The 12 metrics we track internally for Avery NXR's product health

Most companies don't publish their internal product metrics. We're going to share ours.

This isn't a flex about being open. It's an attempt to model what kind of metrics matter for AI agent platforms specifically — and to be honest about what's hard to measure.

If you're a product or company operator in this space, the framework might be useful. If you're a buyer evaluating us, knowing what we measure tells you something about how we think.

Why we measure what we measure

We organized our metrics around three questions:

→ Are agents getting created? (Acquisition) → Are the agents that get created delivering value? (Activation + retention) → Is the platform overall sustainable? (Business health)

The metrics below answer these three questions.

The 12 metrics

1. Active installs (DAU)

Definition: unique installs that processed at least one agent execution in the past day.

Why: tells us whether the platform is being actively used vs. installed and abandoned. Active is more valuable than installed.

Current trajectory: growing month-over-month. Specific numbers we don't share publicly.

2. New install → first agent created (conversion rate)

Definition: of users who install Avery NXR, what % configure their first agent within 7 days?

Why: tells us whether installation friction or "now what?" friction is killing adoption.

Current state: ~65%. The 35% who don't configure are mostly people who installed to evaluate, didn't follow through.

3. First agent → second agent (expansion rate)

Definition: of users who configure one agent, what % configure a second within 30 days?

Why: signals whether the first agent provides enough value to motivate more building.

Current state: ~78%. High because templates are pre-loaded.

4. Active agents per active user

Definition: average number of agents running per active user.

Why: tells us about depth of platform usage. More agents per user = more value being created.

Current state: median ~5. P75 ~9. P95 ~18.

5. Agent retention at month 3, 6, 12

Definition: of agents created in month X, what % are still running at X+3, X+6, X+12?

Why: durability of value created. Detail in [post 215].

Current state: Month 3 ~67%, Month 6 ~52%, Month 12 ~41%. Improving.

6. Time from install to "I trust this" moment

Definition: harder to measure. Operational proxy: time from install to user explicitly recommending to a teammate or upgrading to Pro tier.

Why: signals when users transition from "trying" to "committed."

Current state: Median ~21 days. We want this faster.

7. Audit ledger query frequency

Definition: how often users query the audit ledger.

Why: tells us whether users actually USE the audit transparency feature or just expect it as compliance theater.

Current state: ~40% of active users query at least monthly. We track patterns of when (incident investigation, periodic review, etc.).

8. Connector authentication health

Definition: % of configured connections that are still functional (not auth-expired or breaking).

Why: connectors broken → agents fail → users lose trust.

Current state: ~96% healthy. The 4% issues are mostly users whose external service credentials expired.

9. Multi-user expansion

Definition: of customers on Pro tier, what % have added more than one user within 90 days?

Why: tells us whether champion-to-team expansion is happening.

Current state: ~70%. Higher than we expected.

10. NPS / customer satisfaction (sampled)

Definition: standard NPS measurement, sampled quarterly across active customers.

Why: customer sentiment, not just behavioral metrics.

Current state: NPS in the 50s. Strong but room to improve.

11. Free → Pro conversion

Definition: of Free Desktop users, what % upgrade to Pro within 6 months?

Why: business health. Pro revenue is what funds the company.

Current state: ~14%. Healthy for our model — Free Desktop is generous, so converting 14% to paid is good economic outcome.

12. Customer-reported time saved per agent

Definition: when customers self-report (in surveys, interviews), how much time do they say each agent saves them per week?

Why: subjective but matters. Customer-perceived value is what drives renewal + referrals.

Current state: Median report ~3-5 hours/week per active agent. Power users report 10+.

What these metrics tell us together

When we look at all 12 metrics:

What's working:

→ Acquisition is healthy (DAU growing month-over-month) → First agent conversion is strong (templates working) → Multi-user expansion is good (champion model working) → Connector health is high (engineering investment paying off) → NPS is solid (customers like the product)

What needs improvement:

→ Agent retention at month 12 (41% is not great) → Time from install to "trust" (21 days is too long) → Free → Pro conversion (14% is okay but could grow) → Audit ledger discovery (60% of users don't query it)

What we're working on for each issue:

→ Retention: better maintenance reminders, better ownership prompts, better connector health alerts → Time to trust: improved onboarding flow, better starter pack guidance → Free → Pro: clearer pricing on website, better demonstration of Pro tier value → Audit ledger discovery: week-2 nudges, example queries in onboarding

Metrics we'd LIKE to track but can't (cleanly)

Long-term business impact for customers.

We want to know: did Avery NXR materially improve our customers' businesses? Revenue lift, cost savings, employee retention, customer satisfaction lift?

We can't track this cleanly because business outcomes have many causes and we're one input.

We try to measure via customer self-report, but the data is qualitative.

Cross-team adoption depth.

We can measure how many users a customer has, but harder to measure whether all departments are using agents or just one.

We're working on better signal here. Important because cross-team adoption is the path to durable customer relationships.

Competitor switching.

We hear about it in conversations but can't reliably measure. We don't know precisely how many of our customers come from competitor platforms vs. greenfield.

Metrics other agent platforms might use that we don't

Total executions.

Some platforms emphasize total agent executions as a metric. We don't think it correlates well with value created. An agent that fires 1,000 times but produces low-value outputs is worse than an agent that fires 100 times with high value.

Time spent in platform.

Some platforms measure time-in-app. We deliberately design for SHORT time-in-app — agents should work in the background. High time-in-app is anti-goal for us.

Feature adoption breadth.

Some platforms track how many features a user has tried. We don't think this matters. A user who's using 3 features deeply is more valuable than one who's clicked through all 12 features superficially.

What this transparency cost us

Sharing these metrics has trade-offs:

Competitors learn our state. They can see what we're optimizing for, what we're weak on. Mitigated by the fact that they'd learn similar things from customer conversations anyway.

Customers might worry about retention rate. "41% at month 12 sounds low." It is. We're improving it. Pretending it doesn't exist wouldn't help.

Skeptics use specific numbers against us. Some will. That's fine. The honesty matters more than the easy marketing.

We can't update marketing claims as easily. When you publish metrics, you have to maintain consistency. Easier to be vague.

We think the trade-offs are worth it. Transparency builds trust over time more than marketing fluff does.

What this means for buyers

If you're evaluating Avery NXR, you can ask us about these metrics in customer conversations. We'll share current state.

If you're evaluating other agent platforms, ask them the same metrics. Their willingness to share tells you something.

The honest answer "we don't track that yet" is better than vague answers that paper over real questions.

What this means for vendors

If you're another vendor in this space, the framework above might be useful for your own metrics.

We're not claiming our specific metric definitions are right for everyone. The framework — acquisition, activation/retention, business health — is generalizable. The specific metrics within each category depend on your platform's specifics.

The key insight: metrics shape behavior. Pick metrics that align with the outcomes you actually want to drive.

The principle

What you measure determines what you build. Pick metrics that align with the value you want to create for customers.

For us, that means: not vanity metrics (executions, time-in-app), but metrics that proxy for durable value (retention, customer-reported time saved, audit ledger usage).

The metrics we're proud of are growing. The metrics we're not proud of, we're working on.

In two years, we'll know more about which metrics actually predicted success. For now, this is our best guess.

→ avery.software — Free Desktop tier. The platform built around durable value metrics, not vanity ones.