The strategy kernel is the structure a strategy has to have, and it holds three parts, which are a diagnosis of the situation, a guiding policy for dealing with it and a set of coherent actions that carry the policy out.
Richard Rumelt set the kernel out in Good Strategy Bad Strategy, published in 2011. He arrived at the three parts after years of reading corporate plans that carried the word strategy and contained no argument at all. The kernel is the shape those plans were missing, so a set of choices about where to compete and how to win only becomes usable once it sits inside it.
That structure matters because it records the reason behind each choice. A product manager uses the kernel twice over. The first use is drafting a strategy with the team, and the kernel then says which part of the draft is still missing. The second is reading the deck a founder circulated on a Friday afternoon, and in that reading a missing part is easier to spot than a weak one.
The sections below take the three parts in the order the kernel puts them, starting with the diagnosis. They then show how each part decides what the next one may contain, and what kind of meeting a broken chain produces. Finally they look at the diagnosis itself, which is the part teams leave out most often.
The diagnosis of the situation
A diagnosis is an account of what is happening and of which obstacle stands between the organisation and the result it wants. Rumelt describes that work as figuring out what is going on, which sounds easy until a team tries it. A diagnosis has to simplify. Nobody can act on a situation described in full detail, so a diagnosis leaves most of the detail out. Choosing what to leave out is already a judgement about which facts matter. Naming the obstacle is the expensive part. It means writing down that the organisation has a problem, and most documents stop short of that.
One team did not stop short. Activation on its business scheduling product had fallen from forty per cent to twenty six over three quarters. The diagnosis it wrote was that self serve customers arriving from search could not connect a calendar without an administrator's permission, and that administrators were unreachable inside the fourteen day trial. That sentence is uncomfortable, because it says the product depends on somebody the buyer cannot get hold of. The comfortable version, which is that onboarding needs work, names no obstacle and leaves every action open.
The guiding policy
A guiding policy is the approach the organisation chooses for dealing with the obstacle the diagnosis named. A guiding policy rules out whole classes of action without naming a single action to take. That is what makes it different from a plan. For the scheduling product, the policy might be to make the trial useful with no administrator involved, and to move every permission that blocks a first success out of the trial path.
A guiding policy has force when it rules out proposals that sound reasonable. The policy above refuses a better invitation email to administrators, refuses a guided setup call and refuses anything that shortens the wait for a permission while leaving the wait in place. Each of those is a sensible idea on its own, and a policy that leaves all three available has decided nothing.
The set of coherent actions
Coherent actions are the steps that carry out the guiding policy, and the word coherent carries the weight. Actions are coherent when they support one another, so that the money spent on one makes the next one cheaper. For the scheduling product, four actions carry the policy.
- A trial workspace that runs with no calendar connected.
- Sample data that makes the first session useful on its own.
- A price and a payment route a single user can take without procurement.
- A support rota timed to the first hour of each trial.
Each action makes the others worth more. Sample data earns its cost only because the workspace no longer waits for a calendar, and the support rota reaches people only because the payment route let them in. A list of improvements with nothing holding them together is a backlog, and a backlog carries no argument about why these four came first.
The chain between the three parts
The order of the three parts is what supplies that argument. Each part decides what the next one may contain. If a diagnosis names no obstacle, then any guiding policy will do, because there is nothing for a policy to be wrong about. If a guiding policy rules nothing out, then any action will do, because every proposal satisfies it. The chain also runs backwards, and that is how a reader tests a document. Every action has to name the policy that requires it, and every policy has to name the obstacle it answers.
A strategy is written from the left and tested from the right. If an action names no policy, the break sits between the second box and the third, and the team is funding good ideas that belong to no argument.
A break in the chain stops those answers coming. Each break shows up as its own kind of meeting. If the diagnosis is missing, the argument about what to build never settles, because no agreed account of the problem exists to settle it. If the guiding policy is missing, every proposal gets judged on its own merits and the team funds a scattered collection of reasonable ideas. If the actions are incoherent, the document reads well and the quarter produces nothing anybody can point at.
The diagnosis teams skip
Of the three parts, the diagnosis is the one most often missing from a document. The reasons are practical. Writing a guiding policy feels satisfying, because it sounds decisive, and writing actions feels easy, because the team already keeps a list of them. Writing a diagnosis means saying in public that something is wrong, and then naming the thing. The thing named is often a decision the same leadership took the year before.
The test for a missing diagnosis is short. A document opening on the size of the market has described an opportunity, and an opportunity is no obstacle. A document opening on a list of themes for the year has described activity. A diagnosis appears only when a document says what stands in the way, in terms plain enough for a reasonable colleague to disagree with.
Reading a document for the three parts turns an uneasy feeling into a finding with a name on it. Rumelt used the same structure in reverse to describe the documents that fail. He set out four shapes bad strategy takes, and each shape is named for the part of the kernel it leaves out.