When Your North Star Isn't Yours

You're the PM for a feature, not the whole product. Maybe it's a new checkout flow, a recommendation widget, or a conversational assistant that sits inside a much larger app. The business has a north star—revenue, orders, whatever—and it's tempting to adopt it. More of that thing is better; everybody agrees. So why not make it yours?

Here's the problem: if that number goes up and nobody used your thing, you didn't win. You just got credit for something you didn't do.

I was running a case recently where we had to define success for a cart assistant—conversational AI for grocery shopping. The candidate landed on "orders completed" as the north star. I asked: what if you ship this, nobody uses the assistant, but orders go up 20% anyway? Would you call that success? They said no. Exactly. So we had to find a north star that actually reflected the contribution of the feature.

---

The core tension

When you own a piece of something bigger, you have a marginal impact. The overall number might be huge and noisy. Your slice might be too small to move it in a measurable way, or it might move for a dozen other reasons—seasonality, another team's launch, a change in acquisition. If your north star is the same as the whole product's, you're either claiming credit you don't deserve or freaking out when the number dips for reasons that have nothing to do with you.

So you have to ask: is it appropriate for us to hold the north star of the whole thing when we're only responsible for a part of it—and when our part isn't independently significant or measurable?

---

Four ways to answer

1. Total orders (or whatever the business north star is)
Simple, ultimately essential, but wrong for you. You are responsible for this specific thing, not the whole thing. It's important that you drive the whole thing forward, but your job is about moving your thing forward. In most cases, there is a theory about why your thing drives the whole thing, so having your thing move forward is usually good enough even if the larger thing doesn't (more on that later).

2. Engagement with your feature—conversations had, sessions, etc.
Now you're measuring your thing. The risk is vanity. If the only goal is more orders, then counting conversations is only useful if those conversations lead to orders. If people are chatting about their dog and burning tokens, you've built a metric that looks good and means nothing. You can very easily create a gameable metric, even without realizing it: you could put in features that drive conversations but not orders. So "conversations" can be a vanity metric: engagement with no point. Remember: what you measure affects what you choose to build! That's why it's so important to choose a good NSM.

3. Percentage of orders that came through your feature
My client came away with this as a refinement after I challenge their initial choice. This is better! You're tying success to both (a) the business outcome and (b) usage of your feature. One without the other doesn't work—you don't want a north star that goes up when nobody uses the thing, and you don't want a north star that's "usage" when usage doesn't connect to value. So combining them makes sense. But percentages have a downside: if your share of orders goes up and total orders go down, you still lose. So it's an improvement, but not the cleanest.

4. Orders attributable to the tool
Here you're not saying "we care about total orders" or "we care about conversations." You're saying: we care about orders that happened because of this feature. You define attribution—e.g., an order placed through the assistant, or within a session where the assistant was used—and you count those. More is better. You can still look at total orders as context, but your north star is attributed orders. Your pipeline is a parallel thing to the overall order pipeline. That's how you own a piece without pretending you own the whole.

---

Vanity metrics

People often talk about vanity metrics without completely knowing what that means. This case gives a perfect demonstration. "Conversations had" is a vanity metric because it demonstrates that your app is being used despite the fact that it isn't driving any business value whatsoever.

One way to confirm whether something is a vanity metric is to ask: would a customer or a partner pay for the feature based on this number? If you're selling this to an enterprise and you say "people had a million conversations with our assistant," they'll ask: so what? Did they buy more? If the answer is "no, but engagement is up," you're in vanity territory. If you wouldn't use it to sell the feature, be suspicious of using it as your north star.

That doesn't mean "conversations" or "engagement" are always wrong. There can be legitimate reasons to want usage for its own sake—retention, perception, strategic bets. But then own that. Don't call it a north star for "orders" when it's really a bet that usage is valuable. And watch for gameability: any metric that goes up when people use the thing without doing the valuable thing is dangerous.

---

Why this matters for what you build

The cool thing about this case was that the product didn't exist yet. So the choice of north star wasn't just a dashboard question—it shaped what you'd build. If your north star is "conversations," you optimize for getting people to talk. If it's "orders attributable to the assistant," you optimize for the path from conversation to order. Different north stars, different product decisions. Get the metric wrong and you build the wrong thing.

So when you're the PM of a self-contained flow with its own kind of conversion, your primary job is to make that conversion happen—with the belief that it has a proven relationship to the downstream business outcome. Your north star should reflect that: the outcome you can actually influence and attribute. Not the whole business. Not raw engagement. The thing that says: we did our part, and here's the evidence.