tgroenwals shared this post ยท May 14
Andreas Horn

๐—จ๐—ป๐—ฝ๐—ผ๐—ฝ๐˜‚๐—น๐—ฎ๐—ฟ ๐—ผ๐—ฝ๐—ถ๐—ป๐—ถ๐—ผ๐—ป: ๐— ๐—ผ๐˜€๐˜ ๐—ฐ๐—ผ๐—บ๐—ฝ๐—ฎ๐—ป๐—ถ๐—ฒ๐˜€ ๐—ฐ๐—ผ๐—ป๐—ณ๐˜‚๐˜€๐—ฒ ๐—”๐—œ ๐—ฎ๐—ฑ๐—ผ๐—ฝ๐˜๐—ถ๐—ผ๐—ป ๐˜„๐—ถ๐˜๐—ต ๐—”๐—œ ๐—ถ๐—บ๐—ฝ๐—ฎ๐—ฐ๐˜.

They measure the easy stuff.
They just track usage.

โ†’ How many employees logged in.
โ†’ How many prompts were written.
โ†’ How many tokens were consumed.
โ†’ How many Copilot licenses were activated.

That is not impact.
That is activity.

The real question is different:

โ†’ Did AI increase revenue?
โ†’ Did it shorten time to value?
โ†’ Did it reduce manual handoffs?
โ†’ Did it free up workforce capacity?
โ†’ Did it improve customer satisfaction?

Most companies do not have an AI adoption problem. They have an AI accounting problem. They cannot clearly show where value is created, who owns it, how it compounds, and when it becomes measurable.

You do not scale AI because people are excited. And you do not scale AI because people โ€œuseโ€ it. Adoption matters (still), of course. But adoption is only the starting point. The much more important questions is whether AI creates value that is measurable, repeatable, and attributable.

Clare Kitching nicely summarized below a practical set of KPIs to start measuring AI impact more seriously.

โ†“

๐—œ๐—ณ ๐˜๐—ต๐—ถ๐˜€ ๐˜„๐—ฎ๐˜€ ๐˜‚๐˜€๐—ฒ๐—ณ๐˜‚๐—น, ๐˜†๐—ผ๐˜‚ ๐—บ๐—ฎ๐˜† ๐—ฎ๐—น๐˜€๐—ผ ๐—ฒ๐—ป๐—ท๐—ผ๐˜† ๐—บ๐˜† ๐—ณ๐—ฟ๐—ฒ๐—ฒ ๐—ป๐—ฒ๐˜„๐˜€๐—น๐—ฒ๐˜๐˜๐—ฒ๐—ฟ: https://lnkd.in/dbf74Y9E

43
Gleaming Systems Completely agree. Usage metrics can show interest, but they rarely prove business value. Real AI impact shows up in cycle times, customer outcomes, operational efficiency, and revenue movement. May 14
Gagik Kyurkchyan Great framing, Andreas โ€” activity vs impact is the right cut. The part I keep snagging on is hidden inside the word "accounting": most of those KPIs are subtractions against a baseline. "Post-AI revenue minus baseline revenue" assumes the baseline would've held still, but nothing else in the business froze while AI rolled out. Even measuring it for my own dev workflow, I can't fully separate "AI made me faster" from "I also got better at the problem." It's less an accounting problem than an attribution one โ€” and attribution needs a counterfactual you never get to run. May 14