CPO Days 31 to 60: Instrument the Org Before You Steer It

There is a version of month two where the new CPO starts making calls. The audits surfaced problems, the calendar is full, the org wants direction, and steering feels like the job. I have watched leaders take that road, and it works right up until someone asks "how do you know?" and the honest answer is "people told me." Month two is the fork: either you build instruments now, or you steer by anecdote forever, because there is never a quieter window to build them than this one.

The trap is subtle. The org already has dashboards. They will be offered to you. But they were built to tell the previous regime's story, and adopting them is adopting the story.

The short version

CPO days 31 to 60 convert the four audits from The CPO 30/60/90 into three permanent instruments: cost per outcome by workflow, eval scores with trend lines, and one direction metric. Then three moves turn the instruments into authority: a written coalition map, one visible evidence-backed decision, and a scoreboard renegotiation with the CEO around day 50. The discipline is subtraction. Three instruments leadership reads weekly beat thirty nobody opens. The contradictions you logged in the listening ledger are the first things the instruments should settle.

Instrument one: cost per outcome by workflow

The question this instrument answers: what does one unit of customer value cost us to produce? One resolved ticket. One generated document. One completed workflow run. Not the cloud bill, which is an aggregate that hides everything.

Who to ask. Start with whoever owns the cloud bill, usually a finance partner or an infra lead. Then the engineer who actually tags resources, who is rarely the person finance points you to. You need both: finance has the dollars, the engineer has the attribution.

The 30-day path. Week one: get the raw bill and a list of the top five workflows by volume. Week two: accept crude attribution. Tag by service and team, allocate shared costs proportionally, and write the allocation rule down so it is consistent rather than precise. Week three: divide by outcome counts, which product analytics already has. Week four: publish the five numbers on one page with a date.

What resistance sounds like. "The attribution is too messy to be meaningful." Translation: nobody wants to own a number that might look bad. The answer is to publish the number with its error bars and the allocation rule attached. A number that is wrong by 30 percent but directionally honest beats no number, because it can be argued with and improved. Silence cannot. The full method, including what to do when one workflow turns out to be underwater, is in Gross Margin Is Your Job Now.

Instrument two: eval scores with trend lines

The quality audit told you which AI features have eval scores and which have silence. Month two makes the scores permanent and, critically, adds the trend line. A score is a photograph. A trend is a film. Quality drift only shows up in the film.

Who to ask. The tech lead of each AI feature, directly, not through their manager. The managers will promise a rollup. You want the raw artifact: the eval set, when it last ran, what the score was the run before.

The 30-day path. Week one: inventory, which you mostly have from the audit. Week two: for features with existing evals, get them running on a schedule, weekly at minimum, and logging to one place. Week three: for features with no evals, do not wait for perfect ones. Twenty representative cases graded pass or fail is an eval. Week four: one page, every production AI feature, score and 4-week trend, owner named.

What resistance sounds like. "Our feature does not really lend itself to evals." It does. Every output a customer sees can be graded, even if the first grader is a human with a rubric. The team saying this is usually the team whose score would be lowest, which is exactly the information you need. The argument for treating the eval as the actual spec is in The Eval Is the Spec.

Instrument three: one direction metric

Cost tells you what the product burns. Evals tell you whether it works. Neither tells you whether it is getting better for customers in the way your strategy claims it should. That is the direction metric: one number, chosen by you, that moves only if the product is improving on its actual promise.

Who to ask. Nobody, at first. This one is a choice, not an excavation. Draft it yourself from the audit evidence, then pressure-test it with your most evidence-driven peer before showing the CEO.

The 30-day path. Week one: draft three candidates. Week two: kill two. The survivor must pass the test in Direction Metrics: can it go up while the business gets worse? If yes, it is a vanity metric wearing a strategy costume. Week three: get it computed weekly, even by hand. Week four: put it at the top of the instrument page and start every leadership conversation with it.

What resistance sounds like. "We already have a north star metric." Check whether the existing one is a growth metric in disguise. Most are. Bigger is not the same as better, and conflating them is how products bloat.

The coalition map, as a worksheet

By day 45 the interviews have told you who runs on evidence, who runs on narrative, who is exhausted, and who has waited years for someone to fix their thing. Write it down. An actual document, four columns.

Three rules. First, camp assignments are about your first bets specifically, not about whether people like you. A skeptical CFO who runs on evidence is a stronger future ally than an enthusiastic peer who runs on vibes. Second, the "what they need" column must be concrete enough to act on this month. Third, revisit the map after every visible decision, because the camps move. The long version of this exercise is in the CPO coalition map essay.

Ship one visible decision

Not a quick win. Quick wins are small enough to ignore and cosmetic enough to resent. One decision, three criteria: evidence-backed (it traces to the audits or the ledger), reversible (you can walk it back if the evidence was wrong, see the cost of being wrong for why this matters more for early calls), and watched (people will notice both the call and how it was made).

Three worked shapes I have seen land:

Kill the zombie. An initiative everyone privately calls dead but nobody will bury, usually because its sponsor left and killing it feels like blaming a ghost. You kill it with a short written explanation citing the evidence. Cost: low. Signal: decisions here now get made on evidence and get written down.

Publish the scores. Put the eval page where the whole org can see it, including the embarrassing rows. Cost: two uncomfortable weeks for two teams. Signal: quality is now measured, not asserted, and nobody gets ambushed because the trend was visible to everyone at once.

Replace the status meeting. Reporting moves to a pre-read, live time goes to decisions, using something like the five-question product review. Cost: a few weeks of awkward silence while people learn to read. Signal: leadership time is for judgment, not narration.

Pick one. Shipping two dilutes the precedent and triples the political surface.

The scoreboard conversation

Somewhere around day 50, before the CEO's patience for "still learning" expires unannounced, you open the scoreboard renegotiation. You bring a one-page draft of what you want to be measured on for the next year, anchored in what the audits found. The opening script, roughly as I would say it:

"I want to propose what you hold me accountable for over the next twelve months, and I want to ground it in what I found rather than what I hoped to find. Three things surprised me in the audits. Here is what they imply about what this product org can credibly commit to, and here is the scoreboard I think follows from that. Push back on any of it, but I would rather we argue about the scoreboard now than discover at day 200 that we were keeping different ones."

The structure matters: findings first, scoreboard second, invitation to argue third. You are not asking for easier targets. You are replacing inherited targets, which were set without the information you now have, with informed ones. Most CEOs take this conversation well at day 50 with evidence and badly at day 100 with excuses.

Failure modes

Instrumenting everything. Fifteen dashboards is a museum, not a cockpit. Every gauge must pass one test: does this number change a decision? Three instruments read weekly beat thirty opened never.

The dashboard nobody reads. An instrument that is not embedded in a ritual is decoration. Each of the three gets a standing slot: cost page opens the monthly business review, eval page opens the product review, direction metric opens your CEO one-on-one. No slot, no instrument.

Mistaking visibility for control. The instruments tell you where the org is. They do not move it. I have watched leaders stand up beautiful gauges and then wait, as if measurement were management. The gauges buy you the right to make the visible decision. The decision is still yours to make.

Pick one thing this week

Ask one tech lead, directly, for the eval score and 4-week trend on their feature. Not the manager, the tech lead. The answer, the delay, or the silence is your instrumentation backlog in miniature, and it costs one Slack message to collect.

Month one of this arc is the listening ledger. Month three, where the instruments get spent on kills and bets, is CPO days 61 to 90(coming Jun 24). The worksheets are in the CPO First-90 Kit.

Sources: Michael Watkins, The First 90 Days on securing early wins and alliances, Marty Cagan / SVPG on product leadership and operating cadence, Anthropic on evaluations for standing up evals fast.

Further reading

• The CPO 30/60/90
• Strategy From Signals
• Direction Metrics