Does Productivity Make Money?

The cloud era taught us that activity is not the same as value. AI is about to teach us the same lesson, unless we measure the right things in production.

Ask a leadership team whether their engineering and support functions are productive and you will get a confident yes, usually backed by a dashboard. Commits are up. Tickets are closing faster. The model is shipping more code than any human team could. All true, and none of it answers the only question that matters to the business: did any of this bring money in, or take cost out?

Productivity is not a feeling, and it is not a volume. It is a ratio of value created to spend consumed. The moment you stop measuring the value side, you are not measuring productivity at all. You are measuring effort and calling it progress.

We have made this mistake before

In the early 2010s, engineering productivity was measured by output. Lines of code, then commits, then story points. It was easy to count and it felt rigorous. It was also wrong, for a reason that became obvious in hindsight: code committed does not equal revenue earned. A team could be the most prolific in the company and ship nothing a customer would pay for.

So the industry corrected, and around the middle of the decade it moved to a better question: is the feature live? Shipping to production became the new scoreboard. The hyperscalers set the tone here. The public measure of a cloud platform's momentum became the number of services and features launched, counted off on stage every year, rather than the return of any single one of them produced.

That correction was real, but it stopped one step short. By 2016 the pattern had a name, the "feature factory", and the critique was simple: shipping a feature is still output. A feature live in production is not an outcome. It is a bet that has not yet been measured. Counting launches is just counting commits at a higher altitude.

The honest lesson of the cloud era is therefore not "measure what you ship". It is that every metric one rung below money is a proxy, and proxies drift. Activity drifts from value the moment no one is checking the exchange rate between them.

AI is the same trap with a larger bill

AI productivity is being measured the way engineering productivity was measured fifteen years ago. Tokens consumed. Lines of generated code accepted. Tickets auto-resolved. Agents deployed. These are the commits of this cycle, and they carry the same flaw with one important difference: the spend behind them is large, metered, and visible on a bill.

This is why the value of AI is decided in production, not in development. A coding assistant that accelerates a prototype no customer reaches has produced activity and consumed spend. The same assistant attached to a workflow that shortens cost to serve has produced value you can attribute. The technology is identical. The difference is entirely in whether the work reached a customer and changed an outcome, and whether you measured it there.

Treated as cost, technology spend is something to minimise. Treated as capital, it is something to hold accountable for a return. AI in production is where that return is either earned or quietly lost, and most teams are not looking.

What to measure, and how to read it

The fix is not a longer dashboard. It is a discipline: never let an activity signal stand on its own, and attach every unit of work to the spend it consumes and the value it produces. Sort your metrics into two groups and treat them differently.

The top three are activity signals. They are necessary and they are not the answer. A faster time to resolution that doubles your inference cost per case has made you slower in the currency that counts. The bottom four are value metrics, and they share one property: they are ratios with spend in the denominator. That is the part most teams cannot produce today, because their AI and cloud spend is not attributed to the customer, product, or workflow that consumed it.

Choosing what to measure

Four principles keep a productivity programme honest.

1. Anchor on the unit, not the total. A company-wide cost number tells you nothing about which customer or product is profitable. ROI per customer and ROI per product do.

2. Pair every activity signal with a value metric. If you report cases closed, report cost to serve alongside it. The pairing is what stops a proxy from drifting.

3. Attribute the spend, or the ratio is fiction. ROI per product is only real if you can trace the AI and cloud spend that product consumed. Attribution is the mechanism, not an accounting afterthought.

4. Measure in production, where the money is. Development tells you what is possible. Production tells you what was worth it.

The point

The cloud era proved that you can be extraordinarily busy and create very little, and that the metrics meant to catch this will happily report success the whole way. AI raises the stakes because the activity is cheaper to produce and the spend is easier to grow. The teams that win this cycle will not be the ones generating the most output. They will be the ones who can attribute spend to outcomes and answer the original question with a number: did productivity make money?

That is the question OPTIMAZE exists to answer.