Mindcraft Book audit
Mindcraft / Capabilities / Inbox Triage
Module 03 · Capability

Inbox triage,
on its own.

Customer emails, vendor invoices, sales replies — read, classified, replied to, escalated, or filed. The day starts at 47 unread, not 470. Your team works on judgment calls, not the same five FAQs eighty times a day.

What this fixes

Three reasons
shared inboxes burn people out.

If your team has ever opened info@ at 9am and felt their soul leave their body, you're who we built this for.

Problem 01

Shared inboxes are quicksand.

info@, support@, ops@ — five hundred messages a day, no ownership, FIFO at best, things slipping at worst. Nobody actually likes opening it. So nobody does, until they have to.

Problem 02

Every email gets the same urgency.

An angry customer about a billing error sits in the same queue as a vendor's monthly newsletter. By the time you triage, the angry customer is on Trustpilot.

Problem 03

Replies are 80% the same.

"How do I reset my password." "When is my order shipping." "Do you offer a discount." Your team types the same answer 40 times a day, every day, forever.

Sample transmission

The last eight emails,
handled.

An illustrative triage view, modeled on a typical inbound flow at a small services business. Eight emails arrived in the last 90 seconds. Seven were handled without a human. One was flagged for the manager. Click any row to see the AI's decision.

📥 support@bright-studio.com
347today
96.4%accuracy
1.4savg classify
Live · processing
What ships

What's in the box.

Six things you get with every inbox triage engagement. Real classification, real auto-replies, real escalation rules.

01

Classification you can trust.

Every email tagged on arrival: customer support, sales reply, vendor, billing, complaint, partnership, internal, spam. Confidence scores attached so you know when the AI isn't sure.

02

Auto-replies for the obvious.

FAQs answered instantly with your tone. Order status looked up and replied with real data. Password resets handled. The repetitive 70% goes away — without losing the human touch where it matters.

03

Escalation with full context.

When a human is needed, the email lands in the right human's lap with classification, priority, recommended response, and a one-line summary. No re-reading from scratch.

04

Sentiment + urgency detection.

Angry customer? Detected and bumped to top of queue. VIP account? Routed straight to a senior rep. Threats of cancellation? Manager pinged. Calm question? Standard SLA.

05

Daily and weekly digests.

Summary email each morning: what came in, what got handled, what needs you. Weekly trend report: rising topics, new complaint patterns, suggestions for new auto-replies.

06

Continuous improvement.

Every reply a human writes becomes training data. The agent gets better at your tone, your common cases, your edge cases — automatically, without you teaching it.

How it works

Four steps,
end to end.

Step 1
!

Email arrives.

Gmail, Outlook, Help Scout, Zendesk, Intercom — or via raw IMAP. Whatever you use, we plug in.

Step 2

Read + classified.

Topic, sentiment, urgency, intent — all detected in under 2 seconds. Confidence score attached.

Step 3

Reply or route.

Auto-reply if the answer is obvious; draft for human if it isn't; escalate if it's hot.

Step 4

Logged + learned.

Pattern stored. Agent refines. Next time a similar email arrives, the answer is sharper.

Mission control

Today's inbox,
from above.

A live dashboard your manager pulls up first thing. Volume, classification breakdown, what's pending, what got handled. The state of the inbox in one screen.

MINDCRAFT · INBOX TRIAGE · SUPPORT@ Live · auto-refresh
Inbox today
347
↑ 22 vs yesterday
Auto-handled
282
81% of total
Need attention
11
3 escalations
Avg classify time
1.4s
96.4% accuracy
By category · last 24h
Support
142
Sales
38
Vendor
61
Billing
42
Complaints
11
Spam
53
Built on

A stack you can audit, port, and own.

Everything stays in your tenant. We don't pull email content out — auth is OAuth, scoped to read+send. Source code in your repo.

Gmail / Outlook Help Scout / Zendesk Front / Intercom Claude (classification + reply) n8n / custom backend Slack alerts
8–15 hrs/wk

Per employee, per week — the range workflow automation reliably reclaims across admin, lead routing, scheduling, reporting, and inbox triage. Five well-chosen inbox workflows alone can recover the equivalent of a full FTE and stop the Monday-morning dread.

Source: McKinsey-cited research; multiple industry analyses, 2025–26
Frequently asked

Questions we get a lot.

How do you stop the AI from replying to something it shouldn't?
Hard rules first. Never auto-reply to legal, compliance, or refund-over-threshold emails. Never auto-reply if confidence is below threshold. Soft rules second. Tone-check before send. Holdout sample reviewed by a human weekly. Last line: a draft mode for the first 30 days where the AI suggests replies but a human approves each one.
Does this work with our help-desk tool?
Yes — Help Scout, Zendesk, Front, Intercom all integrate cleanly. We can also work with bare Gmail/Outlook if that's what you have. The classification layer sits before whatever you use; replies and routing go through your existing tools.
What about email security?
Everything stays in your tenant. We don't pull email content out. Auth is OAuth, scoped to read+send. We're SOC2-friendly architecture and can map to your DPA. Optional VPC deployment for regulated industries.
Can it handle non-English?
Yes. Claude handles 100+ languages well. Our team has worked across English, Spanish, Russian, Georgian, French, and Portuguese projects. We pin language on classification, so replies always go out in the customer's language.
What's the typical accuracy?
After 30 days of training on your data: 96–99% classification accuracy, 92–95% reply quality (human-rated, on a sample). We continuously measure both and ship improvements monthly.
Will my customers know it's an AI?
If your transparency standards require disclosure, we add a footer (e.g., "This reply was generated automatically"). Without disclosure, the replies are tuned to match your team's existing tone — most customers don't notice. Either way, the option to escalate is always one reply away.
What's it cost?
Inbox triage is included in the Engine tier ($3,500/mo) for one mailbox. Additional mailboxes are $300/mo each. Plus pass-through Claude API costs, typically $80–400/mo depending on volume.
Often paired with

Related capabilities.

See what's
in your inbox.

The free audit includes a 7-day shadow-mode trial: we sit silently next to your inbox, classify everything that comes in, and show you the report. No replies sent yet, no commitment — just visibility.

Book the audit