There is a kind of business decision that can never be wrong in the way that matters. Not wrong as in incorrect, wrong as in blamed. If you made it from the data and it fails, the failure belongs to the world, not to you. You followed the numbers. This is the quiet appeal of being data-driven, and it has almost nothing to do with whether the data was any good.
A data-driven decision can always be presented as rational, and you will never be blamed for making it.
That is Rory Sutherland's argument, close to verbatim. Its sharpest edge is an asymmetry most organisations feel but rarely name: it is far easier to be fired for being irrational than for being unimaginative. Caution has a defence; originality does not. So the rational move, for the individual, is to reach for the number, the benchmark, the dashboard, anything that can be pointed at later. The trouble is what that habit does to the firm over time.
All big data comes from the past
Data has one structural limit that no amount of volume fixes: it only contains what already happened. It is a record of the decisions you did make, the products you did ship, the prices you did set. It holds no counterfactuals, no trace of the experiment you never ran, the segment you never served, the price you never tried.
It is unrepresentative in a quieter way too: the things that actually move a decision, how a place feels to live in, whether a product is loved, resist the spreadsheet, so they get dropped. And a dropped variable reads, downstream, as a variable that does not matter. A firm that becomes obsessively data-driven, often for defensive reasons, ends up fixated on its own past and structurally unable to imagine anything outside it.
Two ideas wearing one coat
The argument is really two older ideas stacked together, and pulling them apart makes it more useful than the slogan.
| Mechanism | What it is | What it does to the firm |
|---|---|---|
| Goodhart's Law | When a measure becomes a target, it stops being a good measure. | Optimise one metric long enough and it comes loose from the value it once stood for. Year one the new metric is useful; year ten you are optimising the number, not the thing. |
| Mimetic isomorphism | Under uncertainty, firms copy each other (DiMaggio & Powell, 1983). | Imitation feels safe; it is the institutional form of "data-driven," since no one is blamed for doing what everyone does. And imitation converges. |
Stack them and you arrive exactly where Sutherland does. Everyone optimises; everyone optimises the same metric, because the benchmarks are shared; and everyone hedges by copying everyone else.
Everyone's spreadsheet agrees, because everyone's spreadsheet was trained on the same past.
The endpoint is corporate isomorphism, a clumsy name for a simple result: businesses that resemble each other, products that resemble each other, pricing that resembles each other. That is the worst market to be standing in, commodified and poorly differentiated (everything interchangeable), where the only thing left to compete on is the one nobody wanted, price.
Notice who the argument flatters
It is worth being suspicious here, because some of this is too convenient. "Easier to be fired for being irrational than unimaginative" is true, but it is also precisely the thing a creative-industry person needs to be true: it licenses the intuition they were going to back anyway. And "the important things do not fit in a spreadsheet" is half insight, half escape hatch. The insight is real, data has no counterfactuals. The escape hatch is that "you cannot quantify it" is the oldest excuse for refusing to test a belief you would rather keep.
The sharper version is not "use intuition instead of data." It is "use data to kill intuitions, not to generate them."
Data is poor at telling you what to try and excellent at telling you what to stop. The generative move, the new product, the odd price, the untested segment, has to come from somewhere the data cannot reach, because by definition it is not in the record yet. But once you have tried it, the number is the right tool to decide whether it stays. Intuition proposes; data disposes. Run it the other way, with data proposing, and you get the commodified middle.
Where this shows up in security
It maps onto security operations almost too cleanly. A managed security provider (the firm a company hires to watch its alerts) that has optimised how fast it reacts and how many alerts it closes for a decade is optimising exactly what every competitor optimises, against benchmarks the whole industry shares. The defensive, data-driven, unblameable choice is to buy the same tooling everyone else bought:
- the same automation platform, so your playbooks look like the reference playbooks;
- the same detection content, tuned against the same public benchmarks;
- the same headline metrics, time-to-respond and alerts-closed, that every competitor reports.
Make all three choices and you have converged into the same interchangeable middle as everyone else. Response time in year one is a useful stand-in for quality. In year ten, with the whole industry optimising it, it mostly measures how hard everyone is gaming the response-time number. The escape is upstream of the dashboard, in the decision the metric never sees, not in another fraction of a second shaved off a number everyone already reports. See also the SOAR anti-pattern tax and monitoring is not understanding.
Measurement is not a hiding place
None of this is an argument against measurement. It is an argument against measurement as a hiding place. The firms that escape the commodified middle are not the ones that ignore data; they are the ones willing to do the unmeasured thing first and measure it second, to make a decision the spreadsheet could not yet justify and then let the spreadsheet judge it. The defensive move and the original move point in opposite directions. Knowing which one you are making, and why, is most of the discipline.
The missing counterfactual is one column the spreadsheet lacks. For two others, whose job a thing is and what gratuitous cost means, see the companion piece, the value finance can't see.
"Data-driven" is partly a defence. If you decide from the numbers and it fails, the failure belongs to the data, not to you. It is easier to be fired for being irrational than for being unimaginative, so the safe individual move is to reach for a benchmark you can point at later. That habit is fine for one person and corrosive for the firm.
All data is the past. A dataset records the decisions you did make and the products you did ship. It holds no counterfactuals: the segment you never served, the price you never tried, the experiment you never ran. Things that resist measurement get dropped, and a dropped variable reads downstream as one that does not matter. So a data-driven firm stays fixated on its own history and cannot see outside it.
Two mechanisms stack. Goodhart's Law: once a metric becomes a target, you optimise the number, not the thing it once stood for. Mimetic isomorphism: under uncertainty firms copy each other, because nobody is blamed for doing what everyone does. Put them together and the result is convergence.
- everyone optimises;
- everyone optimises the same metric, because the benchmarks are shared;
- everyone hedges by copying everyone else.
Everyone's spreadsheet agrees, because everyone's spreadsheet was trained on the same past.
The endpoint is a commodified market: firms that look alike, products that look alike, pricing that looks alike. The only thing left to compete on is the one nobody wanted, price.
The sharper rule
The fix is not "use intuition instead of data". It is "use data to kill intuitions, not to generate them". Data is poor at telling you what to try and excellent at telling you what to stop. The new product or the odd price has to come from somewhere the record cannot reach, because it is not in the record yet. Try it first, then let the number decide whether it stays. Run it the other way, with data proposing, and you get the beige middle.
Where it bites in security. A managed provider that has optimised time-to-respond and alerts-closed for a decade, on the same tooling and the same public benchmarks as every rival, has made the most defensible and least distinctive set of choices possible. Year one, response time tracks quality. Year ten, it mostly measures who games the number hardest. The escape is upstream, in the decision the metric never sees, not in another fraction of a second.
Sources
- Rory Sutherland: the argument here is close to his verbatim phrasing, the "never fired for a data-driven decision" framing, "all big data comes from the past," and the slide into corporate isomorphism. These recur across his talks and in Alchemy: The Dark Art and Curious Science of Creating Magic in Brands, Business, and Life (2019).
- Charles Goodhart (1975): Goodhart's Law. The popular form, "When a measure becomes a target, it ceases to be a good measure," is Marilyn Strathern's 1997 gloss.
- Paul DiMaggio and Walter Powell, "The Iron Cage Revisited: Institutional Isomorphism and Collective Rationality in Organizational Fields," American Sociological Review (1983): mimetic isomorphism, firms imitating under uncertainty.