Docs Dashboard features Reply Loop
Dashboard features · 07 of 17

Reply Loop — the wedge that makes Mama compound.

Reply Loop watches replies to your outbound, classifies them, and attributes outcomes back to the archetype that produced the brief. Every reply makes the next brief sharper. This is Mama's defensible feature — and the one most users don't discover until week 3.

Time: 6 min·Updated: 2026-05-25·Audience: every Mama user·Deep dive: long-form essay

TL;DR

Connect your sequencer. Reply Loop reads inbound replies, classifies them (interested · objection · unsubscribe · referral · noise), and feeds the outcome back into your archetype scores. The system gets smarter every week. Without Reply Loop, Mama is a one-way signal pipe. With it, Mama is a learning system.

01What Reply Loop is

A closed-loop system. Mama generates a brief → you send outreach via your sequencer → prospect replies → Reply Loop captures the reply → classifies it → attributes outcome to the archetype → archetype confidence updates → next brief in that archetype is informed by what happened.

The result: your archetype matcher gets sharper over time. Reply rates climb. The longer you use Mama, the harder it becomes to switch.

02Sequencer integration

Reply Loop requires read access to your sequencer to see reply events. Supported sequencers:

SequencerConnectionReply visibility
OutreachOAuthFull (with reply text)
SalesloftOAuthFull
ApolloAPI keyReply event + metadata, no text
SmartleadOAuthFull
InstantlyAPI keyFull
LemlistOAuthFull
MailshakeOAuthReply event + metadata
Reply.ioOAuthFull
HubSpot SequencesOAuthFull (via HubSpot connector)

Don't see your sequencer? Use the email-forward fallback: forward replies to your Mama-issued ingest address and the loop works without an API integration. Slower but functional.

03The 4-tier reply classifier

Every captured reply runs through a classifier that bins it into one of 4 outcomes (plus noise):

Interested
Reply expresses interest, asks a question, books a meeting. The win.
Objection
Polite no with a reason ("not now," "wrong contact," "already evaluating X"). Useful signal — the why matters.
Unsubscribe
"Take me off this list" or similar. Mama updates suppression lists across your sequencer + CRM.
Referral
"Talk to [name] instead." Mama prompts you to add the referred person to the account's decision-maker map.

The classifier runs locally on the reply text — no third-party LLM call on your data. Accuracy is ~92% based on internal eval. Misclassifications can be corrected in-app and the model adapts to your team's patterns over time.

04How attribution works

Every brief carries an archetype tag. When a reply arrives, Reply Loop looks up the archetype and updates:

  • Reply rate for that archetype (rolling 90-day window)
  • Outcome mix (% interested / objection / unsubscribe)
  • Confidence score based on volume + outcome consistency
  • Sub-archetype proposals if a pattern emerges within the archetype that replies differently

The archetype's metrics show on its card in the Archetypes screen and influence lookalike ranking.

05The 5-stage flywheel

StageWhat happensLatency
1 · Brief generatedArchetype tag attachedReal-time
2 · Outbound sentMama logs send event from sequencer webhook< 5 min after send
3 · Reply receivedReply event captured from sequencer< 5 min after reply
4 · Reply classifiedClassifier bins into 1 of 4 outcomes< 30s after capture
5 · Archetype updatedReply rate, mix, confidence recalculatedReal-time

06Privacy and data handling

Reply Loop reads inbound emails (from your sequencer's API). Three privacy guarantees:

  • No third-party LLM. Classification runs on Mama-controlled infrastructure.
  • Reply text retention: 90 days by default, configurable down to 0 (metadata-only).
  • No use for model training across workspaces. Your replies inform your archetypes only.

Full data-handling details: /security.

07Common mistakes

Not connecting your sequencer
Without it, Reply Loop has nothing to work with. Mama becomes one-way. The archetype matcher doesn't improve. This is the #1 reason teams plateau in Mama at week 3.
Manually re-classifying every misclassification
The classifier is ~92% accurate. Re-classifying just the obviously-wrong ones is fine — the model adapts. Re-classifying every reply burns time.
Ignoring sub-archetype proposals
When Reply Loop notices a sub-cluster within an archetype that replies 2× higher, it proposes a new archetype. Accept these proposals. They're how the matcher gets sharper.
Was this page helpful?