbound is low-volume and high-stakes — a handful of bespoke matters a day, each needing a partner's full attention. There is little sorting to recapture.
  • It cannot tell a routine fee query from one that is really a costs dispute heading for the Legal Services Commissioner — that judgement stays with a person.
The honest test

If you can't describe most of your inbound mail as a short list of recurring types — new enquiry, existing matter, costs query, scheduling — then this makes your sorting faster, not your judgement safer.

A complaint phrased politely, an urgent matter buried in paragraph four of a forwarded chain, a client downplaying a looming limitation date — these are exactly where a wrong urgency tag does the most damage, and where the tool's confidence score is doing you the favour of admitting it isn't sure.

What it doesn't do — and shouldn't

It sorts and drafts. A person approves and sends. That boundary is deliberate.

WHAT IT DOES
Surfaces the intent, urgency and route it assigned, with a short rationale
Drafts replies for routine traffic, citing the response library used
Flags every message that reads as sensitive — a complaint, a distressed client, a request for legal advice
WHAT IT WON’T
Send any reply to a client on its own
Draft or give legal advice, or answer a request for advice
Decide what a complaint or a costs dispute deserves

A solicitor's duty of confidentiality (under the Australian Solicitors' Conduct Rules) and client legal privilege both attach to inbound correspondence, and a complaint mishandled at the inbox can become a Legal Services Commissioner matter. The accountable solicitor stays on the decision because the consequence — to the client and to the firm's standing — lands on them, not the tool.

What your data has to look like

One inbox the tool can read, routes that mean something, and a current response library to cite.

32%
Typical readiness
across orgs we see, before the first job
A single shared inbox, not five personal mailboxes
Usual weak point
Routes that are real destinations
Needs shaping
A current response library to cite
Needs shaping
A written definition of "urgent" and "sensitive" for the firm
Needs shaping
A new-matter / conflict-check step the tool routes into, not around
Usual weak point
The real first job

The real first job is rarely buying a tool — it is consolidating the inbox, writing down the response library that currently lives in people's heads, and agreeing what "urgent" and "sensitive" mean for the firm. That groundwork is bigger and more durable than the AI layer on top, and it is the part we help with. Once the inputs are clean, every morning's triage is faster and right by default.

Right fit if…
Your inbound mail arrives at one shared or reception inbox a tool can read
High, repetitive volume — booking changes, fee/costs queries, document requests, general enquiries
You can point to a current response library, fee information and document checklists
You can name the destinations a message routes to, and who owns each
Walk away if…
Almost every message is a unique, partner-level matter — little sorting to recapture
Inbound is scattered across solicitors' personal mailboxes nobody else sees
Your standard responses live only in people's heads, not written down
You want a tool that answers clients or gives advice without a person checking
Open questions

The worried-buyer questions, answered straight

No. The tool never sends. It drafts a suggested reply for routine traffic, shows you why it classified the message and which response-library source it used, and every draft sits behind an Approve, Edit, Reassign or Escalate step a person clicks. Anything it reads as sensitive — a complaint, a distressed client, a request for legal advice — is held with no draft at all and routed to a named solicitor. It surfaces and drafts; the person stands behind what goes out.
It tags the easy, repetitive traffic well — booking changes, fee and costs queries, document requests, general enquiries — and that is where most of the volume and wasted time sits. Long forwarded chains, sarcasm, and messages that bury the real request in paragraph four are exactly where its confidence drops, and it shows you that confidence so you know when to trust the tag and when to read it yourself.
No. It replaces the sorting, not the relationship. The person who used to spend the first hour triaging the firm’s inbox now starts on a ranked queue with drafts already prepared, and spends their judgement on the messages that need a human — the complaint, the upset client, the unusual enquiry. Capacity moves to higher-value work; the role stays.
The draft replies cite a source — your standard-response set, fee or costs information, a document checklist — so they are only as right as that source. If your published fees, costs-disclosure wording or document requirements are out of date, the drafts will confidently repeat the stale version. Keeping that reference content current is part of making this work, and usually the first thing we check.
Client correspondence is personal information under the Australian Privacy Principles, and for a law practice it also carries the solicitor’s duty of confidentiality and, often, client legal privilege. The tool reads from your own inbox within your control and classifies against your rules; we scope what is sent to any model, what is retained, and what stays in your tenancy as part of the build. The demo here runs entirely on fabricated mail — none of the senders shown are real clients.
No — and the design is built to prevent it. A new-enquiry message is triaged and routed to your new-matter intake, but the tool never opens a matter, gives advice, or commits the firm. Your conflict check (the duties to former and current clients under the Solicitors’ Conduct Rules) stays a human step the queue feeds into, not one it pre-empts.
What it takes to build
3–4 weeks · 4 phases
Reused from template~70%
Bespoke to this skin~30%
stack · Claude · shared inbox (Outlook/M365) · review UI
What it would cost

Fixed scope, fixed price, fixed dates.

01
Bite-sized first piece
One contained change, low risk
02
Pilot build
Most builds land here
03
Embedded support
Scale on proof

Drowning in the shared inbox?

The honest place to start is a bite-sized first piece — one contained slice of your inbound mail, low risk. Tell us where the sorting hurts; we'll play it back, scope it, and show you what's possible.

More in Professional Services
Contract Clause Review
View →
Project Knowledge Assistant
View →
Policy Precedent Finder
View →
Tender Compliance Matrix Builder
View →
How We Work Proof Talk to us
How We Work Proof Talk to us
Ask us anything