Main site

Customer Success AI agent · Churn risk

Churn-Risk Outreach AI Agent

Say why before proposing anything: the outreach carries the reason and the evidence under it, clears the suppression rules, and waits for the retention lead who owns the relationship.

Typical delivery
4–6 weeks
Deployment
Your stack
Retention lead
Reason first
After launch
Agent Care

What this agent does

Proposes the outreach, never sends it

In

  1. 01

    A risk is raised, and the reason it was raised travels beside it rather than a score standing alone.

  2. 02

    An outreach is proposed, never sent, and the evidence a person could argue with sits underneath it.

Reason

  1. 03

    A signal cannot say why, and nothing is drafted — the account owner is asked what is actually happening.

  2. 04

    A suppression rule fires, and the account it held back is recorded, because an unlogged hold is a lost signal.

  3. 05

    An account is in a live escalation, and no check-in written to sound warm goes anywhere near it.

Decide

  1. 06

    An outreach names a contact who has left, and it stops there rather than reaching a mailbox nobody reads.

  2. 07

    An account was reached by another team this week, and the second message is held and shown to both.

Out

  1. 08

    A risk band moves, and what moved it is set beside the movement rather than a fresh number appearing.

  2. 09

    Execute write actions only inside the approval boundaries agreed during implementation.

Product statement

Proposing, evidencing and suppressing belong to the agent. Sending belongs to a named retention lead, who owns the relationship and answers for the message.

Example workflow

One risk signal, raised to released

  1. 1Risk signal receivedProduct telemetry, ticket history, invoice ageing, sponsor changes or a CSM note
  2. 2Reason assembledWhat moved, when it moved, who it involved and the account record it was read from
  3. 3Outreach proposedThe drafted message, the reason, the evidence and confidence
  4. 4Controls appliedSuppression checks, contact checks, recent-contact checks and confidence threshold

No human action required

Stages 1 to 4 run unaided, and nothing reaches a customer at any of them — the agent is proposing, and the retention lane opens at the suppression gate.

5DecisionBranches at the suppression gate
Clear of every rule

Goes to the retention lead to release.

Anything holding

Logs the hold and names the rule.

Retention lead review

The message is held with its reason, its evidence and any suppression that fired.

Release · Rewrite · Hand to account owner

Released — by the named retention lead

  1. 6Account and outreach records updatedOnly where write access and records policy allow it
  2. 7Outcome evaluatedReason quality, suppression accuracy, lead rewrites and what the account said back

Rewrites

Each lead rewrite is counted in the evaluation.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required

8 items

  • Sending any message to a customer.
    • Deciding that an account is at risk.
    • Offering a discount or a commercial concession.
    • Judging whether a relationship can be recovered.

  • Automation boundary

    Agent acts unaided

    • ✓Draft the outreach and attach the reason to it.
    • ✓Record every suppression and the rule that fired.
    • ✓Carry the evidence a person could argue with under the message.
    • ✓Hold a proposal that cannot say why, and route it to the owner.

    Nothing reaches a customer except by a named retention lead, inside the agreed boundaries.

    • Choosing which accounts get worked this week.
    • Telling an executive that an account is saved.
    • Overriding a suppression rule that has fired.
  • Changes to the signals, bands or suppression rules.

Example output

One proposed outreach, annotated

This serves a success team who may have to explain to a customer why a message arrived at all; below is one proposal exactly as the agent leaves it.

Outreach proposal · single account

Illustrative example
Account
Mid-market, renewal in spring
Recorded as
Drafted, reason attached
Reason
Suppression checks clear
Evidence of record
Rollout owner left, 3 August 2026
Confidence
Held unsent
Held for
The named retention lead

As received

Taken from the account record and the signals under it, and it asserts nothing the record does not carry.

What the record holds

  • Sponsor-change record
  • Support thread history
  • Product usage record

Why this reason

The message names the change, which is the part a lead can actually use.

What the score decides

Below the configured threshold a proposal gets a senior read before the lead sees it.

Action

ReleaseRewriteHand to account owner

Value

Where AI adds value

The same four claims, placed at the point in the workflow where each one applies.

Where the value lands

Value 01 – 04

Every risk signal

From the account it came from

03Evidence

Where the reason comes from

Churn agents in media, fitness and telecom are consumer subscriptions built around the cancel path; this is a B2B account with no cancel button, a renewal months off, and the outreach lands wrong.

01Approved path

Reaching out is a decision

Move the threshold a band opens at and a different set of accounts is written to this week, with no relationship having changed and nobody having lied.

02Human review

What was checked, and not found

Checked and not found: no statute, regulator or published standard says when a supplier must contact an account at risk, or what a message written to save one has to contain, so this page points at none.

04

Build an evidence trail

The outreach, the reason it was raised and the person who sent it stay together.

Integrations

Typical integrations

Five system groups connect to the same agent. Which of them are in scope is decided in discovery.

  • Customer success platforms

    • Gainsight
    • Totango
    • Catalyst
    • ChurnZero
    • Planhat
    • Vitally

  • Product telemetry

    • Amplitude
    • Pendo
    • Mixpanel
    • Segment
    • in-app event feeds

  • Support and ticketing

    • Zendesk
    • Intercom
    • Freshdesk
    • Escalation and live incident records

Agent

Churn-risk outreach

  • Reads the signals
  • Proposes the outreach
  • Holds for the lead

  • CRM and messaging

    • Salesforce
    • HubSpot
    • Dynamics
    • Outreach and email sequencing tools

  • Observability & evaluation

    • OpenTelemetry
    • Langfuse
    • Supported monitoring/evaluation sources

Integration availability depends on the client's existing systems and API access.

Agent controls

Six weighbridges between the model and the customer

Six weighbridges in a row, the last the truest. What weighs in is drawn in the map below.

L6 · Outermost — last line of defence

Inward → L1 · closest to the model

L6Rollback / safe modeNarrow the agent to signal listing when evaluation or production signals degrade.Roll back
L5Version monitoringTrack model, prompt and suppression rules, and note the version each proposal was drafted under.Track
L4TraceabilityRecord each proposal, the reason under it, the holds that fired and every release to a customer.Record
L3Retention releaseHold the message for a named retention lead; the hold governs release, not whether the message reads well.Gate
L2Suppression rulesTest each proposal against the configured holds, and return anything touching a live escalation.Restrict
L1Confidence thresholdsRoute a thin reason to a senior retention read before the message reaches the lead.Require review
Model coreOutreach proposed — the draft, the reason, the evidence and confidence
L1 – L2
Test whether a message may stand
L3
Leaves the sending to a named retention lead
L4 – L5
Keep the outreach and the reason behind it
L6
Sends nothing to the customer when signals degrade

How Nestack evaluates it

Evaluate the whole path to the message — not only the message that goes out.

Coverage runs the whole depth of the workflow, and every layer is cut by slice.

Surface — the message a customer reads

  1. E1Final-output evaluationDid the message say why, in terms the account would recognise?
  2. E2Step-level evaluationDid the agent read the right account, the right signals and the live suppression rules?
  3. E3Tool evaluationDid it read and write the correct account and the correct contact?
  4. E4Confidence calibrationDo low-confidence proposals actually attract more lead rewrites?
  5. E5Slice evaluationHow does performance change across specific risk bands?
  6. E6Business outcomeHow many proposals needed a rewrite before the lead released them?

Floor — the reason a message rests on

Failure modes

Where each failure originates in the agent

Seven failure modes, each fixed to the stage where it first surfaces.

Agent lifecycle

Direction of processing →

  1. 01 · Retrieval

    1 mode

    • NW-03

      Departed contact read as live

      The named contact left and the record kept them.

    Stage gathers

    The signals, the tickets, the owners and the dates

  2. 02 · Reasoning

    2 modes

    • NW-04

      Outreach raised on no reason

      A message appears with no cause behind it.

    • NW-06

      Live escalation missed by a hold

      A rule reads an open incident as clear.

    Stage proposes

    The draft, the reason, the evidence and the holds

  3. 03 · Tool / write

    2 modes

    • NW-02

      Thin reason passed forward

      A proposal moves on without the senior read.

    • NW-05

      Message bound to wrong account

      The draft is filed against another account.

    Stage writes

    Only where write access and approval policy allow it

  4. 04 · Output

    1 mode

    • NW-01

      Reached after the decision

      The save message arrived once the board had been told.

    Stage returns

    The message a customer reads and then answers

  5. 05 · Change / Version

    1 mode

    • NW-07

      Silent threshold drift

      A band widens while the stored proposals keep the old one.

    Stage tracks

    Model, prompt, band rules and hold-rule changes

  • Sev-1 · a message reaches a customer
  • Sev-2 · outreach lands with no reason
  • Sev-3 · a signal degrades, the draft is held
See the complete Customer Success failure-mode catalogue

Affected slices

Sponsor changes absorb the rewrites

A band-level justification figure can read clean while accounts that lost a sponsor carry most of the rewrites. Nestack reports the rewrite rate by risk band, not only in total.

Slice performance — reported separately, not only in aggregate

Illustrative example
SliceFailure rateLiftLift vs. thresholdStatus
Accounts after a sponsor change10.2%3.7×Review
Accounts inside a live escalation7.3%2.6×Review
Multi-team accounts4.5%1.6×Watch
Steady renewals, single owner1.9%0.7×Normal

Bar: rewrite-rate lift vs. steady-renewal baseline · scale 0–4.0× · tick marks the 2.0× review threshold

2 of 4 slices over threshold

Evidence-linked improvement

What a late message costs

A loop closes when the outreach nobody could justify is a standing case. That suite is what the next message sent is measured against.

Improvement cycle · five stages

Switchback — the path turns at Improve and returns at Learn

  1. 01

    Detect

    Rewrite rate rises on accounts after a sponsor change.

  2. 02

    Diagnose

    The save email that arrived the morning after the customer had already told their own board they were leaving is worked backwards until one cause is left standing.

  3. 03

    Improve

    Changes leave numbered, and the messages that forced them are filed underneath.

  4. 04

    Verify

    One message case still failing is enough to stop the release.

  5. 05

    Learn

    One case joins the suite, one line joins the outreach rules.

  6. Learn → Detect

    The return edge. The next message sent is measured against a suite one case longer.

Typical build scope

Twelve workstreams across six weeks

The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, signals, outreach drafting, evaluation, integration, then production validation and handover.

Workstream

Week 1Week 2Week 3Week 4Week 5Week 6

  1. 01Suppression-rule and automation-boundary work.Week 1
  2. 02Signal, ticketing and billing sources.Week 1, Week 2
  3. 03Signal-to-outreach and justification-coverage mapping.Week 1, Week 2
  4. 04Risk-signal ingestion and reason binding.Week 2
  5. 05Reason, evidence and contact binding.Week 2, Week 3
  6. 06Suppression checks and review routing.Week 3
  7. 07Retention lead review workflow.Week 3, Week 4
  8. 08Outreach-system integration.Week 5
  9. 09Reason and suppression cases.Week 4, Week 5
  10. 10Guardrails and outreach controls.Week 4
  11. 11Outreach-trail instrumentation.Week 5, Week 6
  12. 12Deployment, documentation and Agent Care handover.Week 6

12 workstreams · 6 weeks · bar shows the weeks a workstream is active — several run in parallel

Final scope and sequence confirmed in discovery

Engagement tiers

What each tier includes

Rows are the capabilities named in each tier's scope. Higher tiers include everything below them.

Capability✓ in scope · — not at this tierPilotOne account book, one regionProductionProduction retention workflowAdvancedMultiple books / regions
Introduced at Pilot
Outreach drafting to your signals✓✓✓
Named retention lead release✓✓✓
Risk-population baseline✓✓✓
Introduced at Production
Reporting by risk band—✓✓
Lead review workflow in your systems—✓✓
Approved write-back—✓✓
Customer-signal integration—✓✓
Introduced at Advanced
Multi-rule suppression sets——✓
Cross-team outreach co-ordination——✓
Large account books——✓
Multi-channel outreach controls——✓
Build priceFrom $5,000From $8,000Custom quote
Final build price
Confirmed after discovery based on integrations, workflow complexity, risk volume, approval controls and deployment requirements.
Separate from build
Build pricing is separate from recurring Agent Care, which covers managed monitoring, evaluations, incidents and verified improvements after launch.

What we need from you

What you bring, and what we build with it

Each input maps to a piece of build scope and a week in the delivery timeline.

  1. 01Your live accounts and the signals each is watched on

    Signal capture and reason assembly

    Week 1

  2. 02Real messages you sent, including the late ones

    Signal binding, reason logic and the outreach baseline

    Week 2

  3. 03Your suppression rules and the leads they name

    Suppression-rule mapping and the automation boundary

    Week 1

  4. 04Access to relevant APIs, feeds or exports

    Signal, ticketing and billing source assessment, then integration setup

    Week 2

  5. 05Messages you would not want read back

    Suppression cases and the evaluation run

    Week 4

  6. 06What no outreach may repair

    Suppression checks, review routing, guardrails and release controls

    Week 3

  7. 07A named retention lead who sends the message

    Release to the named retention lead, then pilot and production validation

    Weeks 5–6

Nothing else is required

Deployment, documentation and Agent Care handover are ours.

Delivery timeline

Four phases across six weeks

Read the widths as effort spent rather than layout; the fifth week genuinely carries two phases.

Phase

W1

W2

W3

W4

W5

W6

  1. DiscoveryW1
  2. BuildW2 – W3
  3. EvaluateW4 – W5
  4. Pilot & LaunchW5 – W6

Week focus

W1Risk-signal discovery, suppression rules and the automation boundary

W2Source integration and the risk-population baseline

W3Outreach drafting, reason logic and release controls

W4Evaluation suite, suppression cases and failure-mode testing

W5Outreach-system integration, pilot messages and targeted corrections

W6One risk cycle run under the retention lead, then Agent Care handover

Reading the band
Each bar covers the weeks its own work is named in, and nothing else; week five doubles by design.
At the end of W6
When the outreach record validates, Agent Care takes the agent on.
Duration
Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.

Next step · Customer Success AI agent

Build a churn-risk outreach agent around the reason your last save message never gave.

One message, sent the morning after the customer had already told their own board. Show us your risk bands and the last outreach each one produced. If nobody can say today why that message went out, a threshold reached out rather than a person, and that comes back as a finding.

Typical delivery
4–6 weeks
Deployment
Your risk signals
Retention lead
Reason first
After launch
Agent Care available

Nestack Agents · Churn-risk outreach

AGT-CX-07 · Agent Care available after launch

Build a churn-risk outreach agent

Four questions so the call starts from your actual setup — then pick a time that suits you.

About you
Your accounts

No obligation. We’ll map the workflow, name the automation boundary and tell you if an agent is the wrong answer.