Andy Grove ran Intel for two decades and spent a meaningful portion of that time thinking about what managers actually produce. His conclusion was unsentimental: a manager's output is the output of the organizations under their influence. Not their analysis. Not their meetings. Not their strategy documents. The decisions that moved through the system and changed what happened next — that was the product.
The insight is more radical than it sounds. It reframes every activity in an organization as either contributing to a decision or not. The dashboard is not the product — the decision it enables is. The financial close is not the product — the capital allocation it informs is. The weekly operating review is not the product — the course correction it produces is. Strip away every intermediate artifact, and what remains is a sequence of choices, made under uncertainty, that collectively determine whether the business creates value or consumes it.
Most organizations have never oriented themselves around this idea. They've oriented around outputs — reports produced, meetings held, analyses completed — and treated decisions as the thing that happens somewhere in the background after the outputs have been assembled. The result is a business that is extraordinarily busy producing the preconditions for decisions while the decisions themselves happen late, happen poorly, or don't happen at all.
The Decision as Unit of Output
Ajay Agrawal, Joshua Gans, and Avi Goldfarb — economists at the University of Toronto who study the economics of artificial intelligence — proposed a framework that reframes how businesses should think about information technology. Their argument: the fundamental unit of business value is not data, not prediction, not even insight. It's the decision. Data is valuable only to the extent that it improves a decision. Prediction is valuable only to the extent that it reduces the uncertainty around a decision. And the value of reducing that uncertainty is determined entirely by what's at stake in the decision itself.
This sounds abstract until you apply it concretely.
A company invests six months building a sophisticated revenue dashboard. The dashboard is beautiful. It updates daily. It shows revenue by product, by channel, by customer cohort, with trend lines and variance indicators. The investment was substantial — engineering time, data pipeline work, design iteration. And when it's done, leadership looks at it each Monday, nods, and moves on. No pricing decision changes. No channel allocation shifts. No customer strategy adjusts. The dashboard exists. The decisions it was supposed to improve continue to be made the same way they were made before.
The dashboard failed. Not because the data was wrong or the visualization was poor. Because nobody designed the connection between what the dashboard shows and what the organization does when it shows it. The intermediate artifact was produced with care. The decision — the actual product — was never addressed.
Now consider the inverse. A company maintains a simple spreadsheet that tracks three numbers weekly: cash position, collections aging, and committed outflows. The spreadsheet is ugly. It has no trend lines. A summer intern could build it. But it has a rule attached: if the 8-week forward cash projection drops below a defined threshold, the CFO convenes a specific meeting with a specific agenda within 48 hours. The threshold has been crossed twice. Both times, the meeting produced a specific action — once accelerating collections on three large receivables, once deferring a capital expenditure by one quarter. Both actions prevented a cash constraint that would have forced reactive decisions under pressure.
The spreadsheet produced better decisions than the dashboard. Not because it was a better tool, but because it was designed as a decision system rather than an information display.
Why Decision Throughput Degrades
Growing companies don't usually have a shortage of decisions to make. They have a shortage of decisions that actually get made — cleanly, on time, with the right information, by the right person, with accountability for the outcome.
The degradation follows a predictable pattern. When the company was small, decisions happened quickly because the decision-maker had direct access to the relevant information. The founder knew the cash position because she checked the bank account. She knew the margin because she quoted every job. She knew the delivery status because she talked to the team daily. The information-to-decision distance was zero. The decisions were fast not because she was decisive by temperament, but because the infrastructure was simple enough to be transparent.
As the company grows, that transparency erodes. The founder no longer checks the bank account — someone else produces a cash report, which takes two days to assemble because it requires reconciling three systems. She no longer knows the margin on every job — the margin report comes monthly, is usually revised after the close, and reflects a cost allocation methodology nobody has documented. She no longer knows delivery status — it lives in a project management tool she doesn't use, summarized in a weekly email that arrives after the decisions it should have informed.
The information-to-decision distance has grown. Not because anyone made a bad choice, but because growth naturally introduces intermediation — more systems, more people, more steps between the event and the person who needs to know about it. Each step introduces latency, potential error, and the chance that the information arrives too late or in a form that requires translation before it's useful.
McKinsey's research on organizational decision-making quantified this: executives spend 37% of their time making decisions, and 61% of that time is used ineffectively. The inefficiency isn't in the thinking. It's in everything that precedes the thinking — assembling the data, reconciling competing versions, building confidence that the information is current and accurate. By the time the decision-maker has the inputs, the window for the decision may have narrowed or closed.
The companies that make decisions fastest are not the ones with the most decisive leaders. They're the ones with the shortest information-to-decision distance — where the infrastructure produces trustworthy signals in time to act, and the governance specifies what action each signal implies.
The Anatomy of a Decision System
A decision system has four components, and most organizations have invested heavily in the first two while ignoring the second two entirely.
The first component is signal production: the infrastructure that generates the numbers, metrics, and indicators the organization uses. This is where most investment goes — dashboards, BI tools, data warehouses, reporting automation. The investment is necessary but not sufficient, because signals without governance are noise with better formatting.
The second component is signal integrity: the governance that ensures the signals are trustworthy — consistent definitions, reconciled sources, authoritative ownership, documented lineage. This is the work described across several articles in this series, and it's the prerequisite for everything that follows. A decision made on a signal that isn't trustworthy is not a better decision than one made on intuition — it's a worse one, because it carries the false confidence of apparent evidence.
The third component is decision architecture: the explicit specification of which signals connect to which decisions, who makes each decision, what information they need, and what authority they have. This is the component most commonly absent. Organizations produce signals and assume the decisions will happen. They often don't — because nobody specified the trigger, the decider, or the response.
Decision architecture answers four questions for every recurring decision the business makes. What signal triggers this decision? (Not "when we feel like we should discuss it" — a specific threshold or condition.) Who decides? (One person, named, with authority — not a committee that produces consensus through exhaustion.) What information does the decider need, and where does it come from? (Specified in advance, so the information is assembled before the decision window opens, not after.) And what happens after the decision is made? (Who is accountable for execution, what the success criteria are, and when the decision gets reviewed.)
The fourth component is decision memory: the institutional record of what was decided, why, by whom, and what the expected outcome was. This is the component that prevents organizations from relitigating decisions that were already made, losing context when the decision-maker changes roles, and failing to learn from decisions that didn't produce the expected results.
Without decision memory, organizations have a specific and expensive failure mode: the same decision gets made three times. Once when the problem first surfaces. Again six months later when someone new encounters the same problem and doesn't know it was already addressed. And a third time when the original decision's outcome was never evaluated, so nobody knows whether the first answer worked. Each iteration consumes leadership time and organizational energy that should have been available for new decisions.
The Leverage Point Most Companies Miss
Naval Ravikant made an observation about leverage that applies directly to how businesses should think about infrastructure investment. There are three forms of leverage: labor (other people working for you), capital (money working for you), and products with no marginal cost of replication (software, media, code). The third form is the most powerful because it scales without proportional effort.
Decision infrastructure is the closest thing a services business gets to the third form of leverage. A metric definition, once established and governed, improves every decision that touches that metric — forever, without requiring the person who established it to be present. A decision cadence, once installed, produces decisions on a rhythm that doesn't depend on anyone remembering to convene the meeting. An exception threshold, once defined, surfaces problems automatically rather than waiting for someone to notice.
The leverage is in what the system produces without ongoing human effort. Every decision that happens faster because the information was already assembled. Every course correction that happens earlier because the signal fired before the problem compounded. Every meeting that starts with "here's what we need to decide" instead of "let's figure out what the numbers say."
This is the return on infrastructure investment that doesn't appear on any ROI calculation but manifests in the cumulative trajectory of the business. Hundreds of decisions, made slightly better, slightly faster, over years. The compound effect is the difference between an organization that learns and one that repeats.
The Test
There is a diagnostic question that reveals whether an organization is oriented around decisions or around outputs.
Ask any senior leader: what are the three most consequential decisions your organization needs to make in the next 90 days?
If the answer comes quickly — with named decisions, named deciders, and a clear sense of what information is needed — the organization has decision architecture. If the answer comes slowly, or comes as a list of projects and initiatives rather than decisions, or defaults to "we need to get aligned on our strategy" — the organization is producing outputs and hoping decisions emerge.
The second diagnostic is retrospective: for the last significant decision the organization made, can anyone point to the decision log entry that records what was decided, why, who owns the outcome, and when it will be reviewed?
If that entry exists, the organization has decision memory. If it doesn't — if the decision lives in someone's recollection, or in meeting notes that nobody will read again, or nowhere at all — then the organization is spending its decision-making capacity without compounding the results.
The infrastructure that produces better decisions is not glamorous. It doesn't generate impressive visualizations or novel analytical insights. It does something more valuable: it shortens the distance between a question and an answer, between a signal and a response, between a problem identified and a problem resolved. That distance, multiplied across every decision the organization makes, is the gap between a business that feels chaotic and one that feels governable.
Decisions are the product. Everything else is packaging.
Decisions Are the Product explores why decision quality is the true output of business infrastructure. Related: The Visibility Crisis, The New Executive Stack, Calm Is a System.