Systeric / Docs
Open App →

Rumelt’s Kernel

Your car is stuck in mud. The rear wheels are spinning and the smell of burning is getting worse.

Somebody says: our goal is to be out by three o’clock, and we are committed to being a car that does not get stuck.

Everyone recognises that this is useless, and the reason is worth being precise about. It states a desired outcome and a value, and says nothing about what is actually happening or what anyone should do differently. It could be said in any situation. It survives contact with no facts because it contains none.

Now somebody looks under the car and says: the driven wheels are on the wet side and there is almost no weight over them, which is why throttle makes it worse. So we stop adding power, we get weight over those wheels, and we go gently. Three of you sit over the rear axle, the mats go under the wheels, and we rock it out in second.

That is the same situation and a completely different kind of statement. It has the three parts that Richard Rumelt calls the kernel of a strategy.

The three parts#

Rumelt set this out in Good Strategy Bad Strategy (2011), and it is the most useful test we have for whether a document is a strategy at all.

A diagnosis. An account of what is actually going on that simplifies an overwhelming situation down to the thing that matters. Its job is to replace the complexity of reality with a specific, addressable challenge. Diagnosis is a judgement about which fact is load-bearing, and it is where almost all the value is.

A guiding policy. The overall approach chosen to deal with the challenge the diagnosis named. Not a goal and not a plan: a direction that rules some things out. Getting weight over the driven wheels rules out revving harder. A guiding policy that forbids nothing is not one.

Coherent action. The coordinated steps that carry the policy out, which must actually reinforce each other. Coherence is the demanding word. Actions that each make sense alone but pull against one another are the most common failure in real strategies, and they are usually the result of giving every department something to do.

diagnosis what is going on guiding policy the approach coherent action pulling together the same way
The order is not decorative. A policy chosen without a diagnosis is a preference, and actions chosen without a policy are a to-do list that happens to be busy.

The four signatures of bad strategy#

Rumelt’s other contribution is a set of tells, and they are unusually practical because you can run them over a real document in about two minutes.

Fluff. Restating the obvious in inflated language, usually with a fashionable noun doing the work. If deleting the sentence loses no information, it was fluff.

Failure to face the challenge. The document never names what is actually hard. This is the most common signature and the most consequential: a strategy that does not identify an obstacle cannot be evaluated, because there is nothing to be right or wrong about.

Mistaking goals for strategy. A list of desired outcomes with numbers attached, and no account of how they will be achieved or why they are achievable. “Grow revenue 30%” is a goal. The strategy is the part that explains why 30% is available and how.

Bad strategic objectives. Objectives that fail to address critical issues, or a scattered list of unrelated ones. The tell is a list long enough that nothing has been chosen, which usually means the choosing was avoided by giving everyone something.

grow 30% delight customers be best in class goals only the challenge, named the approach actions a kernel
Both documents are the same length and both would survive a planning meeting. Only one of them can be wrong, which is the property that makes it a strategy.

How to tell#

Look for the obstacle in the first page. If nothing has been named as hard, you are holding goals. Ask what the author thinks is standing in the way; if they can answer fluently but it is not written down, the document is simply missing its most important part.

Ask what the policy rules out. A guiding policy that forbids nothing is a slogan. If every reasonable action is still permitted after reading it, no direction was chosen.

Check whether the actions reinforce each other. Take any two and ask whether doing both makes each more effective, or whether they merely coexist. Coexisting is normal; reinforcing is strategy.

Try to falsify it. A real strategy contains a claim about the world that could turn out to be wrong. If nothing in the document could be disproven by next quarter’s evidence, there is nothing in it.

When it goes wrong: the usual repair is not more planning. It is going back to the diagnosis, which is almost always the part that was skipped, because it is the only part that requires admitting something is not working.

Where this stops#

The kernel does not tell you how to find a good diagnosis. It tells you that you need one. Finding it is the work of issue trees and Diagnosis, and Rumelt is comfortable saying that this part is genuinely hard and cannot be proceduralised.

It is easy to write post hoc. You can take a decision already made for other reasons and reverse-engineer a kernel around it, and the result will pass every check on this page. The guard is the same as everywhere: name in advance what evidence would change your mind.

It has nothing to say about execution. A correct kernel that nobody acts on is a document. That is what the Framework cascade and the influence model are for.

How we use it here#

The kernel is the acceptance test for a half strategy and for any two-pager proposing a direction. Before circulating one, check that a reader could point to the diagnosis, state what the policy rules out, and see why the actions belong together.

Our own strategy is a useful worked example. The diagnosis is that the category ships correctness without earning belief, so customers re-check every payroll in a spreadsheet. The guiding policy is to win on explanation rather than coverage, which rules out competing on regulation count. The coherent actions follow from that and reinforce each other: variance analysis, an itemised sequence behind every number, and running our own payroll on it so we feel it first.

That is a strategy in the sense this page means, because it could be wrong. If customers turn out to re-check regardless of how well the numbers are explained, the diagnosis was wrong and we would know.