The mental models that earn their keep

A mental model earns its keep when it changes a decision you would otherwise have got wrong. By that test, most of the hundred-model latticework collapses to about five: stocks and flows, feedback loops, the constraint, Goodhart's law, and second-order effects. The rest are trivia — pleasant to know, rarely load-bearing.

This piece is the working shortlist from my systems-thinking guide for founders: which models pay rent in a 5–50 person B2B service firm, and the procedure for actually using one when a decision is in front of you.

Why do most mental models collect dust?

Because collecting them feels like the work. Reading about inversion or the availability heuristic produces the sensation of getting smarter without the inconvenience of changing anything. A model that never alters what you do on a Tuesday is a bookmark, not a tool.

The filter is blunt: can you point to a specific decision in the last quarter that a given model changed? If not, it is not one of your models. It is one of Charlie Munger's, or Wikipedia's.

Which five models change real decisions?

Stocks and flows. A stock is an accumulation (pipeline, cash, reputation); a flow fills or drains it (outreach, invoicing, delivery quality). Donella Meadows built Thinking in Systems on this distinction, and it explains the commonest founder illusion: revenue looks fine while the pipeline stock quietly drains, because you stopped the outbound flow six weeks ago and the delay hides it.

Feedback loops. Reinforcing loops amplify (referrals beget referrals); balancing loops cap (a full diary stops selling, which empties the diary). If growth keeps stalling at the same headcount or revenue, a balancing loop is doing it — not a lack of effort.

The constraint. Goldratt's argument in The Goal: one bottleneck governs the throughput of the whole system, so improvement anywhere else is cosmetic. In most service firms the constraint is the founder's calendar. Every improvement plan should start with "is this at the constraint?"

Goodhart's law. When a measure becomes a target, it stops measuring. Pay for booked calls and you will get booked calls — with attendees who never intended to buy.

Second-order effects. Every intervention has consequences, and the consequences have consequences. Discounting wins the deal (first order), then anchors the client at a price you resent for three years (second).

How do you apply a model to a live decision?

The mechanism is a five-question pass, run in order, before committing:

1. Name the decision in one sentence. "Hire a second delivery person" — not "sort out capacity".
2. Ask which stock it changes. When a decision only increases a flow that feeds an already-full stock, then it is waste; capacity added away from the shortage does nothing.
3. Trace the loop it sits in. When the decision strengthens a reinforcing loop, expect compounding; when it fights a balancing loop, expect the system to push back and budget for that.
4. Check it against the constraint. When the change does not relieve the bottleneck, then throughput will not move, whatever else improves.
5. Play the tape forward twice. First-order effect, then the second-order response — from clients, staff, competitors, and your own future behaviour.

Ten minutes, on paper. When a decision fails all five questions, then declining it is the decision.

Where do models mislead?

At the edges of what they include. Every model draws a line around "the system" and ignores what is outside it — a choice I unpack in drawing the boundary around your system. Optimise the model of your sales process while excluding delivery capacity, and you will win work you cannot serve. The map is not the territory; a five-model map is still a map.

The corrective is to occasionally draw the whole thing — the full walkthrough of mapping a business as a system forces the boundary question into the open.

How do you keep the collection small?

One page, five models, each with the last decision it changed written next to it. When a model goes two quarters without changing a decision, retire it; when a decision goes wrong, ask which model would have caught it, and audition that one.

And apply the same earn-your-keep test to your commercial thinking. The most expensive un-applied model in B2B is knowing you need a sharp ideal customer profile and never writing it down — because if you cannot name your buyer, you cannot build outbound, and no latticework compensates for that.

---

Keep the thinking coming: The Format — a fortnightly letter on business and personal systems. Subscribe by email.