Sales stalls when every deal runs through the same few senior people and nobody can say which numbers in the pipeline are real. Early on those people are you; past that stage they are a VP and the two closers who own the number. The measurement discipline underneath the fix is the same either way, and this page defines it.
The Celerio Method is the body of knowledge underneath GTM Value Engineering. What follows is the whole structure: the premise, the object under measurement, the estimation requirements, the buyer-side inference, the diagnosis, the decisions, and the boundary of what the method claims. What the page does not carry, because it cannot, is any firm's estimates. Those are built from your own evidence, and they are the one asset a competitor cannot copy.
The premise
A revenue organisation is a stochastic process. Deals move through states at rates. Buyers move through their own states, at their own rates. Both sets of rates can be estimated.
Everything else follows from that. Your deals sit in states, transitions between those states happen at rates, and the rates are properties of your firm selling your product to your market. They can be measured, and mostly they are not.
The profession borrows its rates from a canon calibrated for other companies in a different market, and reads its states from what the seller typed into the CRM. Both are the same mistake: a number nobody checked against your own evidence. A standard pipeline review makes both at once, multiplying a stage probability off a training deck by a stage a seller ticked in the CRM.
The states are borrowed too. The inherited canon assumes one process: find the buying apparatus and qualify against it. Some markets have no such apparatus yet, because the category is still forming. Run the qualification playbook there and the deals you called qualified start dying. So the method starts with a diagnosis of which market you are in, before it estimates anything.
The object
The object under measurement is the deal's chain of states, treated as the state space of a process. A state has to be defined by evidence of a transition having occurred, never by a seller's claim that it has. A stage name in the CRM is only a claim that a transition happened. Proof is the buyer coming back with their own business case and their own numbers, worked inside your structure.
Between each pair of adjacent states you have three quantities to estimate, not one. The conversion coefficient is the probability of moving from one state to the next. The velocity coefficient is the time spent in a state before moving, held as a cumulative-conversion distribution rather than a mean, because deals do not age uniformly.
The value coefficient is the size of the deal that results, heavy-tailed and held as a distribution conditioned on segment, because when three deals carry the quarter, the average deal size tells you nothing useful. A model that estimates conversion alone, or that treats value as a single mean, is underspecified.
And you are running two processes, not one. The seller-side funnel is what your organisation does, visible in your own records.
The buyer-side decision is how belief forms inside the buying group, and it is not visible in your records at all. The profession models the first and assumes the second. The method models both, by different means: the seller process by direct estimation from outcomes, the buyer process by inference from signal.
The four estimation requirements
Your seller-side estimation is governed by four requirements. They look like statistics. Each one is a decision about what your firm will count as a real number.
First, your own evidence, not inherited canon. Your own won-and-lost record is the evidence. An inherited rate, the figure carried in from a previous company or a training methodology, is admissible only as a prior: a starting belief to be updated, never your parameter. Treating a borrowed constant as a measured fact is the error the method exists to correct.
Second, the right level, with pooling.
Estimate at the finest level the evidence supports, and pool toward the coarser level in proportion to how thin the evidence is.
The new joiner makes it concrete. A seller who has closed three deals has no reliable individual rate; the raw ratio is noise, and ranking them on it is unfair and tells you nothing.
So you start them at their team or segment rate and move towards their own number as they close more deals. With three deals the estimate sits close to the segment prior; as outcomes accumulate it migrates towards the individual's own record. The shift happens when the data warrants it, not when a manager decides.
Where even this is too thin, substitute leading indicators, the activity-quality signals that reliably precede outcomes, as a proxy for the missing outcome data, and always report the result as an interval, never a point. This is the part the inherited canon cannot do at all: for the thin book it can only wait for more data or make a number up.
Third, source signal, not self-report. You read states and transitions from signal captured at the moment of the work, buyer behaviour and buyer language, not from the seller's reported stage. Self-report is lossy and biased. It records the flattering version, and it gets overwritten instead of kept, so it decays. An estimate built on it inherits all that error; capturing at source removes it.
Fourth, honest uncertainty. A coefficient is a distribution, not a number. Thirty per cent off eight deals and thirty per cent off eight hundred are not the same number, even though they read the same, and the method carries the difference as a credible interval and will not report a thin estimate as if it were a fact.
Never state a rate more precisely than the evidence allows.
This is the discipline the inherited canon lacks, which is why it states every rate as a constant and far more precisely than it can.
There is a fifth case, the cold start, and it gets its own machinery because pooling carries a hidden assumption: that the new context is exchangeable with the contexts being pooled from. A new market or a new product may not be.
The treatment is to predict from features rather than from history, forecast by analogy against a survivorship-corrected reference class, admit external evidence only as a weak and wide prior, treat the entry itself as a designed experiment, and test for concept shift before transferring any coefficient across. Copy a coefficient across a concept shift and you get negative transfer: the borrowed number does worse than estimating locally from scratch.
The buyer side
The buyer-side process cannot be estimated the same way, because its causes are invisible. The source of a buyer's frame, the meeting, the analyst report, the prior employer that taught them how to think about the problem, cannot be observed. What you can observe is what shows up in their language: the vocabulary and the criteria that show where they learned to think about the problem, and whose playbook that was.
You estimate two things from it. The first is the buyer's position in their own decision process. The second is the provenance of their frame, whose criteria they have absorbed, which you can read from whose language they speak before they tell you who else they are evaluating. The output is a probabilistic read, updated as signal accumulates, held as evidence and not proof. A buyer using your language is evidence, not proof. The words are easy to copy. The criteria underneath them are not.
So you weight behaviour over stated words throughout.
The mapping from observable signal to buyer meaning is itself local. The buyer side needs not only re-estimated numbers but a re-learned mapping, held as a codebook, re-learned per market maturity and per culture.
The instrument reads a small family of engagement signals: depth, energy, breadth across the buying group, seniority, early and self-initiated engagement, and sustained engagement. The direction of each signal holds across contexts; the baseline is local; so you normalise every reading as a deviation from a rolling, like-for-like cohort baseline.
Three rules for reading it: separate leading signals from lagging ones, quarantine the signals whose sign flips across contexts, and read the trend rather than the level.
Is this deal real, and can I move it? That stops being a checklist question and becomes an estimation problem, the same kind as the seller-side coefficients, run on different evidence.
Compound or decay
The method is self-improving, but only conditionally. Signal captured at source and kept makes every later estimate better: more evidence, finer levels, tighter intervals, earlier buyer inference. Signal discarded at capture, self-reported, overwritten, flattened to a status field, makes the estimate decay, and you repeat the same errors and grow more confident in them.
A system that keeps its evidence gets more accurate the more you use it. One that discards its evidence just adds volume and learns nothing.
Which side of that line your revenue system sits on is the one thing that tells you whether your measurement improves or degrades over time.
The two motions and the diagnosis
Everything so far assumes there is a process to estimate. Whether there is one depends on your market, and this is where the method meets the established canons directly. It recognises two process structures, each with an established practitioner motion, and the two are complements rather than rivals.
The qualification structure runs the MEDDIC motion. Established category, demand exists, the buying apparatus is real. The task is to find that apparatus and to qualify and progress against it, and MEDDIC remains the right instrument for that task.
The creation structure runs the Challenger motion. Nascent category, demand must be made, and the apparatus the canon looks for does not yet exist. A category with no name has no budget line. The criteria to qualify against are still being formed. Often there is no single economic buyer. The unit of work is moving a frame, not qualifying a deal.
Its states are awareness, criteria-shaping, category establishment, and then a hand-off into qualification once demand exists. You cannot measure the creation structure by inverting a funnel, because the funnel has not formed yet, so there is nothing to invert. You measure it by frame-movement leading indicators and reference-class projection instead.
The profession already owns both motions; it lacks the diagnosis that tells it which one you are in.
Supplying that diagnosis is what the method adds. It locates your context, a cell, on two axes: how mature the category is, and how much local data you have. The structure follows from where the cell sits, and entry into a new cell is an estimation mode, not a third structure. Run qualification in a creation cell and the failure is characteristic: every deal looks qualified, then dies anyway, a pattern we have written up in the deal that was perfectly qualified and still died.
So the method is a demand-native successor to the qualification canon, not an attack on it. Where demand has to be made before it can be qualified, qualification gives way to modelling your influence over the buyer's frame, and you pick MEDDIC back up the moment the funnel exists. The canon's vocabulary has meanwhile become common property; in April 2026 a US federal court ruled the MEDDPICC mark generic in MEDDICC Ltd v. 01 Consulting LLC.
Teach any method widely enough and its vocabulary becomes common property, which is why your durable asset is measurement against your own evidence rather than the vocabulary itself.
The decision architecture
An estimate is worth nothing until it changes a decision. The method packages their use as eight operating decisions, D1 to D8, each a contract: inputs, knowledge, the estimate consumed, an output, and how you know it was done well. Each decision is made explicit, priced, and instrumented. You still make the call. And the decisions compose: the output of one is an input to the next, and together they run end to end. This is why the engineering underneath instruments the decisions rather than the workflows.
D1 selects the structure and the estimation mode: diagnose the cell before estimating anything, choosing qualification or creation and direct or cold-start estimation, because a wrong choice surfaces later as systematic mis-estimation, not noise.
D2 qualifies a deal in, out, or to self-serve or partner, using the cohort-normalised engagement read and the segment conversion coefficient; a good qualifier shows realised separation between the in and out groups.
D3 allocates selling time across the live portfolio, deciding where each selling-day goes and which opportunities are de-staffed, judged by yield, revenue per selling-day, against the prior allocation.
D4 sizes and shapes territory, allocating accounts and quota per seller and segment to the heavy-tailed concentration of value rather than to an average, judged by attainment dispersion and tail-value capture.
D5 sets capacity and hiring milestones, converting the inverted funnel and the deals-required range into headcount and timing tied to data triggers rather than the calendar, judged by whether productive capacity arrived where and when it was needed.
D6 prices a process intervention: change a coefficient, hold the constraint, re-solve the funnel, and state the expected revenue delta with its uncertainty and its seam, letting the realised delta re-estimate the intervention's true size.
D7 reads pipeline health and forecast, comparing the live pipeline, engagement trends, and ageing against the estimated process and baselines to produce health flags, an early-warning list of decaying accounts, and a forecast stated as a range, judged by accuracy and warning lead time.
D8 senses demand and market-level buying, aggregating the same instrument across the market to read the demand state and, under creation, whether the category is forming, judged by whether sensed demand preceded realised pipeline.
Deal-level inference and market-level sensing are kept distinct on purpose. The instrument is the same; what changes is whether you point it at a single deal or the whole market.
The derivations
Before the decisions were formalised, the method's services were stated as derivations, and they still show most cleanly that it is a superset rather than a single offering: each service is the same estimation, aimed at a different question. Capacity modelling forward-projects the estimated process against a bookings target to size the activity, pipeline coverage, and headcount you need, using all three coefficient types rather than conversion alone.
Territory planning allocates accounts and quota by the estimated process per segment and seller, so you replace equal division and inherited assumption with your own measured rates.
Pipeline health compares your live pipeline's composition and transition behaviour against the estimated process and flags where it deviates, the early warning the self-reported funnel cannot give you, because deviation from a re-derived baseline means something where deviation from an inherited one does not.
Individual conversion measurement runs the same per-seller estimation with its shrinkage treatment and reported uncertainty, and it works for new joiners and low-volume books because it pools the thin evidence instead of faking a number from it.
Buying inference is the buyer-side estimation applied in the live deal and aggregated across the market, so you can read where demand and competitive frames are forming.
One general case sits behind all five: any question of the form what is the rate, at what level, from what evidence, and with what confidence is answerable inside the same framework. A new service is a new instance of that question, not a new method, and that is what lets the method carry more than one offering without splintering.
The formal substrate
Everything above is one formal object at work: a partially observable Markov decision process. The correspondence is exact, and it maps onto what you already do. The hidden state is the deal's true state, the meaning the codebook exists to recover. The action is one of the eight decisions you make.
The observation is the engagement signal family, and the observation model is the codebook, re-learned per cell, which is where the partial observability lives.
The belief is the normalised engagement read you hold on the deal. The transition model is the conversion and velocity coefficients. The reward is the value coefficient net of selling-days, so when you maximise it you are maximising revenue per unit of your binding constraint. The policy is the decision architecture, and the loop you run is: observe, update belief, act, transition, observe again.
You cannot solve a process like this exactly, so the decision architecture approximates it on purpose, running belief-threshold policies and only looking one step ahead at the value of more information. The formalism is there to explain why the method works, not to be solved in the field. The environment reacts to you and keeps shifting, which is why you roll every baseline forward rather than fix it. And the models are where most of your uncertainty is, not the plan, so most of the work goes into learning the models, not into planning against them.
Fulcrum is the engine you run the decisions on. Its atomic unit, the nucleus, is one instrumented operating decision, and its parts sit directly on the formal object. The context contract holds your belief plus the local models; the gates are the thresholds your policy fires on; the journal is the loop that re-learns the models as they drift, so they stay current; the output schema is the action you take; and the grader is the reward you score it against. Across nuclei you prioritise by the value of information, spending your binding constraint, your own attention, on the decisions where making them well is worth the most.
That is the problem the attention supply chain names: the path your scarce attention travels to become paying customers, run as one engineered system.
The boundary
The method is descriptive and estimative. It tells you what your rates are, at what level you can know them, and how confident you are entitled to be. The estimation layer describes; the decision layer supports it.
Neither makes the call for you, and it will not hide a recommendation inside an estimate. The method does not design your pricing, packaging, compensation, product, or brand; it consumes their outputs and prices their consequences.
Its claims are bounded the same way its coefficients are. What holds across scale are the structures and the machinery; the coefficients and the thresholds are local, re-estimated in each cell and over time.
The evidence behind the method today is two worked cells, one seller-side in observability software and one buyer-side in enterprise-AI software, plus a third, less clean, creation-structure engagement. That amounts to expert replication on a sound mechanism: strong, buildable, and to be proven in the running system rather than asserted.
Every counterfactual it prices carries its seam and is offered as a hypothesis carrying a number, not a law. The whole framework is a working model, and we hold it only as far as the evidence goes: never state a rate more precisely than the evidence allows. The machinery stays fixed while the coefficients and thresholds are re-estimated locally, and being clear about which is fixed and which is local is what separates the method from the canon it replaces.
Why it is published openly
It is published in full because the method is not the moat. This page will change as the evidence does; a measurement discipline that hid its own workings would not be worth trusting.
What you cannot lift from this page is the derived layer. Those assets exist only where the work is done: the coefficients estimated from your own won-and-lost record, the codebook re-learned for your market, the rolling baselines your cohorts define, the thresholds your decisions run on, and the practice of running D1 to D8 every week as things change. We publish the method. What you estimate from your own evidence is the part nobody can copy.
Two worked cells and a third, less clean engagement are what stand behind the method today, and we hold them as expert replication on a sound mechanism, to be proven in your running system rather than asserted on this page. The structures and the machinery are what we stand behind across firms; the coefficients and thresholds we re-estimate on your book before we claim any of them for you.