01 · THE GREEN DASHBOARD
Somewhere a dashboard is filling with green. Hours saved. Drafts written. Tickets closed faster. The numbers may be real, and the gain may be worth taking. But the dashboard is answering the first question only: what did the tool produce for us today? Strategy begins with the question the screen leaves blank. After rivals buy the same capability, customers adjust their expectations, and the workflow needs checking and upkeep, what part of this value do we still keep? The first number is gross value. The second is retained value. Selection is the discipline of not confusing them.
02 · FLOOR, NOT CEILING
A 2023 field study by Erik Brynjolfsson, Danielle Li, and Lindsey Raymond gives us a clean gross-value result. The researchers studied the staggered introduction of an AI assistant to five thousand one hundred and seventy-nine customer-support agents at one software firm. Productivity meant issues resolved per hour. Access raised that measure by fourteen percent on average and by thirty-four percent for novice and lower-skilled agents, with minimal effect on experienced and highly skilled agents. That is a substantial, bounded operating gain. It shows the floor rising inside this setting. It does not show whether competitors adopted the same tool, whether prices changed, or whether the firm kept an advantage. The study measures productivity, not durability.
03 · WHERE THE SAVING GOES
So separate the operating case from the strategy case. A shared capability can become lower price, faster service, a smaller queue, or simply the new standard customers expect. It can also create temporary margin while competitors lag. None of those outcomes is automatic. Run the counterfactual instead. If every rival had this capability tomorrow, what would still be ours? If the answer is nothing, bank the gain, but call it table stakes. If something remains, name it before calling the deployment an edge.
04 · THE SEAM
Ajay Agrawal, Joshua Gans, and Avi Goldfarb give the seam an economic shape. They describe machine learning as prediction technology, with prediction serving as an input into decisions. When that input gets cheaper and spreads, advantage can move to a complement around it. The complement might be a proprietary resolution history, a workflow rivals cannot reproduce, exclusive rights, a fast feedback loop, or a customer relationship that turns the same prediction into a different result. Judgment is one complement, not a magic word. The selection job is to identify the scarce complement and explain why it will remain scarce. A common model plus a common process is a floor. A common model joined to a defensible complement may be an edge.
05 · TWO CORNERS
Now set two corners against each other. In the first, the AI capability spreads quickly and the business complement is common. Every firm can buy the tool and attach it to roughly the same process. That is the operating floor. In the second, the same widely available capability feeds something genuinely scarce: a right, a dataset that keeps improving through use, a trusted channel, or an operating loop competitors cannot quickly assemble. That is only a candidate edge. One question remains before it earns the name: who captures the benefit after customers, workers, suppliers, and rivals respond? No capture mechanism, no retained value.
06 · WICKED GROUND
Here is the adversarial turn. Rare and consequential does not mean defensible. Daniel Kahneman and Gary Klein concluded that skilled intuition needs two conditions: an environment predictable enough to contain learnable regularities, and a real opportunity to learn them through feedback. Robin Hogarth called the opposite wicked ground, where feedback is sparse, delayed, or misleading. A firm can call senior judgment its moat when the domain has never taught that judgment whether it is right. Scarcity without honest feedback may be nothing more than scarce confidence. Before funding judgment as the complement, ask for its learning loop.
07 · Advertisement · ElevenLabs Studio
The last mile of content is often sound: the right voice, the clean take, the version that works in another market. ElevenLabs brings AI voice, music, dubbing, and sound tools into one production flow, so teams can shape audio with the same care they bring to the story. Create the version people can actually hear.
08 · RARE FOR YOU
Rarity can diffuse too. A supplier failure may be once in a decade for you and routine across a large documented corpus. A model can propose the reference class and summarize how candidate cases ended. That makes part of the outside view cheaper and less scarce. What may remain differentiated is narrower: validating that these cases are genuinely comparable, joining them to private context, choosing the error tradeoff, and carrying the commitment. Or none of that may be defensible either. The point is not to move value automatically from the model to the human. It is to keep testing where scarcity actually survives.
09 · TWO MAPS, INVERTING
Draw the portfolio twice. The first map ranks deployments by measured operating gain: time, cost, quality, throughput. The second is not observed data. It is the strategy estimate. For each bright cell, discount the gain for how fast the capability diffuses, how much benefit passes to customers or other parties, and what verification and upkeep cost. Then ask whether a scarce complement and a credible capture mechanism remain. A large saving can fade on the second map. A smaller saving can persist if it compounds through something competitors cannot match. The maps need not invert. They simply answer different questions, and both belong in the selection memo.
10 · POINT IT THERE
Take the measurable wins. Refusing a real operating gain because it is not a moat would be its own mistake. But make every proposal carry four answers. Metric: what moves, for whom, and over what horizon? Diffusion: how soon can rivals match the capability? Complement: what do we control that remains scarce? Capture: who receives the benefit after competitive response, verification, and upkeep? If the last two answers are empty, bank the saving and call it table stakes. If they are concrete, test them and fund the edge. The dashboard proves that something moved. Selection decides whether it can still matter after everyone else moves too.