Customer · Customer Account Monitor

A daily written record of every account's health.

So nothing about a customer slips through. Every morning the Customer Account Monitor scans each account you have, pulls the signals from your connected tools into an evidence bundle, and classifies the account green, yellow, or red with a written reason. It then appends one immutable paragraph to that account's diary — an account history that compounds run after run. Every signal it cites has a source. It raises an internal alert when an account's status changes; it never emails your customers.

From connected tools to an immutable diary entry

Daily scan

06:15 UTC — walks every account.

Evidence bundle

Connected-tool signals, deduped by source.

Classify & log

Green / yellow / red, cited → the diary.

Detect transitions

A status change raises an internal alert.

A score is a number with no memory. The diary is a paragraph per account per day, each one anchored to real signals — the accumulating history is the part a clever one-off prompt can never reproduce. Once written, an entry is never edited or deleted.

The contract

What it runs on its own, what it checks with you, what it won't touch.

Runs on its own

Scans every account on your plan once a day — not just accounts near a renewal, every one.
Builds an evidence bundle from the signals on each account in your connected tools.
Classifies each account green / yellow / red and writes the reason in plain English.
Appends one immutable diary entry per account per day — the account history that compounds.
Cites a real source for every signal it references; a fabricated source ID forces the row to abstain.
Runs a second, independent review from a different model vendor when an account's status changes.

Checks with you first

Starts in observe-only mode — it writes the diary from day one, but status changes stay in the review inbox and do not page you. You flip it to active alerts from Settings once you have read the early data.
Abstains when the evidence conflicts or is too thin to support a confident call — the row lands flagged with a reason, never a guessed verdict.
Marks low-confidence classifications explicitly so you can see exactly which accounts it is unsure about and why.

Won't touch

Fire an alert on an unverified signal — every cited source must exist in the account's evidence bundle, or the row abstains.
Flip an account's status without cited evidence — a classification with no anchored signal behind it is rejected by the output validator.
Email your customers — the Customer Account Monitor raises internal alerts only. Customer-facing renewal email is the separate Renewal Manager's job.
Hand you an opaque churn score — there is no '0.34 risk' float; the reason and its citations are the output.
Change a customer's subscription, billing, or contract — it is read-only on every connected tool.

Pairs well with the Renewal Manager — the Monitor watches every account year-round and writes the history; the Renewal Manager reads that history to drive the save play when a renewal comes up. Each is sold and run on its own.

What you get

This is what the worker appends to an account's diary.

Every entry carries the status, the transition, the written reason, and the exact signals it is grounded in. Yesterday's entry stays right above today's — that is the account history.

Account diary · Northwind Apps

yellow → red

Today · 06:15 UTC red transition: degraded cross-vendor review: ran

Northwind moves to red. Two failed invoice retries in the last 30 days, and the support thread opened last week is still unresolved with a frustrated tone from their admin. The champion who ran the original rollout has had no logged activity for 41 days. Last entry on the 14th already flagged the billing retries as a watch item; today's unresolved support escalation pushes this past yellow.

Signals cited — each one resolves to a source

stripe:in_4Qr8 — invoice retry failed
stripe:in_4Qs1 — invoice retry failed
zendesk:tkt_8812 — support thread, unresolved 6d
account_health:no_activity_41d — champion inactivity

8 days ago · 06:15 UTC yellow transition: degraded

Northwind moves to yellow. One invoice retry failed on the 13th and a second failed today — first billing trouble on this account in the 90-day window. Usage and support are otherwise quiet. Watch item, not yet an alert.

What you can trust about it

Append-only

The entry is the evidence. Each one is anchored to real signals, and the diary is never rewritten.

Every signal in the reason resolves to a real source — invoice, ticket, recorded inactivity. A fabricated source ID would have forced the row to abstain instead.
The yellow → red transition is marked explicitly, and a status change triggers an independent review by a model from a different vendor.
Yesterday's entry was not edited or deleted to write today's — the diary is append-only, so the account history is always honest.
It quotes its own prior entry when a status flips, so you see the arc, not just the snapshot.

A status change surfaces in the review inbox first. It pages you as an alert only after you switch the worker out of observe-only mode.

Pricing

Three tiers by account count, plus Enterprise.

Each tier covers the daily scan, the evidence bundling, the green / yellow / red classification, the append-only diary, the source-anchored validator, and the transition digest. The price is set by how many accounts you monitor.

What it replaces

Doing this in-house means a full salary, benefits, and months of ramp — for one function. Ataski ships the output from day one at a flat, predictable monthly price.

Enterprise-grade quality for a fraction of a hire. Scale up or pause anytime — you only pay for the months you use.

Team

$449 / month

up to 200 accounts

≈ $2.25 per customer monitored

Up to 200 accounts monitored
Daily scan + green / yellow / red classification
Append-only per-account diary
Weekly transition digest

Series A B2B SaaS with one or two CSMs and no dedicated CS analyst yet. The entry point.

Choose Team

Most picked

Scale

$1,299 / month

up to 750 accounts

≈ $2.60 per customer monitored

Up to 750 accounts monitored
Everything in Team
Cross-vendor review on status transitions
Source-anchored evidence on every entry

Funded Series B. Adds headroom for the cross-vendor second review on every status transition.

Choose Scale

Business

$3,499+ / month

up to 2,500 accounts

Up to 2,500 accounts monitored
Everything in Scale
Multi-team review handoff
Priority on the daily scan window

Post-Series B with a large installed base and multiple teams sharing the inbox.

Choose Business

Enterprise

Custom

unlimited accounts

Unlimited accounts monitored
BYO LLM keys + regional data residency
99.9% SLA on the weekly digest
Dedicated onboarding

High-volume installed bases needing BYO LLM keys, data residency, and an SLA on the digest.

Talk to sales

A per-customer daily cap bounds the LLM spend on any single account, so a noisy account cannot run up the bill. The LLM cost of every scan is visible on /app/cost. Cancel monthly.

Setup

About ten minutes from sign-in to your first scan.

01
Sign in and open setup. Hire the Customer Account Monitor and land on /app/health/setup — the first-run onboarding.
02
Connect your tools. Your CRM, support tool, and product telemetry. Each one degrades cleanly; connect more later.
03
Confirm the account list. The setup step loads the accounts to monitor. Pick a tier that covers how many you have.
04
Watch the diary fill in. The next daily scan (06:15 UTC) writes the first diary entry for every account. It starts in observe-only mode.

For the first 30 days the worker runs in observe-only mode — it writes the diary but does not page you on status changes. When you have read the early data and trust the calls, flip it to active alerts from Settings.

Under the hood

How the routing and guardrails work.

Model routing — tiered by stakes: Quiet accounts go to a fast, low-cost model; busier accounts to a stronger reasoning model; a status change escalates to a cross-vendor second review
The cross-vendor review: A second, independent AI from a different vendor reviews the call, and a third breaks a genuine tie — the cited reason is the only output, no opaque risk score
Signal sources: Salesforce, Zendesk, Intercom, and product telemetry (Mixpanel, Amplitude, Segment, PostHog) — each source degrades silently
Replay tape: The worker keeps a 90-day replay tape of every signal it has seen
Cost discipline: A per-customer daily LLM cap bounds spend on any single account; a global kill switch bounds the day
Data discipline: Read-only on every connected tool; an append-only diary enforced at the database grant; an audit log on every call; 30-day data deletion on offboarding

Start the diary on your accounts.

Hire the Customer Account Monitor, connect your tools, and the next 06:15 UTC scan writes the first entry for every account. It runs in observe-only mode while you read the early data — nothing pages you until you say so.