In short: At a few hundred people, founders make a small fraction of the decisions. Managers make most of them, many managers have never met the people whose decisions they inherit, and turnover becomes constant. The enterprise brain's job becomes delegation: letting good reasoning travel.
Key points
- Define recurring decision classes with an owner, template, needs list and falsifier convention.
- A precedent slot shows the three most comparable prior decisions, with outcomes, before a manager decides.
- Make a five-part handover pack standard: decisions, open threads, relationships, graveyard, question log.
- Start calibration by adding probabilities in one high-volume class whose outcomes arrive within two quarters.
- Avoid compliance theatre: hold a three-field floor and keep capture time under a minute.
Growth introduces a middle layer. Decisions that a founder once made in a minute, with full context, are now made by a team lead with partial context, often following a precedent set by someone who has since left. The organisation is making more decisions than ever, and each one is made with less of the history behind it.
This is also the first stage at which some decision classes become frequent enough to learn from. Sales deals, support escalations, hiring decisions and discount approvals each run into the hundreds a year.
Decision classes, formally
Define the recurring classes properly now. For each: the owner, the template, the needs list (what a good decision in this class requires), and the falsifier convention. A discount approval should look the same wherever in the company it is made, and be comparable with every other discount approval.
Precedent for managers
The single most valuable feature at this stage is the precedent slot: before a manager decides, the three most comparable prior decisions in that class, with outcomes. It gives a new team lead the benefit of the decisions their predecessors made, without needing the predecessors.
Handovers become routine
With regular departures and internal moves, handovers happen most months. Make the five-part handover pack standard for any role that owns decisions: decisions, open threads, relationships, graveyard, question log. It is the cheapest protection against losing reasoning when people move.
The first calibration
Pick the highest-volume class with outcomes that arrive within two quarters, often sales deals or renewals, and add a probability to each decision. After a year you will know, for the first time, whether the organisation's judgment in that class is well calibrated, and whose calls are reliable.
The trap
The trap at two hundred is compliance theatre. Templates get longer, fields multiply, and people fill them to satisfy the process rather than to record their thinking. Hold the line on the three-field floor: decision, reason, what would change your mind. Draft the rest. Measure capture time and keep it under a minute.
In practice
- Define the ten most common decision classes with owner, template, needs list and falsifier convention.
- Add a precedent slot to the workflow for each class.
- Make the handover pack mandatory for any role that owns decisions.
- Start calibration in one high-volume class with short outcome lag.
- Audit capture time quarterly; cut any field that is not used.
What would prove this wrong. Two-hundred-person organisations without structured decision classes or precedent showing no more inconsistency in similar decisions, and no slower manager ramp-up, than those with them.
Common questions
How do growing companies keep decisions consistent as managers change?
Define recurring decision classes formally and add a precedent slot that shows the three most comparable prior decisions with outcomes before a manager decides.
What should a handover pack include?
Five parts: decisions, open threads, relationships, graveyard, and question log. It should be standard for any role that owns decisions.
How do you stop decision templates becoming bureaucratic?
Hold the line on a three-field floor: decision, reason, and what would change your mind. Draft the rest, measure capture time, keep it under a minute, and cut unused fields.