There is a special kind of silence that shows up when someone asks how engineering is doing compared to everyone else. People look at the ceiling. Someone opens a laptop they already had open. A slide appears with a bar chart labeled "industry" in a font that suggests research happened. The numbers are crisp. The footnotes are missing.
I have been in that meeting as the person who brought the slide, and as the person who had to answer for it. The slide is rarely a conspiracy. It is a social solution to a hard question. The hard question is not "what shipped last month." The hard question is "are we behind?" Behind needs a front-runner. If you do not have one, the room will invent one and call it the market.
Last year I watched a leadership team debate headcount using a single blue bar. The bar said their deploy frequency sat below "industry." Nobody could name a company in that industry sample. Nobody could say whether "deploy" meant production, staging, or a feature flag flip. They still approved a reorganization on the strength of the bar, because the alternative was admitting they did not know.
That is not measurement. That is manners with a chart.
Comparison is the product people are buying
When a CFO or a board member or a forceful founder asks about delivery, they are not usually asking for a tutorial on lead time. They are asking whether the organization is embarrassed. Embarrassment needs a mirror. "Industry average" is a mirror you can buy without standing in front of your own systems long enough to get cold.
That is why vague peers are so popular. They convert an engineering problem into a weather report. Weather is not your fault. Weather does not require you to define failure before the next slide.
Vendors sell weather. Internal heroes sell weather. Sometimes you sell weather to yourself at 11pm the night before the review, because the alternative is to walk in with a local number that can be attacked.
Attackable is the point. A number that cannot be attacked cannot guide you. It can only calm the room for twenty minutes.
How to ruin a bad number without a committee
You do not need a research department. You need to ask out loud, in ordinary English, until the story falls apart.
- Q1Who is in this "industry"?If the answer is a gesture, stop.
- Q2How many teams or companies?If there is no count, you are listening to a mood.
- Q3What did they count?"Lead time" from first commit to production is not "lead time" from idea to revenue. "Failed change" that means rollback is not "failed change" that means any page that looked sad. Definitions are the whole game.
- Q4Which weeks?A freezes and launches and hiring waves world does not have a single climate.
- Q5Did people type this into a form, or did machines emit it?Surveys and pipelines both lie. They lie differently. Knowing which liar you hired matters.
- Q6Are you describing what happened, or claiming what caused it?"Deploys went up" is a description. "Deploys went up because we bought the platform" is a story. Stories need more than a logo between two charts.
If those questions make people angry, good. Anger is often what theater does when the lights come up.
What you say instead when it is your neck
"We are roughly at industry median on DORA, trending elite."
"Checkout services, last ninety days. Median first-commit-to-prod was five days across a couple hundred changes. One in ten took longer than twelve days, almost all of that was review queue. Failures that needed a hotfix inside a day were about one in twenty. I am not going to rank us against a peer group I cannot name. I am going to run a six week limit on open pull requests for that group. If the median does not drop by at least a day, or failures get worse, we undo it."
Same room. Same fear. One of those sentences can be checked after lunch. The other can only be believed.
Notice the live answer does not sound cosmopolitan. It sounds local. Local is what you actually manage.
What goes on the slide Monday
If you need five lines and nothing cuter:
Add the sentence: "What would falsify this?" If you cannot answer, you are decorating. Decorating is how you win this review and lose the next one.
Twenty-two people, no costume
We surveyed engineering leaders about measurement and improvement. Twenty-two people finished. About half (11 of 22) were CTOs, VPs, or directors. Ten industries showed up on the form. They chose to answer. That is the sample. That is all of the sample.
Do not measure delivery systematically
Only optimize with best effort
Believe better delivery could save 10%+ of cost
Would use a tool that helps them optimize on purpose
Self-selected sample · 22 respondents · 10 industries · 11 of 22 CTO/VP/director
I will not promote that into a map of the industry. I will say the smaller thing: among people willing to talk about this, incomplete measurement is common, and suspicion that the status quo is expensive is also common.
Use that as a prompt in your own org. Do not paste it into a board deck as "market data" unless you want to become the person who brought the costume.
Certainty is a management style
The reason this keeps happening is not stupidity. Looking certain is rewarded in the hour of the meeting. Looking careful is rewarded in the quarter when the roadmap still misses and someone asks why the industry chart did not save you.
I would rather take the careful hit early. "I can defend our last ninety days. I cannot defend a percentile I borrowed." That sentence has cost me comfort. It has never cost me as much as a confident fiction.
Your job, when you hold the engineering bag, is not to own the landscape. It is to own claims that still make sense when a skeptical person asks where the number came from.
If the answer is a slide with no source, the meeting is not about delivery. It is about manners.