No major ESP processes inbound. Replies are pure signal — purchase intent, churn risk, support escalations, partnership pings — and the entire industry throws them away. Email Digit listens to every reply, classifies it, and triggers the right thing.
Once you've connected your inbox, you don't see Reply Intelligence until something interesting happens. Then it's already triaged.
Connect Gmail, Microsoft 365, or any IMAP server. We ingest the raw MIME, preserve threading, keep attachments, and store the canonical message ID — so when a reply lands two weeks later, we still know which campaign it belongs to.
Each reply runs through our intent + sentiment + extraction model. Intent is one of 9 curated classes today (not generic positive / negative). Sentiment is a scalar. Extracted fields update the contact record automatically — company, team size, current vendor, geo.
Each classification can branch your automation — pause a drip when a reply lands, fast-track to sales on purchase intent, fork into retention flow on cancellation signals, hold a send when a recipient is out of office. Configure once per flow, then forget.
Every Friday morning: a digest of the week's replies, grouped by intent. 47 replies — 12 high-intent (listed). 3 cancellations (with the reason extracted). 2 partnership inquiries. Your inbox stops being a tax and starts being a feed.
Real examples from our beta cohort — anonymized. Each one shows the inbound message, what Email Digit classified, and the automation it triggered.
“Quick question — what does Pro tier look like for a team of 12? Moving off Klaviyo this quarter.”
“We're not seeing the open-rate uplift we expected. Considering pausing for a few months and revisiting in Q3.”
“Love what you're building. We're a Shopify app with overlap — want to chat about a co-marketing motion?”
“DKIM is failing on mail.acme.com. Already tried rotating once — same result. Need help today.”
“I'm out until June 12. For urgent matters, please contact priya@…”
“Can I send WhatsApp messages from your platform to numbers in Bangladesh, Pakistan, and UAE?”
Every classified reply emits a typed event with the same shape — whether the inbound came from Gmail, IMAP, or your own SMTP. Subscribe via webhook, query via API, or branch automations directly.
UNCLASSIFIED{
"id": "evt_8f3k2p4xN",
"type": "reply.received",
"created_at": "2026-05-28T14:32:53Z",
"data": {
"reply_id": "rep_amelia_9k2",
"from_email": "amelia@stripe.com",
"subject": "Re: your demo",
"intent": "PURCHASE_INTENT",
"sentiment": +0.78,
"confidence": 0.94,
"extracted": {
"company": "Stripe",
"team_size": 12,
"current_vendor": "Klaviyo",
"geo": "US"
},
"channel": "email",
"contact_id": "cnt_amelia_9k2"
}
}Override it from the dashboard or via API. Corrections feed back into the next training round — your overrides become the gold-label set for your workspace, so the model gets better the more you push back on it. We never silently retrain without surfacing the diff.
No PII leaves our infrastructure for classification by default. Today we route through Anthropic Claude with PII-redacted prompts only, with OpenAI as fallback. Raw reply content stays in our database, encrypted at rest. The DPA spells the chain out. Self-hosted Groq inference is Q4 2026.
Q3 2026 — Pro and Business will be able to define workspace-scoped intent classes with 5–20 example replies. The model picks up new intents within ~3 hours. Today, classes are global to the platform.
English, Spanish, Portuguese out-of-the-box today. Arabic, Urdu, Hindi, French, German Q3 2026. We classify by detecting language first, then routing through the matching prompt set.
P50 ~22s · P95 <60s · P99 <90s, measured from inbound MIME timestamp to reply.received webhook delivery. Long tail is dominated by Gmail API delivery delay, not our inference.
Connect your inbox. Email Digit handles the rest. Most teams see classifications within an hour of setup.