Before accounting could work, it needed a classification. Not a ledger. Not a system. A way to look at a transaction and say: this is an asset, that is a liability, this is revenue, that is an expense. Without that, a ledger is a list of numbers and a list of numbers cannot balance, cannot be audited, cannot tell you anything.
Decisions have never had that classification. Every decision a company makes, whether it reshapes the organisation or buys printer cartridges, is treated as the same kind of event. Same inbox. Same approval thread. Same absence of a record. Not because the ledger was wrong. Because the chart of accounts was missing.
STEP is that chart. It is the single most important idea in everything I have been building, and I want to give it the space it deserves.
What STEP says
Every consequential decision a company makes falls into one of four types. Not forty. Not a spectrum. Four, and each one behaves differently, fails differently, and needs to be kept differently.
S is for Structural.
A structural decision changes the shape of the company itself. It alters who is in the organisation, how authority is arranged, or what the company is bound to on the outside.
Hiring is structural. You are adding a person, a role, a reporting line, a cost. Promotion is structural. You are changing someone's standing, their authority, their compensation, their trajectory. Reorganisation is structural. Vendor onboarding is structural. You are binding the company to a new external party. Delegation of authority is structural. Partnership agreements. New entities. Every one of these decisions rearranges the organisation in a way that is difficult to reverse and that other people's work will be built on top of.
Structural decisions carry a particular weight because they affect people's lives directly. A hiring decision determines somebody's livelihood. A promotion decision shapes a career. A reorganisation changes who someone reports to, what they work on, how their day feels. These are not abstract entries. They are commitments that the people affected remember precisely, even when the company does not.
When a structural decision is not properly kept, it gets disputed. Not immediately. Months later, sometimes years, when somebody asks why a reporting line was drawn the way it was, or who agreed to a vendor relationship that has since gone wrong, or what the basis was for a promotion that someone else believes they deserved. The dispute cannot be resolved without the record, and the record does not exist, so the dispute runs on memory and seniority and politics instead of on facts.
Structural decisions need authority, sequence, and reasoning preserved. Not because the company is being careful. Because these are the decisions people will come back to, and when they do, the company needs an answer that is more than somebody's recollection.
T is for Transactional.
A transactional decision moves money or resources from one place to another. It is the daily traffic of corporate approval.
Budgets. Purchase orders. Capital expenditure. Contract renewals. Discounts. Credit terms. Payment releases. Expense approvals. These arrive in volume, they cross thresholds, and they have deadlines, often external ones set by vendors, customers, or regulators.
Transactional decisions are the ones companies understand best, because they look most like the transactions accounting already handles. They have numbers. They have limits. They have a natural home in procurement systems and finance tools. Most companies, if they have systematised any part of their decision-making at all, have systematised this one.
What makes transactional decisions distinct is that their primary failure mode is delay. A purchase order that sits in an inbox for eleven days is a project that sits still for eleven days. A vendor's quoted price that expires because the approval came two days late is money lost to a clock the company could not see. A payment release that waits because nobody can tell whether the underlying approval actually happened is a supplier relationship quietly eroding.
Speed and visibility are everything here. The person who submitted the request needs to know where it is. The person whose desk it is sitting on needs to know it is sitting there. The finance team releasing the payment needs to see the approval chain in one place, not reconstruct it from three forwarded emails. Transactional decisions do not need heavy governance. They need a clear queue, a visible path, and a fast cycle.
The danger of transactional decisions is not that they are risky individually. It is that their sheer volume creates an illusion of completeness. A company that has systematised its purchase orders believes it has systematised its decisions. It has not. It has systematised the ones that already had numbers attached. The three other kinds, the ones without a natural threshold or a pre-built tool, are still running on memory and email.
E is for Exception.
An exception is a decision to knowingly bend a rule the company already has. It is the most important letter in STEP, and it is the one companies handle worst.
An out-of-cycle increment above the policy cap. A purchase approved without the required three quotes. A discount extended beyond the standard terms for one specific customer. Access granted outside the normal procedure. A process skipped because the situation was urgent and the process was slow.
Each of these is individually defensible. A competent person looked at the specific case, judged that the rule did not fit, and made a call. That is good judgment. That is what you hire senior people to do. The problem is not the exception.
The problem is that nobody counts them.
When the same rule gets bent a second time, it is still an exception. When it gets bent a third time, it is a pattern. When it gets bent a fourth time without anyone knowing it is four, something fundamental has shifted. The written policy says twelve percent. The actual practice says eighteen. The company is running two versions of reality simultaneously and nobody decided this. Nobody wrote a new policy. Nobody approved the change. It happened by accumulation, one defensible bending at a time, and the company's stated rules and its actual behaviour have quietly divorced.
This is the most dangerous failure mode of any kind of decision: silent precedent. Not a dispute, which is at least visible. Not a delay, which is at least felt. A slow, invisible drift between what the company says it does and what it actually does, discovered only when somebody new arrives, or an auditor asks, or two employees compare notes and find they were treated by different rules for no recoverable reason.
Exceptions need three things that no other kind of decision needs. A link to the rule being bent, so the deviation is explicit rather than implicit. A reason specific to this case, so the basis is visible and not just assumed. And a count. How many times has this rule been excepted this year. That number, surfaced automatically, is the difference between an organisation that grants exceptions with judgment and one that rewrites its policies by accident.
P is for Process.
A process decision changes how future decisions get made. It is the company editing its own operating system.
A new policy. A revision to the approval limits. An update to the competency framework. A change to the procurement rules. A new SOP. A modification to who is allowed to approve what and up to what amount.
Process decisions are the rarest of the four and the most consequential, because every future action inherits them. When you change the competency framework, every promotion decision for the next five years runs through the new version. When you revise the approval limits, every purchase order routes differently from that day forward. When you write a credit policy, every regional manager applies it to every distributor in every conversation.
The failure mode of a process decision is the longest-fuse one of all: the orphaned rule. The policy exists. It is followed. Thousands of choices are made within it every year. And the reasoning behind it is gone.
A competency framework decided in 2019 by a working group of four people, two of whom have left, adapted from a consultant's model nobody can name, is now the basis for every promotion in the company. A manager reads out the criteria. A person asks why. The manager cannot answer, because the reasoning was never written next to the words.
An approval limit set at two crore by somebody nobody has worked with, in a different revenue environment, at a different scale, is now determining how thousands of purchase decisions are routed. Nobody knows whether two crore was principled or simply a round number. The business has tripled since. The number has not moved.
Process decisions need their reasoning preserved with more care than any other kind, because the gap between the moment they are decided and the moment their reasoning is needed is the longest of any decision type. Sometimes years. And by the time the reasoning is needed, the people who had it are three companies away.
They also need something unique: a measure of weight. How many decisions currently stand on this policy. How many promotions run through this framework. How many purchase orders are routed by this limit. That number tells you how load-bearing the process decision is, and it should be visible before anyone tries to change it. A person updating a competency framework should see that forty-seven promotions were made on the authority of the version they are about to replace. That changes how carefully they think about the revision.
Why STEP matters more than anything else I have written
I have written about Decision Debt, about the half of the lifecycle nobody built, about the seventh hat, about choices versus decisions. All of those are descriptions of the problem. STEP is the beginning of the solution.
Without STEP, recording decisions is a flat exercise. Everything goes into the same register. The reorganisation sits next to the stationery purchase. The exception slips through with no link to the rule it bent. The policy change gets the same treatment as a travel reimbursement. And the system, within three months, is either too heavy for the fast things or too light for the consequential ones, and people abandon it, and the company concludes that decision-keeping does not work.
With STEP, each decision enters the system already knowing what it is. The classification is not a field somebody fills in. It is a property of the category the decision belongs to. When someone selects "out-of-cycle increment," the system already knows this is an exception, and it already knows to link it to the policy being bent, and it already knows to surface the count. When someone selects "competency framework revision," the system already knows this is a process decision, and it already knows to show the weight. The person making the decision does not think about STEP. They pick a category. The intelligence follows.
That is what a chart of accounts does. It disappears into the infrastructure and makes the right thing happen without anybody having to remember the rules.
Part of The Other Books, an ongoing series on the decisions companies forget
