AI
AI’s Saved Hours Don’t Show Up as Savings
The hours AI gives back are real. They just get spent on the work people used to put off, which is worth a lot, but it isn’t a cost line, and it shouldn’t be scored like one.
“I’m not sure what to say, but we aren’t seeing efficiency, you know?”
That’s me, partway through writing an update on how our customer-facing teams were using AI. The tools were clearly saving people several hours a week each. I stopped mid-draft, because the next line was supposed to turn those hours into money, and I couldn’t make it true.
Every AI pitch deck has the same slide: hours saved, times a loaded rate, equals dollars back. It’s a clean number, and I don’t trust it.
The hours are real. They just never show up on the P&L, because nobody goes home early. All that recovered time is indirect. People are working as hard as ever, or doing even more, and the extra goes into tasks they would have normally put off or never done. The support lead finally writes the audit-prep guide she’s put off for a year. The PM answers a feature request with a real write-up instead of two lines. The doc that would’ve been two bullets turns into a real doc, and the nice-to-have backlog finally gets touched.
That’s value, and I’d argue it’s a lot of value. It just isn’t savings.
We’ve seen this before. In 1987 Robert Solow wrote that you could see the computer age everywhere but in the productivity statistics. The numbers took about a decade to catch up, because companies spent that decade on the slow, invisible part: retraining people and reorganizing the work so the technology could actually pay off.
It looks to me like where AI in software development was a year ago, and support and customer success are in the middle of it now. If you score AI on headcount, you’ll conclude it failed right while it’s working. Score the backlog that finally moves and the quality floor that went up instead.
I don’t have the whole answer. Maybe this time the numbers never catch up, and I can’t rule that out. What I did in that update was skip the savings claim entirely. I put what got done that otherwise wouldn’t have right next to what the tools cost, and let people judge the trade for themselves.
So take hours-saved-times-rate off the slide and ask the harder question: what got done that wouldn’t have been done at all?