Usage declining
Weekly active users down 34% over six weeks, concentrated in one team.
How it works
From scattered SaaS signals to coordinated Customer Success action.
Four layers, in a fixed order. Your data is validated before it is scored. Deterministic, versioned logic decides what needs attention. The Risk agent explains and enriches on top of that decision — never instead of it. And nothing reaches a customer without a person approving it.
Your data, structured
Customer Success data arrives as spreadsheets, exports and things people remember. The first layer turns that into a weekly, per-account record the rest of the platform can be held accountable to.
Collect
Weekly account data arrives by spreadsheet upload or direct entry. Integrations come later; nothing waits on them.
Define
Each measure is defined for your business, in your vocabulary. There is no built-in catalogue you have to bend your data into.
Normalize
Values are mapped, typed and stamped to a week, so week 14 always means the same thing.
Validate
Data is checked for completeness, validity, consistency and freshness before anything is calculated from it.
Governed decisioning
Calculated, not generated. The same inputs produce the same decision, every time, against a configuration version you can name — which is what makes the outcome reproducible and the audit meaningful.
Data Quality Gate
Every week is graded Ready, Limited or Blocked. Incomplete or stale inputs never produce a confident-looking wrong answer — they produce a stated one.
Health
KPIs and health components calculated from your definitions, weightings and thresholds, not ours.
Attention
Accounts ranked by what actually needs a person this week, weighted by revenue, lifecycle and renewal proximity.
Playbook routing
Deterministic rules map the account's state to a CS motion, stamped with the exact rule and configuration version applied.
This is the part of AgentIQ that is built and tested today, and the part that runs without a language model at all.
AI intelligence
The agent layer reads what the engines already decided and does the work a CS manager does not have time for: assembling the account's evidence and working out what it means. It never produces the numbers — it explains them.
Investigate
Pull together the account's calculated evidence, its onboarding context and the history of what was tried before.
Interpret
Turn that evidence into a business reading, with the stated confidence and the limits of what the data supports.
Explain
Show the records behind every conclusion, and what argues against it.
Recommend
Sharpen the routed motion into a specific next step, grounded in the evidence rather than in a template.
The Risk agent is live today. Adoption, Expansion, Support Intelligence, CS Copilot and Executive Reporting follow on the same runtime. And the layer below stays deterministic — if an agent is unavailable, the health score, the routing and the report are still produced.
Controlled execution
A recommendation is a proposal, not an instruction. The gap between the two is deliberate.
Prepare
The motion is drafted with its evidence attached, ready to review.
Approve
A person decides. The approval gate is part of the pipeline, not a setting.
Act
The approved motion runs, and what ran is recorded.
Close the loop
The next week is unlocked only once you record what was actually done and what came of it. That record becomes an input to the next decision.
What this looks like on one account
How the Risk agent reasons over the evidence the deterministic engines produce. Four signals arrive from four sources. The agent does not simply add them up — it forms a hypothesis, looks for what contradicts it, states what it is missing, and proposes a next step.
Signals in
Support escalation
Three unresolved P2 tickets on the reporting module, oldest open 19 days.
Renewal approaching
62 days to renewal, with no expansion discussion logged.
Sponsor inactive
The executive sponsor has missed the last two quarterly reviews.
Analysis out
Risk hypothesis
Renewal is at risk because the reporting workflow that justified the original purchase has degraded, and the sponsor who championed it is no longer engaged.
Contradictory evidence
Seat count is unchanged, invoices are paid on time, and a second team increased usage 18% over the same period.
Missing information
Whether the sponsor has changed role, and whether the reporting tickets have a committed fix date.
Recommended next action
Open a sponsor re-engagement motion before the renewal window, with the support fix date as the reason to meet.
Illustrative. Northstar Labs is not a customer; the account and its evidence are constructed to show the shape of the reasoning.
One account picture. One governed decision trail.
See it run on an account you know.
The reasoning is easier to judge against a real account than a constructed one. Bring one you already have an opinion about.