The software numbers leaders should watch

The dashboard had forty charts. The meeting still started with "so how are we doing?" Nobody could answer from the screen. They had numbers. They did not have the ones that change a decision.
I would tell a founder to watch four things. Is the product up for the people who pay you. How long a real change takes to reach a customer. What the cloud costs per customer. Which risks are still open, and which of those have no name.
The rest can wait. A chart that looks busy is not a review.
Uptime for the path people use
The product was up, or it was not, for the person trying to use it. Pick the path they actually take. Login. The job they run. The page they wait on. "The site is up" while that path is down is a number that makes the room feel fine and leaves the customer stuck.
Write the window. Last month. Not a lifetime average that hides Tuesday afternoon. Say what broke, who was called, and what you changed so the same thing does not break the same way.
A name owns that number. If the name is "the team," you do not have uptime. You have a hope that someone was watching. The night that tests the name is who owns production after launch.
How long a change takes to ship
Not how many tickets closed. How long from "we decided" to "a customer has it."
Pick one kind of change and keep it. A bug fix. A small feature. The same kind, every month, so the number means something. If a small change takes a quarter, you do not have a speed problem you can fix with a pep talk. You have a calendar problem. The longer piece on what that time costs is the real cost of technical debt.
Count the change that reached a customer. A deploy that stayed on a branch is not shipped. A ticket moved to done is not shipped if the customer cannot see it yet.
Cloud cost per customer
Take the bill and divide it by the customers who pay, or by the accounts that use the product. Pick one definition and do not change it next month so the chart looks better.
The total bill going up can be fine if the customers went up with it. The total bill flat while customers left is not fine. One cell that says AWS does not tell you which. If you cannot split production, the extra environments, and the thing you will turn off, you do not have a cost per customer. You have a bill. How to put that bill on a sheet with a name is planning next year's software budget.
Open risks nobody owns
A risk with no name is a surprise you have already agreed to.
Keep the list short. The thing that takes the product down. The account only one person can open. The vendor you cannot replace this quarter. The data you are not allowed to lose. Each row has a person, or it is not a risk you are watching. It is a sentence in a slide.
"The team" is not a name. A backup sits next to the name. If both are out, the risk waits. It does not become whoever is free on Friday.
| Number | What good looks like | Not good enough |
|---|---|---|
| Uptime | The path customers use, last month, and what broke. | A lifetime average while that path was down on Tuesday. |
| Ship time | One kind of change, from the decision to a customer having it. | Tickets closed, or a branch that never shipped. |
| Cloud cost per customer | The bill, split, divided by the customers you picked and kept. | One cell that says AWS. |
| Open risks | A short list. A person, and a backup, on every row. | "The team." |
Why the pretty numbers mislead
Lines of code. Hours logged. Story points. Tickets closed. A green box that means the average. They move when nothing a customer feels has moved.
A team can close forty tickets and still take a quarter to ship the change that matters. A status page can be green while the one path customers use was down. A cloud bill can look "under control" because nobody divided it by the customers you still have.
If a number cannot change a decision this month, it does not belong in the review. Leave it on the big dashboard. Do not make people pretend it is the meeting.
A short monthly review
One sitting. Half an hour. The same four numbers, in the same order.
Uptime for the path customers use, and what broke. How long the last real change took to ship. Cloud cost per customer, and which line moved. Open risks, and which still have no name.
One decision leaves the room. Turn something off. Name an owner. Fix the path. Or say, out loud, that you are waiting. A review that ends with "good discussion" did not happen.
Write the four down before the meeting, on one page. If you need forty charts to explain them, you do not have the numbers yet. You have a tour.
Write the four, or bring someone who will
A founder who still ships, or the engineering lead, can run this. The page is short. The names are real. That is enough.
Bring someone in when the dashboard is busy and the meeting still starts with a question the charts do not answer. Uptime is a lifetime average. Ship time is a ticket count. The cloud line is one cell. The risk list has no names. That conversation is CTO advisory. Thirty minutes is enough to tell whether the four numbers are real.
Do not start a rewrite to avoid the review. A new platform does not tell you what broke last Tuesday, and it does not name who owns the risk.
Same idea, other angles
The numbers worth watching are the same week. The seat changes.
- App metrics that actually matter, on Maxiom Apps. Crash-free sessions, retention, store ratings, and how fast a release ships.
- How to measure an outside dev team, on Maxiom Dev. Shipped work, review turnaround, and bugs found late. Not hours, and not lines of code.
- Healthcare app metrics that matter, on Maxiom Labs. Portal adoption, uptime, access reviews done, and time to fix.
FAQ: the software numbers leaders should watch
Which software numbers should a founder watch?
Four. Uptime for the path customers actually use. How long a real change takes to reach a customer. Cloud cost per customer, with the definition held still. Open risks, each with a name. If a number cannot change a decision this month, leave it off the page.
What counts as uptime?
The path a paying customer takes, over the last month. Login, the job they run, the page they wait on. A lifetime average, or "the site is up" while that path is down, is not the number. Write what broke and who was called. That night is who owns production after launch.
How do you measure how long a change takes to ship?
From the decision to a customer having it. Pick one kind of change, a bug fix or a small feature, and use that kind every month. Tickets closed and branches that never shipped do not count. If that time is a quarter, read the real cost of technical debt.
What is cloud cost per customer?
The bill, split into lines a person can explain, divided by the customers who pay or the accounts that use the product. Pick one and keep it. A single cell that says AWS is a bill, not a cost per customer. The sheet that line belongs on is planning next year's software budget.
What belongs on the open-risk list?
A short list. The thing that takes the product down. The account only one person can open. The vendor you cannot replace this quarter. The data you are not allowed to lose. Each row has a person and a backup. "The team" is not a row you can review.
Which numbers should we stop putting in the review?
Lines of code. Hours logged. Story points. Ticket counts. A green average that hides the path customers use. They move when nothing a customer feels has moved. Keep them if a team needs them for their own work. Do not pretend they are the review.
What does the monthly review look like?
Half an hour. The same four numbers, written on one page before you sit down. One decision leaves the room: turn something off, name an owner, fix the path, or say out loud that you are waiting. "Good discussion" is not a decision.
Who should be in the room?
The person who owns each number, or the backup. A founder can run it. An engineering lead can run it. A room full of people who cannot say what broke, what shipped, what the bill did, or who owns the risk is the wrong room.
When should we bring someone in?
When the charts are busy and the meeting still starts with a question they do not answer. That is CTO advisory. If you can already point at the four numbers and the names, run the review yourselves.
What if the product is a mobile app, an outside team, or a clinical system?
Same week, different seat. A phone or tablet product is app metrics that actually matter. An outside team is how to measure an outside dev team. A clinical system is healthcare app metrics that matter.
If you want that conversation, see CTO advisory or talk to an engineer. Thirty minutes is enough to tell whether the four numbers are real.
I write from the seat of a working engineering company, not a tool vendor. More at antoniochagoury.com.



