Account Health Score AI Agent

What this agent does

Computes the score, never calls the account

A score is produced, and the signals that produced it are carried with it rather than left behind.

A signal drops out, and the score says how many inputs it was built from, not only what it came to.

Reason

A weight is applied, and the judgement that chose it is named as a judgement rather than as arithmetic.

A telemetry feed stops reporting, and the account it fed is marked thin instead of quietly staying green.

A score moves between reviews, and the component that moved it is shown beside the movement.

Decide

A new account is scored, and the cohort it is being compared against is stated on the record.

A score reaches a meeting, and a manager can argue with the components, not only with the colour.

Out

A signal cannot be measured at all, and that is written up as a gap rather than scored around.

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

→ Product statement

Computing, weighting and marking the gaps belong to the agent. Deciding an account is at risk belongs to a named success lead, who prioritises the week and owns that call.

Example workflow

One account, signals to decision

AgentHuman

1Signals receivedProduct telemetry, ticket history, sentiment records, sponsor changes and invoice ageing

2Coverage assessedWhich signals reported, which were stale, which never arrived and the day each was last read

3Score computedThe score, its components, the weights they carried and confidence

4Controls appliedCoverage checks, cohort checks, weight-provenance checks and scoring confidence

No human action required

Stages 1 to 4 run unaided, and no account is prioritised at any of them — the agent is computing, and the success lead lane opens at the coverage gate.

5DecisionSplits at the coverage gate

Coverage complete

Goes to the named success lead to read.

Anything thin

Adds a senior success read first.

Success lead review

The score is held with its components, its weights and the inputs it never received.

Accept · Adjust · Send to senior review

Accepted — by the named success lead▼

6Account and success records updatedOnly where write access and records policy allow it

7Outcome evaluatedCoverage completeness, weight stability, lead corrections and what review found

Corrections

Each lead correction is counted in the evaluation.

What should not run autonomously

Human approval stays in control

Outside the boundary — human approval required8 items

Deciding that an account is at risk.

Prioritising the accounts a manager works this week.

Choosing the weight a signal carries.

Triggering outreach on a falling score.

Automation boundaryAgent acts unaided

✓Compute the score and carry its components with it.

✓Mark which inputs were stale and which never arrived.

✓Show what the score would come to if one signal moved instead.

✓Name the cohort an account is scored against and when it was set.

No account is prioritised except by a named success lead, inside the agreed boundaries.

Judging why an account actually went quiet.

Telling a board that a book of accounts is healthy.

Ruling that a colour means the same to two teams.

Changes to the signals, the weights or the bands.

Example output

One account score, annotated

This serves a success team who may have to explain a colour to a manager who disagrees with it; below is one score exactly as the agent leaves it.

Health score output · single accountIllustrative example

Account

Recorded as

Components

Evidence of record

Confidence

Held for

Mid-market, second term

Computed, one feed silent

Components carried

Telemetry last read 3 August 2026

Held unactioned

The named success lead

As receivedTaken from the signals that reported and the ones that did not, and it concludes nothing beyond them.

What the record holds Product telemetry Ticket history Invoice ageing

Why no action hereCalling an account at risk is a judgement the success lead makes.

ActionAcceptAdjustSend to senior review

What the score decidesBelow the configured threshold a score gets a senior read before the lead sees it.

Value

Where AI adds value

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

Where the value landsValue 01 – 04

Every scoreFrom the signals under it

03Evidence

Where the components come from

Churn-risk outreach is a sibling page in this batch, and acting on a score belongs there. This page stops at the score, its components and the inputs it never received.

01Approved path

A score is not a diagnosis

Change what usage is worth against tickets and the same account changes colour, with no signal having moved and nobody having lied about anything.

02Human review

What was checked, and not found

Checked and not found: no statute, regulator or published standard governs an internal account health score, so there is no external rule to point at here and this page claims none.

04Build an evidence trail

The score, the signals under it and the manager who acted on it stay together.

Integrations

Typical integrations

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

Product telemetryPendo · Amplitude · Mixpanel
Feature usage and login frequency

Support and ticketingZendesk · Intercom · Front
Ticket volume and reopen rates

CRM and account recordsSalesforce · HubSpot · Gainsight
Sponsor changes and account tiering

Agent

Account health scoring

Reads the signals
Computes the score
Holds for the lead

Billing and invoicingNetSuite · Zuora · Chargebee
Invoice ageing and collections history

Observability & evaluationOpenTelemetry · Langfuse
Supported monitoring/evaluation sources

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

Agent controls

Six sluices between the model and the manager

Six sluices down one race, the last the narrowest. What runs on is drawn in the map below.

L6 · Outermost — last line of defenceInward → L1 · closest to the model

L6Rollback / safe modeWithhold the score and show only the signals when evaluation or production signals degrade.Roll back

L5Version monitoringTrack model, prompt and weighting rules, and note the version each score was computed under.Track

L4TraceabilityRecord each score, the signals under it, the weights applied and every read of the record.Record

L3Success lead releaseHold the score for a named success lead; the hold governs release, not whether the number is right.Gate

L2Coverage guardrailsTest each score against the signals it actually received, and refuse one whose feed has gone silent.Restrict

L1Confidence thresholdsRoute a thin score to a senior success read before it reaches a manager.Require review

Model coreScore computed — the components, the weights, the missing inputs and confidence

L1 – L2Test whether a score may stand

L3Leaves the account call to a success lead

L4 – L5Keep the score and the signals behind it

L6Withholds the score entirely when signals degrade

How Nestack evaluates it

Evaluate the whole computation — not only the score that comes out.

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

Surface — the score a manager acts on

Depth of coverage ▼

E1Final-output evaluationDid the score record the components and the weights it was actually built from?

E2Step-level evaluationDid the agent read the live feeds, the right period and the current weighting?

E3Tool evaluationDid it read and write the correct account and the correct score?

E4Confidence calibrationDo low-confidence scores actually attract more lead corrections?

E5Slice evaluationHow does performance change across specific account tiers?

E6Business outcomeHow many scores needed a correction before the lead accepted?

Floor — the signals a score rests on

Failure modes

Where each failure originates in the agent

Seven failure modes, each placed at the stage where it first shows.

Agent lifecycleDirection of processing →

01 · Retrieval1 mode

NM-03

Silent feed read as live

A feed that stopped reporting is read as current.

Stage gathersThe feeds, the tickets, the tiers and the dates

02 · Reasoning2 modes

NM-04

Score raised on thin coverage

A number is computed from fewer inputs.

NM-06

Weight applied without provenance

A weighting is used that nobody can source.

Stage proposesThe components, the weights and the missing inputs

03 · Tool / write2 modes

NM-02

Thin score passed forward

A score moves on without the senior read.

NM-05

Score bound to wrong account

The number is filed against another account.

Stage writesOnly where write access and approval policy allow it

04 · Output1 mode

NM-01

Green all quarter, then notice

The score never moved and the account left.

Stage returnsThe score a manager reads and a book is run on

05 · Change / Version1 mode

NM-07

Silent weighting drift

A weight changes while the stored score keeps the old one.

Stage tracksModel, prompt, weight rules and signal changes

Sev-1 · the agent prioritises an account Sev-2 · a thin score reaches a manager Sev-3 · a feed degrades, the score is held

See the complete Customer Success failure-mode catalogue →

Affected slices

Silent feeds absorb the corrections

A tier-level signal-coverage figure can read clean while accounts with silent feeds carry most of the corrections. Nestack reports the correction rate by account tier, not only in total.

Slice performance — reported separately, not only in aggregateIllustrative example

Slice	Failure rate	Lift	Status
Accounts with silent feeds	9.6%	3.7×	Review
Multi-product estates	6.9%	2.7×	Review
Accounts inside first term	4.3%	1.7×	Watch
Established single-product accounts	2.3%	0.9×	Normal

Bar: correction-rate lift vs. established-account baseline · scale 0–4.0× · tick marks the 2.0× review threshold 2 of 4 slices over threshold

Evidence-linked improvement

What a green account costs

A cycle ends when the score nobody could explain is a standing case. That suite is what the next score published is measured against.

Improvement cycle · five stagesSwitchback — the path turns at Improve and returns at Learn

01Detect

Correction rate rises on accounts with silent feeds.

02Diagnose

The account that scored green all quarter and gave notice in the first week of the next one is worked backwards until one cause is left standing.

03Improve

The change ships numbered, and the scores that forced it ride with it.

04Verify

One score case still failing is enough to hold the release back.

05Learn

It is kept for good, and the scoring rules are amended in the same commit.

Learn → DetectThe return edge. The next score published 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, sources, score computation, evaluation, integration, then production validation and handover.

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

01Signal-inventory and automation-boundary work.

02Telemetry, ticketing and billing sources.

03Signal-to-component and weight-source mapping.

04Signal ingestion and coverage checks.

05Component, weight and cohort binding.

06Coverage scoring and review routing.

07Success lead review workflow.

08Success-platform integration.

09Score and signal-coverage cases.

10Guardrails and scoring controls.

11Score-trail instrumentation.

12Deployment, documentation and Agent Care handover.

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 tier PilotOne account book, one product ProductionProduction success workflow AdvancedMultiple products / regions

Introduced at Pilot

Score computation to your signals✓✓✓

Named success lead release✓✓✓

Account-population baseline✓✓✓

Introduced at Production

Reporting by account tier—✓✓

Lead review workflow in your systems—✓✓

Approved write-back—✓✓

Product-telemetry integration—✓✓

Introduced at Advanced

Multi-signal weighting sets——✓

Cross-tier score packs——✓

Large account books——✓

Multi-product score controls——✓

Build price From $5,000 From $8,000 Custom quote

Final build priceConfirmed after discovery based on integrations, workflow complexity, account volume, approval controls and deployment requirements.

Separate from buildBuild 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.

You bringWe build with it

01Your live accounts and the signals each is scored on → Signal capture and coverage checksWeek 1

02Representative accounts, including ones you lost → Signal binding, weighting logic and the score baselineWeek 2

03Your review cadence and the success leads it names → Signal mapping, weighting review and the automation boundaryWeek 1

04Access to relevant APIs, feeds or exports → Telemetry, ticketing and billing source assessment, then integration setupWeek 2

05Scores you would not want defended → Signal cases and failure-mode testingWeek 4

06What no score may conclude → Coverage scoring, review routing, guardrails and release controlsWeek 3

07A named success lead who reads the score → Release to the named success lead, then pilot and production validationWeeks 5–6

Nothing else is required Deployment, documentation and Agent Care handover are ours.

Delivery timeline

Four phases across six weeks

Weeks are counted here, not drawn evenly; the fifth carries two phases because it genuinely does.

Phase W1W2W3W4W5W6

Discovery W1

Build W2 – W3

Evaluate W4 – W5

Pilot & Launch W5 – W6

Week focus W1Signal discovery, weighting review and the automation boundary W2Source integration and the account-population baseline W3Score computation, coverage logic and release controls W4Evaluation suite, signal cases and failure-mode testing W5Success-platform integration, pilot scores and targeted corrections W6One scoring cycle run under the success lead, then Agent Care handover

Reading the bandA bar sits only on the weeks its own work is named for, and the fifth week is doubled on purpose.

At the end of W6When the score record validates, Agent Care takes the agent on.

DurationSix-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.

Next step · Customer Success AI agent

Build an account health agent around the signals your last green account never showed.

Show us one book of accounts and the score each one carries. If a score in that book has never gone amber, and nobody could say today which of its feeds are still reporting, then it is being trusted rather than read. A score nobody can argue with comes back as a finding.

Discuss this agent →

Nestack Agents · Account healthAGT-CX-01 · Agent Care available after launch