Your Single Source of Truth Is a Person, and She Just Took PTO

Somewhere in your company there is one person who can answer questions about the numbers. Not the CMO. Not the person whose title says analytics. One specific human, usually mid-level, usually underpaid, who knows which report is right and which one is haunted.

Everyone knows who it is. When the CEO asks how paid search did last quarter, the question does not go into a tool. It goes into Slack, addressed to her. And she answers, because she always answers, and the number is correct, because it always is.

That is not a reporting stack. That is a person doing an impression of a reporting stack.

The org chart says analyst. The architecture says load-bearing wall.

Nobody designed this. It happened the way most operational disasters happen, which is gradually and then all at once. Somebody bought a marketing automation platform. Somebody else bought a CRM. Somebody in ecommerce set up Shopify, somebody in finance set up Stripe, and web analytics got reinstalled twice by two different agencies who both swore the previous setup was broken.

None of these systems agree with each other. They were never going to. They count different things, on different clocks, with different definitions of what a customer is.

So a human stepped in to do the reconciliation. She learned that the CRM double counts anything that came through the partner form. She learned that revenue in the ad platform is self reported and should be ignored on principle. She learned that a data migration a couple years back broke everything before a certain date, so any trend line that crosses it is a lie. She learned that one campaign got tagged wrong for six weeks and has to be manually pulled out of every channel report forever.

None of that is written down. It is not written down because writing it down was never anybody’s quarterly goal, and because half of it she does not consciously know she knows. It is muscle memory now. She just does the adjustment on the way to the answer.

You find out three ways, and none of them are pleasant

She takes a week off. Somebody else tries to pull the number. They pull it straight out of the tool, because why wouldn’t you, and it is wrong in a way nobody catches until it is already in a board deck. The correction happens quietly, later, with a lot of throat clearing.

She is in back to back meetings. The question sits in Slack for six hours. A decision gets made without it. This is the most common version and the most expensive, and it never shows up in any incident log because nothing technically broke.

She leaves. This is the one that actually hurts. She gives two weeks. Somebody schedules a knowledge transfer session. She sits in a room and tries to explain, out loud, a set of exceptions she has been carrying for three years, to a person who does not yet know enough to ask the right questions. It goes badly. It always goes badly. Then she is gone and the company is genuinely, materially blind for a quarter or more while the next person rebuilds the same undocumented map by making the same mistakes.

You did not lose an analyst. You lost your data layer. It just happened to have a birthday and a parking spot.

“We should document this” is a way of not fixing it

The standard response to this problem is a documentation initiative. Somebody creates a Confluence page called Reporting Definitions. It gets forty percent filled in during a slow week in August, and then it rots, because documentation describes a system without controlling it. The doc drifts from reality the moment somebody renames a field, and nobody notices for months.

Then it becomes a slightly darker joke, because now you have two sources of truth: the tools, and a stale wiki page that people cite in arguments.

The other standard response is to buy a BI tool and declare victory. Now the tribal knowledge is not in her head, it is in the tool. Sort of. Except the joins, the filters, and the eleven exceptions live inside dashboards that she built, using logic she chose, with no comments. Congratulations, you have promoted the key person risk from a Slack dependency to a software dependency that only one person can maintain. When she leaves, nobody will dare touch her dashboards. They will build new ones next to them. You have seen the folder.

The fix is boring: put the logic in the system, not the human

The actual problem is not that she is smart. It is that the reconciliation work happens in a place no system can see. Every adjustment she makes in her head is a business rule your company depends on and cannot inspect, version, or test.

The fix is to move that layer down into infrastructure. One place where all the sources land. One definition of a customer, a lead, a conversion, and revenue, applied the same way every time, applied on its own without anyone remembering to apply it. The exceptions become rules that are visible and editable rather than folklore. When somebody disagrees with a number, the argument is about the rule, which is a good argument to have, instead of about whether anybody typed the filter correctly, which is a stupid argument to have.

The test is simple and slightly rude. Pick your most important marketing number. Now ask whether a competent new hire, on day three, with no access to that one person, could get it right. If the answer is no, that number does not exist in your company. It exists in her.

The upside is not just risk reduction, though it is nice to stop having a single point of failure who can be hit by a bus or a better offer. The upside is that she gets her job back. She was hired to figure out what is working. She has been spending her week being a query engine with a pulse. Nobody was ever excited to do that, and the good ones leave over it, which is how you get the third scenario above.

Nobody in the room wants to say it

There is a version of this where the company knows it has the problem and quietly likes it. A person you can Slack is faster and more flexible than any system, and she is cheaper than the project it would take to replace her. That is a real calculation. It also has a ceiling, and you find it the week she is unreachable and a decision cannot wait.

If your reporting has a name and a calendar, that is worth noticing before the calendar fills up with interviews somewhere else.

THE DASHBOARD exists because we got tired of watching smart people spend their careers being the join between two systems. It unifies the whole stack into one source of truth you can ask questions in plain language, so the answer does not depend on who is online. Have a look at ihatemarketing.ai if that sounds like your Tuesday.