Agency Routing

Your Metrics Are Making Decisions Without You

In mid-2023 I was running Flow at Pluralsight. We had a crib sheet for the second half of the year with your standard set of financial objectives. Book more ARR than we did in H1, improve renewal rate substantially. And a specific strategic objective I want to talk about: maximize the number of customers whose teams use Flow weekly.

Weekly usage went up. We tracked it. We celebrated it. The number moved and the slide deck looked great.

But using Flow weekly did not mean it was changing anything for our customers. The product (now owned by a different company) tracked DevOps metrics for engineering teams: cycle time, deployment frequency, etc. The problem was that for most actually improving those metrics required serious process reengineering inside the customer’s org. And we were not measuring that. We were measuring logins.

Customers were using the product. They were not all getting breakthroughs. The metric was green. But their outcome was flat.

So we did the thing the metric could not measure. We went to our most important customers and helped them process-reengineer sat with their teams, worked through their workflows, did the hard unglamorous consulting work that actually moved the DevOps numbers. For a tiny handful, it worked. But it worked despite the metric, not because of it. The dashboard never told us to do that. We had to override it.

That is when I started formulating something I now term agency routing: it’s how decision-making power migrates from the people you trust to the numbers you track.

People will call this Goodhart’s Law. It is related but it is not the same thing. Goodhart tells you the metric will get corrupted: when a measure becomes a target, it ceases to be a good measure. That is about measurement itself. Sociologists like Merton call a version of this goal displacement: when the means become the ends. It’s much closer, but still too clean.

Agency routing is about power.

It is not just that the number gets gamed, or that the means become the ends. It is that the number starts making decisions nobody specifically authorized. Product roadmap, engineering priorities, customer success playbooks, hiring plans all reorganize around the metrics, and nobody really decides that should happen. It looks like good execution. And that is what makes it hard to see.

Here is how it happens

You scale. You add dashboards because you cannot sit in every room anymore. At first the dashboards are a gift. They answer things you used to learn in hallways.

Then the dashboards start becoming a steering wheel.

A VP says we need to improve activation by five percent this quarter. That becomes an OKR. The OKR becomes the thing teams are judged on. Judgment becomes compensation. Compensation predicts behavior.

The org stops asking what would help customers most? and starts asking what would move activation this quarter?

That second question is not evil. It is often reasonable. But once a metric becomes the language of success, it becomes the language of power. And once it is the language of power, it becomes the routing layer for agency.

Your team does not need to disagree with you to diverge from you. They just need to pursue the most rewarded local interpretation of their org’s scoreboard.

Your metric as the true decision-maker

Delegation is compression. You cannot give someone the full, rich intent behind a goal — make the product feel inevitable to a new user, without confusing or manipulating them — so you compress it into a proxy. Increase activation. Reduce time-to-value. Improve NPS.

Proxies are necessary. You cannot coordinate forty or four hundred people around a paragraph of nuance.

But this compression discards information. The information discarded is almost always the part you care about most: the tradeoffs, the constraints, the edge cases, the subtle reasons you chose one path over another.

When you remove that information, you create a vacuum. Incentives fill vacuums. This is not a moral story about bad employees. It is a system story about what you reward.

In a previous meditation I called this intent drift: how the meaning of a change gets lost between the person who conceived it and the code in the product that ships. This is the same thing at the org level. Intent gets compressed into a metric. The metric gets optimized. The optimization drifts from the intent. Few really notice when the number turns green.

Why scaling is hard in a way that never shows up in hiring plans

At small scale, the drift is visible. You see the work. You hear the reactions in real time. A misaligned win gets corrected socially before it becomes standard.

At larger scale, those corrections do not propagate. The local team sees the metric moved. The slide deck looks good. The quarterly review goes well. The work gets reinforced. The pattern repeats.

It can run for months because it feels like progress. Motion creates morale. Morale creates confidence. Confidence creates narrative. Narrative creates momentum. By the time someone realizes the company has drifted, the drift has been normalized into process. You have a KPI. You have weekly check-ins. You have headcount. You have career paths built around that number continuing to matter.

At that point, even if you want to change course, you are not just changing a target. You are trying to reroute agency away from a well-funded, well-defended decision-making channel.

You learn to recognize it by what people say.

We hit the number, but customers still complain.

Or: We improved retention, but product quality is worse.

Or: We reduced support tickets, but churn went up.

These are not paradoxes. They mean the org has learned to satisfy the metric without satisfying the intent. It learned which levers move the proxy fast. It also learned, without anyone deciding to, which costs get externalized and therefore never punished.

Leaders often respond by adding more, complementary metrics. A quality metric to balance the growth metric. A churn metric to balance the activation metric. A satisfaction metric to balance the efficiency metric. The dashboard grows. The meetings multiply. The organization becomes a bureaucracy of counter-balanced metrics.

It helps at the margins. But it does not fix the core problem. The problem was never the absence of measurement. It was the absence of a definition of good that is richer than numbers.

I saw this during my time at Dropbox (a few companies before my Pluralsight stint). Dropbox Paper was a fantastic product. The team was talented and the work was genuinely good. The metric was MAU grow monthly active users. And it worked. Paper usage grew.

But Paper was a proprietary format. The .paper files were not really files they were links to a web-based experience. You could not sync them the way you synced everything else in Dropbox. You could not open them outside of Dropbox. In a company whose entire value proposition was your files, everywhere, portable, yours, the org was pouring resources into a product that structurally contradicted the core business.

The MAU number was green. The strategic fit was broken. Nobody that I saw had the metric that said stop, this is pulling us away from what we are.

The agency had already routed.

If it’s in your control, why do you do it? If it’s in someone else’s, then who are you blaming? Atoms? The gods? Stupid either way. Blame no one. Set people straight, if you can. If not, just repair the damage. — VIII. 17

Instruments, not goals

I do not think the fix is ignoring metrics. You and I could agree that is childish. Metrics are necessary at scale.

The fix is to stop pretending metrics are goals and start treating them as instruments. An instrument helps you see. It should not be allowed to steer by itself.

What I have learned (mostly by getting it wrong) is that when you delegate, you are not only assigning responsibility. You are authorizing a set of tradeoffs. If you do not specify which tradeoffs are unacceptable, the team will make them anyway, and the metric will choose for them.

So you give the team a target and a set of constraints that make the target safe:

Increase activation by X. New users should reach a meaningful outcome quickly, not just click through steps. Do not use deceptive UX. Do not increase support burden downstream. Do not sacrifice long-term retention for short-term activation. We will check with qualitative reviews, cohort retention at 30/60/90, and support ticket tagging for post-onboarding confusion.

The team still gets freedom to be creative. The proxy does not get to be the boss.

Who gets to say no

The part I have not solved, and I do not think most teams have either is that constraints need enforcement. Someone has to be empowered to reject a metric win that violates the intent. Most orgs have execs and dashboards. Very few have someone with the mandate and the incentive to say: the number went up but we did it wrong, and we are reverting.

Without that role, metrics become self-driving. Nobody has a mandate to slow them down.

Alignment is not free

Executives want a scalable way to stay aligned without being involved in every single decision. They want to say here is the goal, go execute and trust that execution will reflect the company’s values and intent.

This works when the company is small enough that cultural feedback loops are fast.

At scale, you do not get alignment for free. You get it by building a control plane. Metrics are part of that control plane but they are not sufficient. People stay aligned when they know, in advance, which wins will be celebrated and which wins will be rejected: even if they move the number.

If you never reject a metric win, you are teaching the company that the metric is the truth. Soon it will be.

Scaling does not fail because people stop working hard. It fails because decisions get distributed and incentives become the routing layer.

The loop between intent and reality is the job. Not the metric. Not the dashboard. The loop.

If you do not own it deliberately, the scoreboard will own it for you. And you will wake up one day, look at the number, and see green.

Both of those can be true at the same time.