A dashboard tells you what happened. The next decade turns on what gets decided tomorrow

What does it feel like when someone across the chain describes your thesis without knowing you?#

Relief, and a certain urgency.

I read an interview with Dr. Cassiano Bevilaqua, Marketing and Technical Service director for Latin America at Cobb, conducted by AgroEditorial S.C.C. for Avicultura Ecuatoriana (2026). I have no relationship with the company or the publication, and none is needed: I read it the way you read someone who has walked far more farms than you have and who, coming from genetics, ends up pointing at the same gap I found coming in from the building.

One sentence is worth the whole interview:

"A dashboard full of data does not improve results on its own; it improves them when it produces the right decision at the right moment."

I have spent a long time building exactly that, and I have explained it at tables where the answer was "interesting, but we already have a dashboard." Finding it stated that way, from that side of the chain, does not prove me right — no quotation proves anyone right — but it confirms the gap is real, that it is regional, and that I am not imagining it from Armenia, Colombia.

This article is what I believe comes after that sentence.

Why does a dashboard not produce a decision?#

Because between the data and the decision there is work, and that work has an owner.

Someone looks at feed intake in house 7, compares it with flock age, remembers the heat two days ago, estimates how many kilos the catch will yield, weighs it against what the plant asked for on Thursday and what is left in the silo, and out of that comes an instruction: move it forward, wait, thin, shift feed. That work happens in a spreadsheet and in the head of a farm manager, every day, on every operation I know.

A dashboard does none of it. It shows the five variables and leaves the hard part untouched: combining them under this company's constraints — its plant slot, its capital, the commitments it has already signed — and issuing an order someone can carry out and confirm.

Two decades of industry investment went into measuring better: sensors, controllers, scales, integrations. That was the correct half of the problem. The other half — turning that measurement into tomorrow's decision — still lives in the heads of people who retire, change companies, or simply have a bad Tuesday.

Where is the money: in the record or in the tail?#

The interview puts it plainly: the regional step forward "will not depend only on reaching records, but on making good results more frequent and uniform." It is the most subversive sentence of all, because it changes the object of desire: no longer the star flock shown at the management meeting, but flock 16 — the one in the old house, the one nobody presents.

It can be turned into arithmetic. Take an operation running 20 flocks a year of 25,000 birds, with a catch weight of 2.5 kg. That is 1,250,000 kg of live weight a year. Suppose 15 of those flocks close at an FCR of 1.65 and 5 drift to 1.80.

You can redo the arithmetic in a minute: the 15 good flocks produce 937,500 kg, and 0.03 of FCR over those kilos is 28,125 kg of feed. The 5 laggards produce 312,500 kg, and 0.15 of FCR over those kilos is 46,875 kg.

Fixing the tail is worth 1.7 times what improving the already-good flocks is worth. And it is the cheaper of the two, because it needs no new nutrition and no new equipment: it needs flock 16 to be decided the way flock 3 was.

That is the point where a conviction becomes a plan. If the value sits in the tail, and the tail exists because the decision is not repeatable, then the thing to build is not another sensor. It is the mechanism that makes the decision repeatable.

What part of this can a machine do, and what part cannot?#

The interview has a sentence I sign in full: artificial intelligence "will help identify patterns and predict risks, but it will not replace field knowledge."

I take it one step further, and this is the design decision I have had to explain most and regret least: the model must not decide. Projecting a flock's weight fourteen days out is a statistical problem, and there a model beats a person. Choosing the catch day is not: it depends on what this company is after — kilos, margin, keeping its promise to the plant — and on what it is never willing to breach. That is not learned from historical data. It is declared.

So in the system I build, management declares the policy, a deterministic calculation decides within what is allowed, and natural language shows up only at the end, to write in one line the reason behind an instruction that was already settled. The same data and the same policy always produce the same decision. A projection is never presented as a measurement, and a declared constraint is not loosened because it would be convenient.

It sounds like a limitation and it is the opposite: it is the only thing that makes real delegation possible. Nobody hands over the catch day of their whole operation to something that answers one way on Tuesday and another on Thursday with no one able to explain why.

And who gets this: five large groups or five hundred mid-sized ones?#

That is the question that keeps me up, and it is where my reading parts ways with any global vendor's.

The interview says Latin America has extraordinary advantages: grain availability, production knowledge, export capacity, and a consumer who recognises chicken as affordable protein. All true. And around those advantages there are thousands of mid-sized producers too big for Excel and too small for the global integrators' systems: poor connectivity on farm, no room to carry three area managers, and their entire operating judgment living inside one or two heads.

If the decision layer only reaches those who can already pay for it, the gap the interview wants to close widens instead. And the one who loses is the least at fault: the producer who does the basics well, who knows their houses one by one, and who has no way to turn that into something that outlives the person who knows it.

I work for the other outcome. So that opening farm number 12 does not depend on finding person number 12. So that the reading the stockman takes at six in the morning, with no signal, inside the house, does not die in a notebook. So that the judgment of a manager with twenty years of craft ends up written down — not to replace them, but so it can be continued, compared and argued with when they are not in the room.

What does poultry look like when judgment is written down?#

It looks boring, and that is the highest praise I can give it.

It looks like a company where management declares once what it is after and what it will not breach, and that declaration governs the next two hundred flocks without anyone reinterpreting it each Monday. Where the difference between house 7 and house 8 is explained by what was known when the call was made, not by who was on shift. Where a son coming back from university finds his father's knowledge written down, instead of relearning it at the cost of three bad cycles.

Above all, it looks like a region whose advantage is not cheap grain but predictability: being able to commit kilos, deliver them, and commit again. That is the currency you compete with for the export markets the interview mentions, and it cannot be bought ready-made. It is built out of repeatable decisions.

What I am building, said plainly#

I am not making a better dashboard. I am making the missing layer between the data and the order: management declares the policy, the system decides within it, issues the instruction to a named person, and keeps a record of what was known at that moment so it can be reviewed later. It works with no signal, because inside a poultry house the signal does not reach. And it replaces neither the vet nor the farm manager in what is professional judgment: it takes off their hands the arithmetic they do today, by hand, at six in the morning.

Someone who travels the poultry industry across the continent describing the gap in the same terms does not prove my path is the right one. It confirms the gap exists, that it is regional, and that whoever fills it well will change how this industry decides. My bet is to fill it from Colombia, for the mid-sized producer, and to do it so that every decision the system makes can be audited afterwards.

A dashboard tells you what happened. The question that pays the payroll — and the one that will define the next decade — is what I do tomorrow with house 7.