What you can see, and what you can't.
Measure is the function with the most software and the least reconciliation. It's the difference between two dashboards nobody trusts and one read both populations are scored against. What you count, what you ignore, what each number tells you about whether the operation is working. Humans and AI together.
v 02 · live
What Measure means.
Measure is the work of making the operation visible. Both halves of it. Not the work of building dashboards. Not the work of monitoring KPIs. The work of choosing what to count across humans and AI, what to ignore, what each number actually tells you about the operation, and producing one read instead of two reports.
Most CX teams measure too much on the human side and inherit whatever the AI vendor surfaces. CSAT, NPS, FCR, AHT, ASA, abandonment, escalation rate, transfer rate, contact-per-customer, repeat contact rate, response time SLA, resolution time SLA. On the other side: deflection, containment, intent recognition, false-resolution. Twenty metrics, two dashboards, three different read-outs of the same operation. The metrics are ritual on both sides, not signal.
That's the gap Measure names. Most metrics are inherited rather than chosen. Most dashboards are populated rather than designed. The vendor's dashboard is accepted as truth without verification. Most reviews discuss what changed on one side without asking why, or whether the metric is even decision-grade, or whether the other side moved in the opposite direction at the same time. Measurement becomes performance, not signal.
Haven's Measure module starts with the metric audit across both populations. Every metric tied to a decision it could trigger. Every metric without a decision retired. One read of the operation, not two. The dashboard becomes deliberate.
Decision-grade metrics get the attention. Performance metrics get the right altitude. The operation responds faster because the signals are cleaner and reconciled. When the CEO asks how CX is performing, the answer is one number with the working underneath it, not a 20-slide deck stapled to a two-page bot performance section.
The progression. Four levels.
Two dashboards, neither trusted. Human metrics inherited from previous leaders. AI metrics inherited from the vendor's defaults. Most reviews are theatre on one side and silence on the other. Decisions happen on instinct, despite the data.
- Inherited human metric set
- Vendor dashboard accepted without verification
- Review theatre
- Decisions despite data
Some metrics are trusted on the human side. The AI's still on its own. The leadership team has favourites. The vendor dashboard reports on its own terms. The two are reconciled by hand in a slide before each board meeting. Reviews surface obvious changes but rarely ask why, and rarely connect across the two.
- Trusted handful of human metrics
- AI vendor dashboard runs in parallel
- Reconciliation happens manually in slides
- Why questions rare; cross-population questions rarer
Every metric has a named decision and is read across both populations. The dashboard is shorter. Human and AI metrics are reconciled into one read. Reviews ask why before what. Metrics without decisions get retired without ceremony.
- Decision-tied metrics
- One reconciled read across humans and AI
- Why-first reviews
- Retirement is routine
The metric set evolves with the operation and reads across both populations natively. New decisions trigger new metrics. Solved problems retire their metrics. AI scope changes adjust metric baselines so quarter-on-quarter comparisons hold. The dashboard is a living instrument, not a museum, and not two museums.
- Living metric set, native across humans and AI
- New decisions create new metrics
- Baselines adjust automatically as AI scope changes
- Dashboard as instrument; one read, always
What Measure builds.
The metric audit
Every metric tied to a decision, across humans and AI both. Metrics without decisions retired. Vendor defaults verified or dropped. The dashboard becomes a tool, not a wall, and not two walls.
- Each metric mapped to a decision
- Human and AI metrics reconciled into one read
- Vendor-default metrics audited before they're trusted
- Owner & cadence per surviving metric
The decision register
A named list of operational decisions and the metrics that trigger each one. Same triggers apply whether the case was handled by a human or the AI. The bridge between data and action.
- Decisions named, not just metrics
- Trigger thresholds per decision
- Same triggers apply across humans and AI
- Action owners & escalation path
The review cadence
A weekly or fortnightly review structured around decisions, not numbers. One read across both populations. Why before what. Action before report.
- Structured around decisions, not numbers
- One read across humans and AI
- Why before what, action before report
- Decision log published every week
See it cascade.
When the AI's scope expands or volume shifts, Measure is what keeps the baselines honest. Otherwise next quarter's numbers look like progress when they're just a different denominator. Haven adjusts the read so a like-for-like comparison still holds. See how the cascade keeps Measure honest →