How agencies design their KPIs
No one sat down and decided to design a KPIs framework that would obscure performance. That’s not how these things happen.
Agency KPIs emerged the way most professional standards do: organically, over time, shaped by what was possible to measure, what was possible to influence, and what was possible to demonstrate to a client in a quarterly review. The result is a framework that centers variables where human expertise shows and attributes clearly. Reach. Frequency. CPM efficiency. Campaign setup quality. Reporting turnaround. Strategic recommendations delivered.
These are real inputs. They reflect genuine work. The problem isn’t that they’re dishonest, it’s that they’re complete within a specific model of how advertising operates. A model where decisions move at human speed, where the review cycle is the unit of work, where humans create value in the moments they can document and present.
That model has a structural consequence: decision latency never becomes a variable. The time between what the data shows and what the campaign does becomes normal operating procedure. It doesn’t have a column because in this model it isn’t a cost. It’s just the rhythm of the work.
So nobody measures it. And what nobody measures, nobody questions.
What those KPIs make invisible
Run the standard agency report through this lens and the pattern becomes clear.
CPM efficiency measures how well the team purchased inventory. Not what happened in the hours between when it started underperforming and when someone acted. Campaign setup quality measures how well the team built the architecture at launch. Not how many times it needed to change mid-flight and how long each change took. Strategic recommendations delivered measures the volume and quality of advice. Not the distance between recommendation and implementation.
The framework has precise instruments for the moments of visible human work. It has no instruments for the intervals between them. Not because the intervals don’t matter, because in the model the framework assumed, the intervals were operationally empty by design. Nobody treated the weekly review cycle as a limitation to measure. It was the cadence. Nobody tracked decision latency as a cost. It was the time the process required.
Build your measurement framework inside that assumption and you get an accurate picture of how well the process runs. You get no picture of what the process costs.
When the client adopts the agency’s framework.
This is where the structural problem compounds.
At some point in most client-agency relationships, the client begins evaluating their advertising investment using the same framework the agency uses to report on it. The KPIs migrate. The benchmarks become shared. The quarterly review structure becomes the client’s own rhythm of evaluation.
Once that happens, the client is measuring their investment with an instrument calibrated for a different purpose. Not because the agency imposed it, but because it was the available framework, it was coherent, and it was what both sides knew how to discuss.
The consequence is invisible but significant. When a client runs a pilot of an autonomous system alongside their existing model, they evaluate it with the tools they have. The autonomous system’s advantage the decisions made at 3am, the budget recovered between platform reports, the compounding effect of continuous optimization none of that has a column.
The pilot looks incremental. The performance gap looks smaller than it is. The spreadsheet says the difference is manageable.
It isn’t the agency defending the framework. It’s the framework being the only instrument in the room and instruments, by definition, can only show what they were built to see.
The question nobody designed the framework to answer.
There is no agency KPIs for what happens between the last review and the next one.
That absence is not an oversight. In the agency model, the intervals between reviews are administrative space. Reports get compiled. Data gets cleaned. The next presentation gets prepared. Value is created at the moments of visible work the recommendation, the optimization request, the strategic call. Decision latency is absorbed into the overhead of the process, not measured against it.
In an autonomous system, those same intervals are where the decisions happen.
The campaign that started losing performance at 3am handled before the team arrives on Monday. The inventory pricing shift that opened a reallocation opportunity for forty minutes on a Thursday afternoon, captured before any review cycle could schedule it. The compounding effect of thousands of micro-corrections, each operating in the space between human checkpoints, none of it waiting for a meeting, an approval, a report.
In an agency model, the intervals are empty by structural necessity. In an autonomous system, the intervals are the process.
A framework built around visible human work has no column for that. The metrics that protect the agency model don’t do so by design. They do so by omission. And the client who adopts them as their own is evaluating their investment with a ruler that was built to measure something else.
Mainkore operates in those intervals. 200+ variables. 20 milliseconds. 24/7. Guaranteed by contract.
