Ground Truth EP 03: The Business Logic Was Never In The File. It Was In Someone's Head.

A piece published this week gave the industry's favorite illusion a name: automation theater, impressive dashboards while the real work still happens in spreadsheets and email, because the business logic that actually decides a number "exists only in someone's head." We hold that next to a data engineer's line, "alignment by agreement has no enforcement mechanism," a stat that says finance leaders burn five-plus hours a week just moving data by hand, and a VP of Marketing's real story of a decision that froze for months because nobody trusted the numbers anymore. Brought to you by Entrecore.

Transcript

SPEAKER_00 0:06

A VP of sales pulls up her dashboard, $2.4 million in Q2 revenue. The CFO opens his $2.1 million. Same company, same quarter, same underlying transactions. That is a real scene from a piece published in July. And the same piece puts a number on what happens next. Analysts spend anywhere from 30 to 60% of their time not analyzing anything, just hunting down why two reports disagree. This is ground truth. I am Maya, that is Theo. And every week we hold what the enterprise AI market is saying against what is actually happening on the ground. This week, the sharpest thing anyone wrote was not about a tool at all. It was about where the actual business logic lives.

SPEAKER_01 0:58

Right, and I want to start with the exact phrase because somebody finally named the thing we have been circling for two weeks. A piece from a writer named Leonard Coy calls it automation theater. His definition. That is exactly it. And he goes one line further than that, which is the part I keep coming back to. Critical business logic, he writes, exists only in someone's head.

SPEAKER_00 1:33

Not in a file anywhere.

SPEAKER_01 1:35

Not in a file, not in a catalog, not with anybody's name on it. There is no author. And once you notice that, a lot of this week's other findings stop looking like separate problems and start looking like the same one.

SPEAKER_00 1:48

Give me an example.

SPEAKER_01 1:50

A data engineer named Aline Oliveira wrote the cleanest version of it. Sales counts a deal closed when it hits the CRM. Finance counts it closed when the invoice is paid. Her words. Nobody is wrong. They are looking at different slices from different tools with no shared definition. Then she names why the usual fix does not hold. The usual fix is a meeting, she writes. It works for a week. Then a new quarter starts, edge cases pile up, and each team reverts to whatever their tool shows.

SPEAKER_00 2:22

Because the agreement lived in the meeting, not in a file either.

SPEAKER_01 2:26

Exactly the same shape as Kuey's line. And she lands it in one sentence I want to say twice. Alignment by agreement has no enforcement mechanism. The room agrees. Nobody wrote it down with an owner attached. Three months later, the agreement is gone, and nobody decided to break it. It just had nowhere to live.

SPEAKER_00 2:49

Did it show up anywhere else this week?

SPEAKER_01 2:51

It did. And I do not think it is a coincidence. A piece from a company called DLT Hub, cross-posted with a firm named Xenos, is literally titled Data Contracts, Agreement versus Enforcement. In their world, enforcement means something specific. A schema gets agreed between a producer system and a consumer system, and the pipeline breaks loudly if either side violates it.

SPEAKER_00 3:21

So that is enforcement too, just aimed somewhere else.

SPEAKER_01 3:25

Aimed at the wire, not the room. It will stop a malformed field from breaking a pipeline. It will not stop finance and sales from quietly meaning two different things by revenue, because nobody violates a schema by disagreeing on a definition. The schema never learns the definition changed. Only the humans do, and nothing enforces them.

SPEAKER_00 3:47

Is that actually costing real time? Or is this a philosophical complaint?

SPEAKER_01 3:52

It is costing real time, and there is a stat for it now. A finance research piece found that 69% of finance leaders spend at least five hours a week recreating reports, and 58% spend at least five hours a week just transferring data between systems by hand. That is not analysis. That is two people retyping the same business logic into two different tools because it was never written down anywhere, both of them could read it.

SPEAKER_00 4:23

What does that actually cost a company? Not a stat, an actual story.

SPEAKER_01 4:28

There is one, and it has a name attached, which is rarer than you would think in this kind of writing. Cassandra Gill is VP of marketing at a company called Superside. She described reports from two different tools, Marketto and Salesforce, that produced significantly different results. Her words There was a total mistrust in the data, and it resulted in team members sticking to the same strategies they had been running for months, uncertain of where to further focus their efforts.

SPEAKER_00 5:01

So the mistrust did not just slow a meeting down.

SPEAKER_01 5:04

It froze a decision for months. That is the part I want people to sit with. The failure mode is not a louder argument. It is a team that quietly stops trying to figure out what is working, because it no longer trusts the tool that would tell them. And there was nobody whose job it was to say which number was right.

SPEAKER_00 5:25

Is there a way to see how big that is across a lot of companies, not just one VP's story?

SPEAKER_01 5:31

There is a stat from a practitioner named Joe Rice, worth putting right next to her story. He found that three out of four data teams have nobody who owns their data products. And the teams running what he calls full anarchy, no owner at all, spend 45% of the week firefighting. Teams with an owner spend 27%.

SPEAKER_00 5:55

18 points of a work week. Just because nobody's name is on the definition.

SPEAKER_01 6:00

Just because nobody's name is on it. That is the superside story, multiplied across every team, with nobody responsible for what a number means once the meeting ends.

SPEAKER_00 6:12

What does that look like at real scale, not just one team?

SPEAKER_01 6:16

A BI founder named Ethan Ding described it plainly this week. His enterprise clients run 10,000 to 20,000 dashboards spread across systems, and he said it is extremely hard to prevent people from making infinite copies of data. Every one of those dashboards is somebody's private belief about what a number is, and none of them has an author either.

SPEAKER_00 6:38

10,000 versions of someone's head.

SPEAKER_01 6:41

10,000 versions of someone's head is exactly right. And Reese makes one more point worth sitting with because everyone is racing to put agents on top of exactly this mess. He says point a swarm of agents at a siloed company, and you will not end up with fewer silos. You will end up with more of them, faster, with better test coverage. The agent has no gut feel about your data, nobody to overhear at lunch, and no reason to hesitate. So it hands you a confident answer built on questionable foundations.

SPEAKER_00 7:17

Which is the 78% number we talked about a couple weeks ago. Confidently wrong at scale. That is a lot of places for one number to travel.

SPEAKER_01 8:10

It is, and it is a real improvement. Tools passing the same definition back and forth instead of each one guessing is better than the alternative. But look at what it does not touch. It keeps a definition consistent everywhere once that definition exists. It says nothing about the moment before that, whether finance and marketing agreed on the definition in the first place. You can make a wrong number perfectly consistent across 10 systems. It is still wrong in all ten.

SPEAKER_00 8:42

Did any vendor just skip the fixing part and go straight for the feeling?

SPEAKER_01 8:47

One did, and I have to give them credit for honesty about what they are actually selling. Kalibra's homepage leads with stop paying the hallucination tax. Context and control for AI. And they trademarked a name for the human feeling underneath this whole conversation. Data confidence.

SPEAKER_00 9:07

You can trademark confidence now.

SPEAKER_01 9:09

Apparently. And look at what is actually for sale under that name. Control for AI, a product aimed at the machine. Confidence is the word on the sign out front, aimed at whoever is reading the homepage, not at the thing that would actually earn it. Which is somebody's name on the definition.

SPEAKER_00 9:29

Okay, say I run a company and I do not want to wait for someone to finally write the file down. What do I actually do this week?

SPEAKER_01 9:38

Run Oliveira's test on yourself before you run it on anyone else. Pick the number you are most afraid someone will question in your next meeting. Ask a plain question about it. If the person who agreed to that definition left the company tomorrow, is it written down anywhere with their name on it, or does it only exist in their head? For example, say your bookings number counts a deal at signature in one system and at first invoice in another. And the only reason everyone currently agrees which one counts is that the same person has explained it in every meeting for a year.

SPEAKER_00 10:15

And if that person is on vacation next week.

SPEAKER_01 10:18

Then the number is not real yet. It is a belief one person is carrying around, and Kui's line applies exactly. It looks complete. It is not.

SPEAKER_00 10:29

We do not have this solved either, for what it is worth. We are just two people noticing that this week, several different writers described the exact same missing thing without ever using the same word for it.

SPEAKER_01 10:42

Automation theater.