How to Go from an AI Product Strategy to Quarterly OKRs

​Dear Reader,​

Useful OKR sets stand on the shoulders of high-quality inputs like Product Strategy. That doesn't change when there's a model behind the product. What changes is how tempting it becomes to measure the model instead of the customer.

Turn your Product Strategy Choices into Statements

Summarize the choices you make in concise statements. Incorporate the most significant bits about your strategic narrative, defined playing field, and winning moves per choice. Depending on the level, you need different levels of granularity. Make sure they're real choices, not business-as-usual dressed up. "We will use AI" is not a choice. Everyone in your category wrote that sentence this year.

A B2B example:

​Grow from 60% to 85% of invoice line items processed without human review by 2028, by serving mid-market finance teams who still treat accounts payable as a headcount problem, through models trained on each customer's own approval history instead of a generic document parser.

Derive a lagging Proxy Metric

Ask yourself: How would we know this choice was successful, in the form of a metric? Think beyond the "usual" revenue or activity metrics and go broader. How can you take on different perspectives on your choice?

Watch out for the AI version of the Metrics Buffet: model accuracy, eval pass rates, tokens served, share of users who touched an AI feature. Those are health metrics for AI usage. They don't represent changes in customer behavior.

For the statement above, the lagging proxy sits closer to review hours removed per 1,000 invoices, per account. By design, that's an annual-ish company- or portfolio-level metric. It's also uncomfortable, because you can't hit it by shipping a model.

Identify the leading indicators driving the lagging proxymetric.

Using tactics like reverse-engineering the customer journey from the lagging events you just identified, or KPI trees, identify more leading indicators. The same two criteria determine whether they become useful OKRs: How autonomously can your team influence them, and how quickly can you detect changes?

With AI in the product, the indicators worth having usually measure earned trust rather than usage:

• Share of suggestions accepted without an edit, by week of an account's tenure
• • Share of approvers who stopped opening the line-item detail view at al

Of course, this is overly simplified and leaves out all the messy Discovery, discussions, and iterations. But at the core, that's the path you want to follow, even though you might have to wander into the forests from time to time.
*Mind you, I said simple, not eas

Thank you for Practicing Product,

​Tim​

Who is Tim Herbig?

As a Product Management Coach, I guide Product Teams to measure the real progress of their evidence-informed decisions.

I focus on better practices to connect the dots of Product Strategy, Product OKRs, and Product Discovery.