Our approach

Everyone manages metrics. We manage decisions.

A metric is a sensor: it tells you what happened. A decision is the only thing in a business that costs money, accumulates as debt and stays in memory — so it's the only thing you can actually manage. This is the method we built around that.

Pick your first step Jump to the Decision Loop

Executive summary

Most companies look at the dashboard and do one of two things. Nothing — that's reporting. Or they push on the metric itself — and the metric stops meaning anything.

We do a third thing. We manage the decisions the metrics should lead to.

How fast a company grows is how fast it decides. There is a limited number of decisions you can make in a year; every postponed one is subtracted, with interest. So we built a framework in which a decision can't be postponed without a date, can't be made without being written down, and can't grow big by waiting.

It has five principles, nine entities, five types of signal, a seven-step loop, four meetings and three roles. And one number — Decision Age — that tells you how your decision-making compares with a company of your size.

Two lines we'll stand behind:

Big decisions are a symptom, not a virtue.

A decision without a record is a decision you'll make again.

What we see — and what we do

Most companies compared with Data Never Lies
Most companies	Data Never Lies
Manage metrics	Manage decisions
Learn to make big decisions better	Build a system in which big decisions don't arise
Collect data to decide "properly"	Decide on what's there, with a threshold set in advance and a review date
React only to pain (a metric fell)	See five types of signal: threat, opportunity, silence, external, debt
A decision lives in the head of whoever made it	A decision is recorded: signal, threshold, choice, date, precedent
Decision experience leaves with people	It stays in the company
Post-mortems look for who's to blame	Reviews recalibrate thresholds — no names
Pride in complex strategic decisions	Forced big decisions treated as a process defect

The diagnosis we make is Decision Debt. The number we measure it with is Decision Age. The method we pay it down with is the Decision Loop.

Five principles

Principles are the things you can't break without breaking the framework. Each is written so that breaking it shows up in the data.

A decision is never postponed. Three outcomes: do / don't / wait until a date. "Wait" without a date and a condition ("we'll decide when X, but no later than D") isn't a decision — it's debt. A company with a long queue of undated decisions has a bad process, whatever the quality of the decisions themselves.
Decide on the data you have now. In our experience, a preliminary decision on today's data matches the final one about 85% of the time. Waiting for complete data mostly buys confirmation of what you already know, at the price of time. The review date insures the rest.
Many small decisions instead of rare big ones. The error rate per decision is roughly the same at any level. The only lever is the size of the bet. A hundred £1k decisions and one £100k decision carry the same expected loss — but the first gives you smooth losses you learn from, the second gives you one data point you can't. Big decisions you chose the date for belong in the calendar, and there are few of them. Big decisions the queue forced on you shouldn't exist.
A decision is a record, not an event. Without the signal, threshold, choice and date written down, a decision doesn't exist for the organisation: it can't be found, repeated or challenged. Decision experience belongs to the company, not the employee.
The system learns, not the person. After every verdict three things change: metric corridors, hypothesis thresholds and the precedent base. Reviews don't look for culprits; they calibrate thresholds. Before any new decision, step zero is a search for a precedent.

A consequence of 1 and 3: if your company often has to make big tactical decisions with high stakes, the system is built wrong. That's not resilience to be proud of — it's a signal of debt.

Nine entities

The framework is made of nine entities. Each has mandatory fields, and an entity without a mandatory field doesn't enter the system. That is the main mechanism that stops postponing.

The nine entities
Entity	What it is	Mandatory fields	Rule
1. Goal	What the business wants, as a number with a date	number, date, owner, at least one metric	A goal without a number is a wish
2. Metric	The quantity that shows movement towards a goal	data source, frequency, corridor (expected range: lower and upper bound), owner	A metric without a corridor can't give a signal — it's just a report
3. Signal	A metric leaving its corridor, failing to move, or an external event	type (5 types), source, date, deviation, interpretation owner, hypothesis deadline	A signal is a fact, not an opinion. "I think sales are falling" isn't a signal until there's a corridor. A signal without an owner older than 7 days = debt
4. Hypothesis	A testable interpretation of a signal: "if we do X, metric Y changes by Z by date D"	variable, expected effect, success threshold set before starting, sample size, deadline, verdict owner, cost (£ and hours)	Without a threshold set before the start, it's hope, not a hypothesis. Sales and marketing hypotheses are the same entity
5. Decision	A choice on a hypothesis or signal: do / don't / wait until a date	signal, options considered, choice, decision date, review date, precedent, owner, class (operational / tactical / strategic), chosen / forced for big ones	There is no fourth outcome. "Wait" only with a date and a condition. Step zero: search the ledger
6. Initiative	The container for executing a decision, tied to a goal	goal, thesis, kill criteria (set at creation), owner, status	One active bet per owner. No kill criteria, no initiative
7. Task	An atomic action	owner, date, the initiative or decision it belongs to	A task with no decision above it is noise
8. Verdict	The result of a hypothesis against its threshold	outcome (confirmed / refuted / unclear), actual vs threshold, residue (what we keep), date	Due on the deadline, even if "unclear". Overdue = debt. No names in the review
9. Ledger	The memory of all decisions and verdicts, searchable by metric, signal type, initiative, period	every Decision and Verdict; every change to a corridor or threshold, with date and reason	The only source of "how we decided before". Open to everyone who decides. Never edited after the fact — only added to

Decision class is set by the cost of reversal, not by importance: operational reverses in a day; tactical in a quarter; strategic takes longer. The thresholds are calibrated to the company.

Initiatives and tasks are execution entities. Your project tools (Jira, Asana, Planner) already hold them. The framework adds the link that's usually missing: task → decision → signal.

Five types of signal

Most companies see one type of signal: pain. Opportunities, silence and debt pass by, because there's no corridor to compare against.

Five types of signal
Signal	Definition	Who usually sees it	Leads to	Example
Threat	Metric below the lower bound	Everyone — the only one people react to	Fix / stop	Project margin falls below 12%
Opportunity	Metric above the upper bound	Almost nobody — a positive deviation gets called luck	Repeat / scale. This is where growth lives	Tender-to-contract conversion twice the norm in one region
Silence	Metric didn't move when the hypothesis said it should	Nobody — the absence of an event isn't in any report	Kill or reformulate the hypothesis	New sales channel launched; six weeks later the pipeline is unchanged
External	An event without a metric: market, client, competitor, regulator, a key person leaving	The leader, in their head	Create a metric and corridor, then a hypothesis	A key client replaces its head of procurement
Debt	A signal about the process itself: undated "wait", overdue verdict, ownerless signal	Nobody — a process doesn't report on itself	Decide the postponed decision now	14 decisions in the queue with no review date

A positive signal is still a deviation. If a metric is twice its corridor, that's not "well done", it's a question: what changed, and can we repeat it? Companies that don't review their successes can't reproduce them.

Silence is the most expensive signal. An initiative with no effect keeps spending money and attention until someone notices the absence of a result. A corridor for the expected effect, with a date, turns silence into a signal automatically.

The debt signal is what separates the framework from any dashboard. A dashboard shows the business. The framework also shows the state of the decision process itself.

The Decision Loop

Seven steps. It isn't linear: a signal has three entrances, and memory feeds back twice.

Goals → Metrics. Each goal gets two or three metrics with corridors. Done once at the start and every quarter at Goal review.
Metrics → Signals. A metric leaves its corridor or doesn't move — a signal appears with an owner and a hypothesis deadline. Two more entrances: an external event (gets a metric and corridor first) and a process signal (goes straight to the top of the decision meeting agenda).
Signals → Hypotheses. The owner writes a testable hypothesis with a threshold set before starting. Step zero: search the ledger — if something similar was decided before, its verdict is the first option on the table.
Hypotheses → Decisions. At the decision meeting: do / don't / wait until a date. Recorded at the moment it's made.
Decisions → Actions. The decision becomes an initiative or tasks with owners and dates.
Actions → Verdicts. On the hypothesis deadline: a verdict against the threshold, and a residue — even if the result is "unclear".
Verdicts → Memory → Signals. Verdict and residue go into the ledger; corridors and thresholds are recalibrated; the next signal is caught on the new corridors.

Three invariants. If any one is broken, the process is broken — however good the individual decisions:

a signal without an interpretation owner older than 7 days
a "wait" decision without a review date
a hypothesis past its deadline without a verdict

These three are Decision Debt in measurable form.

Four meetings, three roles

A meeting exists to move an entity one step round the loop. There are no meetings for information — the ledger and the dashboard do that. Each meeting has an exit rule it can't end without.

The four meetings
Meeting	Rhythm	In	Out	Exit rule	Who
Signal review	Weekly, 15 min	Metrics against corridors; external events of the week	Signals, each with an owner and a hypothesis date	No signal without an owner	Signal owner, function heads
Decision meeting	Weekly, 45 min	Hypotheses with thresholds; the queue of postponed decisions — always item one	Decisions: do / don't / wait until a date	Every item gets one of three outcomes; the queue stays on screen	Decision owners, Ledger keeper
Verdict review	Every two weeks, 45 min	Hypotheses past their deadline	Verdicts, residues, changes to corridors and thresholds	No names; zero overdue verdicts afterwards	Verdict owners, Signal owner, Ledger keeper
Goal review	Quarterly, 2 hours	Goals and initiatives; kill criteria; debt counters; share of forced big decisions; Decision Age	Initiatives closed or continued; goals revised; the chosen big decisions for the quarter	Big decisions are taken here and only here	Leadership, Ledger keeper

About an hour and a half a week per function, plus a quarterly half-day. Usually less than a company spends today on status meetings that decide nothing.

Three roles — roles, not job titles; one person can hold two:

The three roles
Role	Responsible for	Usually	DNL can cover it
Signal owner	Corridors; signal detection; data for every meeting	Data team, analyst; in small companies the CFO	Yes — an analyst or data engineer from our team
Decision owner	Decisions in their function; review dates; choice of precedent	Head of the function	No — always your person
Ledger keeper	Completeness of records; process signals (debt); recalibration after verdicts; agendas	COO, chief of staff; in small companies the CEO	Yes — a fractional role we run in the first months until you take it over

Sales and marketing don't get a separate rhythm: their hypotheses live in the same decision meeting as a separate view. Two rhythms are two frameworks, and one of them dies.

Decision Debt: three counters

Decision Debt is the sum of everything postponed in the decision process. Like technical debt, it charges interest: an operational decision not taken on time becomes tactical in weeks and strategic in months. The cost grows; the error rate per decision stays the same.

Decision Debt counters
Counter	What we count	How	Target	Where it shows
Backlog	Decisions in "wait" with no review date, or past it	count × average age in days	0 undated; average age with a date ≤ 14 days	First item of every decision meeting
Orphans	Signals without an interpretation owner, older than 7 days	count	0	Signal review
Overdue verdicts	Hypotheses past their deadline with no verdict	count × days overdue	0	Verdict review

A fourth indicator is a consequence, not a counter: the share of forced big decisions in a quarter. Target: zero. Above 30%, the company is run by its queue of undecided things, not by its leaders.

How debt compounds. An unresolved question about a drawing approval (operational, £0 to reverse) → two weeks later a schedule slip and a decision to resequence the works (tactical, £20k) → two months later a disputed variation (strategic, £200k and lawyers). The mistake in every link is the same: nobody decided. The debt just accrued interest.

What we do with debt at the start. We don't work through the whole queue. At the first decision meeting every postponed decision gets either an outcome or a date. One hour, and the undated backlog is zero. After that the system doesn't let it grow, because "wait without a date" doesn't exist as a status.

Decision Age

One number, twenty minutes, calculated from your last ten significant decisions.

You name the ten. For each, three yes/no questions:

Was it made within a week of the signal? → Speed
Can you show, in writing, why you decided and what came of it? → Memory
Did you choose the date — or did circumstances? → Size

A fourth question, outside the score: how many of the ten did you later reverse? It checks principle 2 against your own data, and it's usually the strongest moment of the conversation.

The three shares multiply, not add. Zero on any one zeroes the result: fast but unrecorded decisions score zero, on purpose. The raw score is compared with the framework's reference and expressed as an age — "your decisions are made at the level of a company a fraction of your size." The three sub-scores matter more than the total: "overall fine, but Memory is low" is already a diagnosis and a plan.

The three multipliers map one-to-one onto principles 1, 4 and 3: don't postpone, remember, don't let it get big.

We measure Decision Age in the first free session, and again after the framework is installed. Data Therapy

The ledger: how the system learns

The ledger is what the framework learns with. Not the person getting more experienced — the system: corridors, thresholds and precedents. It belongs to the company and doesn't leave with an employee.

One row per decision, written at the moment of deciding, never after the fact:

One ledger row, with an example
Field	Content	Example
Date	When decided	2026-10-14
Signal	Link to the signal and its type	Threat: project margin 9% against a corridor of 12–18%
Class	Operational / tactical / strategic	Tactical
Chosen / forced	For tactical and strategic only	Forced (the signal sat for 5 weeks)
Options	What was considered, 2–4 lines	1) negotiate a variation; 2) replace the subcontractor; 3) accept the loss
Choice	What was decided and why, 1–2 sentences	1) — grounds under clause 4.2 of the contract
Threshold	What counts as success, by when	Variation agreed at ≥ £30k by 2026-11-30
Review date	When we come back if the threshold isn't met	2026-11-30
Precedent	A similar past decision, or "none"	Decision #41 (2025) — option 1 worked in 6 weeks
Owner	Who answers for the verdict	Commercial director
Cost	£ and leadership hours to execute	£4k, 12 h

The verdict is added to the same row: outcome, actual against threshold, and the residue — what we now know that we didn't.

Three learning mechanisms.

Precedent search — step zero of any decision. Before discussing options, the decision owner searches the ledger by metric and signal type. If a precedent exists, its verdict is the first option on the table. This removes most repeat decisions and halves the decision meeting.
Recalibration. At every verdict review, each verdict answers "was the threshold right?" A threshold confirmed too often is too soft; one refuted always is unrealistic. Every change is recorded with a date and a reason.
Reversal rate. Once a quarter: how many decisions changed after review. Below 20%, you're waiting for data longer than you need — decide faster. Above 40%, thresholds are badly set or signals are false.

The residue is the most important field. A refuted hypothesis without a residue is lost money; with a residue it's bought knowledge. One sentence: "what we now know that we didn't before this decision."

The ledger is a table, not a system. At the start, one table in a tool you already have is enough. It's also the context that makes an AI assistant useful: a company's decision history is what lets AI answer about your business instead of in generalities.

What the framework doesn't do

It doesn't make decisions right. It makes them faster, smaller and recorded, and it makes the system learn from each one — but whether a decision is right is still your judgement.

It doesn't predict. Corridors and thresholds, not forecasts. A forecast can be one source of a corridor, not a replacement for the loop.

It doesn't replace the leader. The decision owner is always your person.

Where it comes from

We use this framework to run Data Never Lies: one ledger, hypotheses with thresholds set before the start, kill criteria on every initiative, verdicts with residues. The practice behind it comes from our Data Therapy work with leaders and leadership teams, and from building the data systems that corridors and signals depend on.

The eight-week installation The weekly sessions that work on your decisions directly

Glossary

Glossary
Term	One line
Decision Intelligence Framework	Our method: nine entities, a seven-step loop, four meetings
The Decision Loop	Goals → Signals → Hypotheses → Decisions → Actions → Verdicts → Memory → Signals
Decision Debt	The sum of everything postponed; charges interest; three counters
Decision Age	Your decision-making compared with a company of your size, from ten decisions and three questions
Corridor	The expected range of a metric: lower and upper bound
Signal	Leaving the corridor, silence, an external event, or debt
Threat / Opportunity / Silence / External / Debt	The five types of signal
Hypothesis	"If X, then Y changes by Z by D", with a threshold set before the start
Threshold	What counts as success, set before the test begins
Decision	Do / don't / wait until a date
Chosen / forced decision	Who set the date: you, or the queue of the undecided
Verdict	Confirmed / refuted / unclear, plus a residue
Residue	What we now know that we didn't before the decision
Ledger	The searchable memory of all decisions and verdicts
Precedent	A similar past decision from the ledger; step zero
Backlog / Orphans / Overdue verdicts	The three Decision Debt counters
Reversal rate	How many decisions changed after review
Signal owner / Decision owner / Ledger keeper	The three roles
Signal review / Decision meeting / Verdict review / Goal review	The four meetings
Operational / Tactical / Strategic	Decision class, by cost of reversal

What's your Decision Age?

Pick your first step

The first session is free. Ten decisions, three questions each, one page back.