The pillar · v3.2 · 2026-05-31

The GTM World Model

The unified theory of go-to-market, for systems thinkers.

Go-to-market is a causal identity wrapped in a correlational membrane. This model formalizes it across 6 tiers and 27 state variables, with 26 equations separating what is true by identity from what is causal-but-regime-dependent and what is merely correlational — so an agentic GTM system can act on it without confusing a law for a coincidence.

v3.2 Last updated 2026-07-05 32 theses · 27 failure modes · 219 terms

The tiers

GTM runs at different tempos. The model separates them so a fast loop never overwrites a slow, human-owned decision.

Tier	Tempo	What it decides
Pre-funnel / Buyer-state & Brand tier (where the outcome is largely decided)	slow (brand: quarters-years) to medium (buyer-state: weeks-months); mostly INVISIBLE to the seller	Whether you are even in contention before a seller engages. Encodes the empirical reality that ~95% of B2B buyers are out-of-market in any quarter (LinkedIn 95-5 Rule) and that winners are overwhelmingly on the buyer's Day-1 shortlist (6sense: ~85-95%, first seller contact at ~61% of journey in 2025). The seller funnel (Tier 1) is DOWNSTREAM of and CONFOUNDED by this tier.
Strategy / Configuration tier (slow-moving, human-owned)	quarters to years	WHO you sell to, WHAT you claim, HOW you sell, WHERE you reach them, and WHETHER the product earns demand. These are largely FORCED MOVES given the economics, not free preferences.
Unit-Economics ESTIMATOR tier (model-dependent; only the MRR walk is exact)	continuous; reviewed monthly/quarterly	Whether the Tier-3 configuration is VIABLE and SOLVENT. CRITICAL v2.0 CORRECTION: this tier is NOT 'mostly identities'. The ONLY exact identity is the MRR walk (MRR_t = MRR_{t-1} + New + Expansion - Contraction - Churn). Every other metric -- LTV, NRR, GRR, CAC-payback, Magic Number, Rule-of-40 -- is an ESTIMATOR that summarizes the walk under contestable assumptions. v1.1 mislabeled these as identities, which let the model claim more certainty than its primitives support and propagated upward into a circular T7.
Execution / Behavioral tier (fast, increasingly agent-owned)	per-lead, per-account, daily/weekly	The actual per-lead and per-account loop: detect signal -> decide whether to act -> gather context -> qualify -> draft/personalize -> route -> hand off -> follow up -> expand/retain. This is where the behavioral coefficients (conversion rates, response, reply rates) actually get produced, and where they DRIFT.
Operations & Systems substrate (the rails all tiers run on)	always-on
Macro regime scalar Psi (sits ABOVE Tier 2; re-weights the objective)	slow (shifts with the rate/capital cycle over quarters-years)	WHICH weighted combination of growth and free cash flow the market prices. Not a Tier-2 input coefficient -- it changes the OBJECTIVE FUNCTION itself, which is why the latent PMF multiplier Phi cannot absorb it. Resolves T3 (payback-vs-ratio) as REGIME-CONDITIONAL rather than a fixed law.

State variables

Symbol	Name	Type	What it is
C	Customer/revenue stock	stock	The central stock the whole system grows or leaks. Everything is a flow into or out of C.
a	Acquisition inflow rate	flow_in	New customers per period. A LINEAR lever on the steady state.
delta	Churn rate	flow_out	Fraction of C lost per period. Sits in the DENOMINATOR of steady state, hence nonlinear.
g	Expansion rate	flow_modifier	Revenue growth within existing C. Net retention = 1 + g - delta.
V	Top-of-funnel volume	flow_source	Raw demand entering the funnel.
r_i	Stage conversion rates	coefficient	The behavioral coefficients. THIS is what an execution agent like LangChain's moves (lead->oppty up 250%). They DRIFT and saturate. REFINEMENT (v1.1): r_i is not one number but a DISTRIBUTION across reps; agentic augmentation compresses its variance (raises the floor) rather than only shifting the mean — Coefficient Compression.
p_bar	Average price / ACV	coefficient	Set in Tier 3. Determines which MOTION is viable (the threshold).
m	Gross margin	coefficient	Converts revenue to contribution; gates LTV.
S	Sales+marketing spend	control_input	The primary control input. CAC = S/a.
x	Market penetration / saturation	state	Cumulative share of TAM captured. Drives CAC convexity.
Phi	Product-market fit (latent)	latent_multiplier	Unobservable. Inferred from joint pattern of low churn + high win-rate + low CAC + high expansion moving TOGETHER. Enters MULTIPLICATIVELY.
k(tau)	Lag/memory kernel	function	Distributed-lag kernel. Why last-touch attribution structurally misattributes brand/demand-creation.
Psi	Macro regime scalar	exogenous_index	Capital-tightness index (10yr Treasury, Cloud multiples, VC pace) re-weighting the growth/FCF objective. Sits above Tier 2.
B	Buyer-state vector	latent_state	B=[shortlist S_p, in-market A_m, journey-completion J_c, group size G]. Pre-funnel; confounds every Tier-1 conversion coefficient. Latent, observed via intent proxies.
B_r	Brand stock	stock	Nerlove-Arrow stock of mental availability. dB_r/dt = alphaI - deltaB_r. Drives shortlist probability. B2B decay ~10-20%/yr.
S	Switching-cost moat	latent_modifier	Hazard-rate modifier h_eff = h0*exp(-S). High in enterprise/regulated/integrated; ~0 in PLG-SMB. Makes T7 additive when high. Unmeasured publicly; proxied via GRR + low-satisfaction confounder.
k	PLG viral coefficient + loop state	loop_state	u_{t+1}=ku_t; k_eff=k_rawretention. B2B k<1 always (~0.2 excellent); divergent k>=1 case never binds.
alpha	Autonomy level of an agentic execution chain	control_input	alpha in [0,1]; 0 = every step human-reviewed, 1 = fully autonomous. The lever the operator actually sets. New in C6+ agentic extension.
G	Governance / HITL maturity index	exogenous_index	G in [0,1] or staged (ad hoc / defined / disciplined). Operationalizes T10. Enters the model as TWO parameters — gamma_max(G) (catch-rate ceiling) and rho(G) (fatigue-rot) — because a single scalar cannot explain the T29 sign-flip in optimal autonomy.
gamma	Effective error catch-rate of human review	function	gamma(alpha,G) in [0,1]. Fraction of agent errors human review actually catches at autonomy alpha under governance G. Degrades as review LOAD (1-alpha) rises past what triage discipline sustains — the formal 'HITL rubber-stamp rot' failure mode. See E-A2.
p_step	Effective per-step error rate of an agent-executed chain	function	p(alpha,G) in [0,1]. Base atomic error p0 net of catches: p = p0*(1-gamma). See E-A1.
R_chain	Chain reliability across n sequential steps	function	R(alpha,G,n)=[1-p]^n. Bottleneck-theorem analogue to the funnel identity: chain depth n compounds a fixed per-step error into unreliability. Formal reason T11's 'compounding failures' is structural, not merely empirical. See E-A3.
C_a	Agent execution spend	flow_out	Inference, orchestration, tooling. USAGE-scaling, distinct from headcount-scaling S. Operationalizes T11 (FinOps->CAC edge): an agentic motion can look cheaper per rep while CAC rises if C_a scales faster than S falls. See E-A5.
lambda_hitl	HITL-as-data-collection rate	coefficient	lambda in [0,1]. Fraction of reviewed agent actions that convert to labeled training signal for the stage coefficients r_i. Operationalizes T9: HITL as a data-collection mechanism, separate from and additive to its safety-gate role. See E-A6.
theta	Category-wide noise floor	loop_state	theta(kappa), unbounded above. The signal floor an outbound/inbound touch must clear to register with a buyer. A Nerlove-Arrow adstock analogue to brand-stock B_r, but accumulating with aggregate category agent-adoption kappa. Formalizes T23 'agents are leverage then pollution'. See E-A7.
kappa	Aggregate agent-adoption share across the competitive category	exogenous_index	kappa in [0,1]. NOT the firm's own alpha — the category-level driver of theta. A shared-resource variable no single firm's governance G can internalize. Drives the T30 category externality.
B_agent	Agent-evaluator presence within a buying committee	state	B_agent in [0,1]. Share of agent evaluators inside the buying committee. Extension of the existing buyer-state vector B. Flagged correlational until measured (Gap 7). See E-A9.

The equations

Each carries an epistemic type — identity, causal, or correlational — and a plain-English reading. That tag is the whole point: it tells an operator (or an agent) how much to trust the edge.

The conservation law (the spine) identity

\frac{d C}{d t} = a - (δ - g) C

steady state

C^{*} = \frac{a}{δ - g}

Reads as: Customers are a stock with inflow (acquisition) and net outflow (churn minus expansion). The equilibrium size is acquisition over net-churn.

Implication: Acquisition sets the numerator (linear lever); retention sets the denominator (nonlinear lever). As net retention -> 100% (g -> delta), C* diverges with ZERO new acquisition. This is why retention dominates acquisition structurally, not as opinion.

The funnel as a log-linear operator identity

R = V i \prod r_{i} ⟺ ln R = ln V + i \sum ln r_{i} + ln \overset{p}{ˉ}

Reads as: Revenue is volume times the product of stage conversions times price. In log space, contributions are additive.

Implication: BOTTLENECK THEOREM: because stages multiply, the marginal return to fixing the WORST-converting stage dominates improving an already-good one. Doubling a 5% stage doubles throughput; +5pts on a 50% stage adds 10%.

The unit-economics master inequality estimator

L T V > C A C ⟺ \frac{m p ˉ}{δ} > \frac{S}{a} ⟺ \frac{m p ˉ a}{δ S} > 1

Reads as: Lifetime contribution must exceed acquisition cost. Expanding LTV exposes churn in the denominator-of-the-denominator, so retention enters quadratically vs how acquisition enters.

Implication: GTM is CONSTRAINED optimization, not maximization. And the BINDING constraint at venture stage is CAC PAYBACK (cash timing), not the LTV:CAC ratio (equilibrium). A 5:1 ratio can still kill you if LTV arrives over 5 years and CAC is due now.

Channel saturation (CAC convexity) causal_regime

C A C (x) = \frac{C A C _{0}}{1 - ( \frac{x}{T A M} ) ^{β}}

Reads as: As you exhaust a finite market, cheap demand goes first, so CAC rises convexly with cumulative penetration.

Implication: Every channel has a carrying capacity. Optimal strategy is a PORTFOLIO of channels each run only to its rising-marginal-cost knee. 'Be everywhere' pays fixed costs below scale; 'master one channel' rides it up a convex cost curve. Mix must SHIFT over time as channels saturate — which is why GTM strategies have half-lives.

Motion selection as a forced move (step function) causal_regime

human sales viable ⟺ A C V \cdot m \cdot LTV_{mult} > cost of one human sales cycle

Reads as: Cost-to-serve per human touch is ~fixed; revenue scales with ACV. Below a threshold ACV, a salesperson is unprofitable BY ARITHMETIC.

Implication: Motion is not a preference — it is forced by where ACV sits vs cost-to-serve. The 'PLG <5K / hybrid 5K-50K / sales-led >50K' bands are the footprint of this inequality crossing zero. The hybrid zone is where it's near-equality and the answer is 'let the deal self-select' = product-led sales.

Distributed lags and the attribution problem causal_regime

r (t) = \int k (τ) A (t - τ) d τ

Reads as: Observed conversion today is a convolution of PAST activity with a decaying memory kernel.

Implication: Naive last-touch attribution is STRUCTURALLY guaranteed to misattribute: it assigns 100% to the impulse, 0% to the kernel. Brand/demand-creation will always look underperforming to click-attribution, and over-cutting it is a predictable recurring error. Self-reported attribution exists to recover the kernel.

The product-market-fit multiplier (the master confounder) latent_multiplier

R \approx Φ \cdot f (GTM levers) [multiplicative], not R \approx Φ + f (GTM levers)

Reads as: Product-market fit multiplies the effectiveness of every GTM lever simultaneously; it does not add to them.

Implication: Most cross-metric GTM correlations (high NRR <-> efficient CAC, etc.) are CONFOUNDED by Phi driving both. GTM tactics have near-zero causal effect below a Phi threshold and large effect above it. Anything times near-zero Phi is near-zero — the graveyard of well-funded companies with great GTM machinery and no fit. GTM Fit pays off only once PMF clears threshold.

Sales velocity identity identity

Sales Velocity = \frac{n _{opps} \cdot win rate \cdot avg deal size}{sales cycle length}

Reads as: Revenue throughput is opportunities times win-rate times deal-size, divided by how long deals take.

Implication: Four independent levers, and CYCLE LENGTH is in the denominator — compressing the cycle (multi-threading, mutual action plans, faster speed-to-lead) raises velocity as much as winning more. Agents that compress research/drafting time compress cycle length directly.

The MRR walk (the ONE true identity) identity

M R R_{t} = M R R_{t - 1} + New_{t} + Exp_{t} - Contr_{t} - Churn_{t}

Reads as: Ending recurring revenue equals starting plus new plus expansion minus contraction minus churn. Exact on a consistent recognition basis.

Implication: This is the only Tier-2 identity. Every other 'metric' is an estimator summarizing trajectories of this walk under assumptions.

Corrected LTV under heterogeneous churn (Fader-Hardie sBG) estimator

L T V = t \geq 1 \sum A R P A \cdot m \cdot S (t) (1 + d)^{- t}, E [T] = \frac{α + β - 1}{α - 1} (α > 1)

Reads as: Lifetime value integrates discounted margin over a survival curve fit to cohort retention, NOT ARPA*margin/churn. Aggregate retention rises over time because high-churn accounts sort out early (heterogeneity), not because individual churn falls.

Implication: Naive exponential LTV overstates 2-3x on a ~10%-churn cohort. Worked: alpha=2.5, beta=6 -> E[T]=5mo -> LTV $800 vs naive $1600 (2x). Use sBG for contractual/time-invariant churn, Beta-discrete-Weibull (BdW) for non-monotone.

Piecewise revenue: multiplicative PMF (low-S) vs additive switching-cost moat (high-S) causal_regime

R = {Φ f (GTM) Φ f_{1} (GTM) + S f_{2} (installed base) S < S^{*} S \geq S^{*} h_{eff} = h_{0} e^{- S}

Reads as: Where switching costs are low (consumer, PLG-SMB, transactional B2B), revenue is fit AMPLIFIED BY go-to-market: multiplicative. Where switching costs are high (enterprise multi-year, regulated, deeply integrated), a moat term S adds revenue largely INDEPENDENT of product love.

Implication: T7's strong multiplicative form is wrong in ~the most valuable third of enterprise software. Oracle/SAP-class: NPS near/below 0, 30-50% unused seats, yet GRR >95% -- revenue tracks S (lock-in), not Phi. Classical IO (Klemperer; Farrell-Klemperer) is explicitly ADDITIVE: switching-cost rents are a premium ON TOP OF current quality.

Macro regime objective re-weighting (Psi / Rule-of-X) causal_regime

U = w_{g} (Ψ) growth + w_{f} (Ψ) FCF, X = M (Ψ) growth + FCF margin, M \approx 2 - 3

Reads as: The market's objective function is regime-dependent. Psi (a function of 10yr Treasury, Cloud multiples, VC pace) sets the growth weight M. In ZIRP, growth is weighted ~3x and long payback is fine; in high-rate, M falls and payback must shorten.

Implication: Resolves T3: 'payback over ratio' is true ONLY when capital is expensive (M low) or runway < ~2x payback. In ZIRP it was false; for capital-abundant incumbents it is false. Payback is regime-conditional, not a fixed king.

Buyer-state confounds the seller funnel (the de-confounded estimand) conditional_coefficient

Naive (confounded): True estimand: Win = Leads \times Conversion P (win ∣ seller touch, S_{p}, A_{m}, J_{c}, Φ, Ψ, S)

Reads as: The seller funnel is DOWNSTREAM of an invisible buyer-state B = [shortlist position S_p, in-market A_m, journey-completion J_c, buying-group size G]. Funnel conversion coefficients are mixtures over B, not measures of execution.

Implication: THE behavior-changing edge. In v1.1, doubling SDRs doubled pipeline. Here, doubling SDRs into a fixed in-market pool (A_m=1 is only ~5% of accounts) causes GEOMETRIC conversion decay. Stops the model recommending infinite GTM scale-up; shifts capital toward brand (B_r) and PLG. Cold/non-shortlist outbound ~3-5% win vs shortlist/inbound ~30-50%+.

Brand as a stock (Nerlove-Arrow adstock), not a lagged flow causal_regime

\frac{d B _{r}}{d t} = α I (t) - δ B_{r} (t), P (S_{p} = 1 ∣ B_{r}) = σ (λ_{0} + λ_{1} B_{r})

Reads as: Brand is a STOCK that accumulates mental availability across category entry points and decays slowly. It is HOW you get on the Day-1 shortlist. Past investment yields long-term baseline demand; you cannot switch it on quickly.

Implication: Differs from v1.1 T6, which treated demand-creation as a lagged FLOW inside the attribution convolution. Here brand is the stock that GENERATES the flow. With ~95% of buyers out-of-market, a funnel-only model is blind to where mental availability accumulates.

PLG loop (recursive), distinct from the funnel (multiplicative) causal_regime

u_{t + 1} = u_{t} (1 + k - c), pure viral: u_{t + 1} = k u_{t} \to \frac{u _{0}}{1 - k}, k_{eff} = k_{raw} \cdot R

Reads as: Product-led growth is a recursive loop, not a chain of funnel conversions. Closed-form divergence requires viral coefficient k>=1. Retention is the UPSTREAM CAUSE of virality (dead users send no invites), reversing the usual arrow.

Implication: Two hard couplings: (1) expansion-as-acquisition under usage pricing BREAKS NRR's intra-account assumption (cross-team adoption is acquisition wearing a retention label); (2) k_eff = k_raw*R, so Phi (driving retention) is upstream of the loop. Hands off to sales-assist at a usage/MRR threshold, where buyer-state B activates.

Ex-ante PMF -- the testable (but not-yet-runnable) form of T7 latent_multiplier

lo g (growth_{i t}) = b_{0} + b_{1} lo g Φ_{i, t - 1} + b_{2} lo g GTM_{i t} + b_{3} lo g Φ_{i, t - 1} lo g GTM_{i t} + ε

Reads as: To de-circularize T7, fix Phi BEFORE the growth window using a measure that cannot be re-inferred from revenue: cohort-retention floor at week 8-12 (primary), or independent G2/Gartner satisfaction at t-1 (secondary). Then the interaction term b3 settles multiplicative vs additive.

Implication: This is the CORRECT falsifiable core of T7. But the cross-company panel (independent t-1 Phi + subsequent growth + comparable GTM spend) DOES NOT EXIST publicly. Until a private aggregator (Bessemer, SaaS Capital) runs it, T7 remains a LOCALIZED HEURISTIC, not a tested claim.

Irreducible-noise floor (prediction-refusal guardrail) correlational

θ = \frac{Var ( noise )}{Var ( growth )} \approx 0.25 - 0.40

Reads as: ~25-40% of mid-stage B2B SaaS growth-rate variance is unexplainable by any GTM lever (competitor surprise, macro shock, viral inflection, champion exit). The model must output confidence bands and 'prediction refused' states rather than false point estimates.

Implication: Decision rule: refuse LTV before the cohort-retention floor flattens (month 6-12); refuse forecasts spanning a competitor surprise; widen intervals through Psi regime transitions. This marks the BOUNDARY of the model's own predictive power.

E-A1 · Effective per-step error rate under partial autonomy causal_regime

p (α, G) = p_{0} [1 - γ (α, G)]

Reads as: The per-step error an agent chain actually ships is its zero-oversight base rate p0 knocked down by whatever fraction human review catches.

Implication: Autonomy only lowers cost through reliability if gamma is real. Raising alpha raises p unless governance holds gamma up.

E-A2 · Reviewer-fatigue-adjusted catch-rate causal_regime

γ (α, G) = clip [γ_{m a x} (G) (1 - α) - ρ (G) (1 - α)^{2}, 0, γ_{m a x} (G)]

Reads as: Review catches errors up to a governance ceiling gamma_max(G), but catch efficacy DEGRADES as review load (1-alpha) rises past what triage discipline sustains.

Implication: Governance enters as two parameters, not one: gamma_max (ceiling) AND rho (fatigue-rot). Mature governance is defined partly by a LOWER rho — discipline that resists alert fatigue as volume scales. This two-parameter split is what drives the T29 sign flip.

E-A3 · Chain reliability identity

R (α, G, n) = [1 - p (α, G)]^{n}

Reads as: Reliability of an n-step agent chain is the per-step success rate compounded n times.

Implication: Bottleneck-theorem analogue to the funnel identity R=V*prod(r_i)*p_bar: chain DEPTH compounds a fixed per-step error. Autonomy safe on a 3-step task can be unsafe on an 8-step one. This is why T11's compounding-failure claim is a structural property of chain depth.

E-A4 · Effective cost per completed task estimator

C_{eff} (α, G) = n [c_{infer} + c_{human} (1 - α)] + [1 - R (α, G, n)] C_{fail} (1 - κ_{gov} (G))

Reads as: Cost per finished task = execution cost (fixed inference plus human review scaling with load 1-alpha) plus expected failure cost, discounted by the share of failure cost governance actually absorbs.

Implication: kappa_gov(G) is the fraction of failure COST governance absorbs (rollback, incident response) — distinct from gamma, the fraction of raw errors it PREVENTS. The two enter separately. Minimizing C_eff over alpha is the operator's real decision; its argmin flips across the G threshold (see worked example / T29).

E-A5 · Agent execution spend inside CAC estimator

CAC = \frac{S + C _{a}}{a}, C_{a} = tasks \sum n c_{infer} (volume)

Reads as: Split the CAC numerator: headcount-scaling spend S plus usage-scaling agent spend C_a, over acquisitions a.

Implication: v3.1's CAC estimator implicitly buried agent spend inside S. Separating C_a makes T11's FinOps->CAC edge explicit: an agentic motion can cut S per rep while CAC still RISES if C_a scales faster than S falls.

E-A6 · HITL as a learning-rate term on coefficient drift causal_regime

\frac{d r _{i}}{d t} = f_{existing} (r_{i}, \dots) + λ γ (α, G) (target_{i} - r_{i})

Reads as: Stage coefficients r_i drift toward a target at a rate set by lambda (share of reviews that become training signal) times the review volume gamma actually producing catches.

Implication: Makes the 'Coefficient Compression' drift rate ENDOGENOUS. Operationalizes T9: HITL is a data-collection mechanism (this term), separate from and additive to its safety-gate role (E-A2/E-A4).

E-A7 · Category-wide noise floor from aggregate agent adoption causal_regime

\frac{d θ}{d t} = η κ - δ_{θ} θ, θ^{*} = \frac{η κ}{δ _{θ}}

steady state

θ^{*} = \frac{η κ}{δ _{θ}}

Reads as: The buyer's noise floor accumulates with aggregate category agent-adoption kappa and decays slowly — an adstock for outbound saturation.

Implication: Formalizes T23/T30: as kappa rises across the category, theta rises and the marginal GTM alpha any single firm's agent can extract falls. A category-level tragedy of the commons no single vendor's governance G can undo.

E-A8 · Controlled agentic-CAC holdout estimator estimator

Δ CAC_{agentic} = CAC_{agentic} - CAC_{manual}

Reads as: The matched-cohort cost-per-qualified-opportunity difference an agentic motion would have to prove — net of deliverability decay and operator maintenance.

Implication: Written down specifically so its inputs can be named as UNMEASURED. This is the formal shape of the comparison T15 says has no public holdout. Included so a vendor sees exactly what a validating experiment must hold constant.

E-A9 · Buyer-state extension for AI-mediated buying committees correlational

B_{ext} = B \oplus [B_{agent}]

Reads as: Append an agent-evaluator share term to the human buyer-state vector B.

Implication: Restates Gap 7 as a variable: no source tests whether T12's ~95% pre-contact primacy strengthens, weakens, or holds when the buying committee itself includes AI evaluators.

The 32 theses

The contested claims the model takes a position on. Each has its own page with mechanism, evidence, and operator implication. Load-bearing theses carry the structure.

#	Claim	Type
T0	GTM is a causal identity wrapped in a correlational membrane whose coefficients are functions of regime and of a latent PMF multiplier.load-bearing	meta
T1	Acquisition is a linear lever; retention is a nonlinear one (it shrinks the steady-state denominator). A field obsessed with top-of-funnel is optimizing the lower-exponent term.
T2	Bottleneck theorem: because funnel stages multiply, fixing the worst-converting stage dominates improving a good one.
T3	[v2.0 REGIME-CONDITIONAL] CAC payback binds harder than the LTV:CAC ratio ONLY when capital is expensive (Psi tight, M low) or runway < ~2x payback. In ZIRP it was false; for capital-abundant incumbents it is false. Payback is not a fixed king -- it is a function of Psi.
T4	Every channel has a carrying capacity and convex CAC; the optimum is a time-shifting portfolio, which is why GTM strategies have half-lives.
T5	Motion is a forced move set by ACV vs cost-to-serve, not a preference; the PLG/hybrid/sales-led bands are that inequality's footprint.
T6	GTM has distributed lags, so last-touch attribution structurally misattributes and over-cutting brand is a recurring institutional error.
T7	[v2.0 PIECEWISE] PMF is a multiplicative amplifier of GTM ONLY in low-switching-cost segments (R=Phif). In high-switching-cost segments (enterprise, regulated, integrated) revenue is ADDITIVE: R=Phif1 + S*f2, where a switching-cost moat S generates revenue largely independent of product love (Oracle/SAP existence proof). The strong multiplicative form was circular (Phi proxied by the NRR it predicts) and is wrong in ~the most valuable third of enterprise software.load-bearing
T8	An agentic GTM system executes the behavioral tier and is uniquely good at moving conversion coefficients and compressing cycle time — but only within the configuration and constraints set by the human-owned strategy and economics tiers.	architecture
T9	Human-in-the-loop is not just a safety mechanism; it is the data-collection mechanism by which the behavioral coefficients are learned. A GTM model that can't ingest its own correction signal is dead on arrival.	architecture
T10	Every serious agentic-GTM vendor independently converges on the same micro-architecture (Sense-Reason-Act-Learn under human-in-the-loop, governed, on a unified platform) — which validates the three-tier model from the outside: they are all building the SUBSTRATE that lets agents execute Tier 1 while reading Tiers 2-3. The competitive edge has shifted from WHETHER you adopt agents to how DISCIPLINED and GOVERNED the implementation is.load-bearing	architecture
T11	Agents invert software's cost/reliability profile (cheap to build, expensive+unreliable to run, compounding failures), which adds a FinOps->CAC edge to the economics tier and makes human-in-the-loop a structural necessity (error-chain interruption) rather than merely a safety nicety.load-bearing
T12	BUYER-STATE PRIMACY: the seller funnel observes only ~5% of value creation; the outcome is largely determined pre-contact by Day-1 shortlist position. Funnel conversion coefficients are CONDITIONAL (mixtures over buyer-state), not causal. Doubling seller activity into a fixed in-market pool yields geometric decay, not proportional pipeline. This is the single biggest behavioral change from v1.1.load-bearing
T13	BRAND IS A STOCK, NOT A FLOW: mental availability accumulates (Nerlove-Arrow) and is HOW you reach the Day-1 shortlist. Brand stock B_r -> buyer-state B -> outcome is the real 'Tier 0' causal chain. Demand-creation is not a lagged flow inside attribution (T6) but a stock that generates it.load-bearing
T14	PLG IS A LOOP, NOT A FUNNEL, AND k<1 ALWAYS BINDS: B2B virality never sustains k>=1, so the divergent closed form is a non-binding limit; PLG is a highly efficient channel, not perpetual motion. Retention is upstream of virality (k_eff=k_raw*R), and expansion-as-acquisition breaks NRR's intra-account assumption.
T15	[AGENTIC, v2.0 RECALIBRATED] Agentic GTM is technically calibrated but commercially over-claimed. Cost-per-ACTIVITY collapses via model-routing; cost-per-QUALIFIED-OPPORTUNITY net of deliverability decay and 15-20hr/wk operator maintenance is UNPROVEN (no public holdout). The autonomous-AI-SDR thesis is falsified in public (11x collapse; ZoomInfo 'worse than our SDRs'); capital flipped to AE-augmentation (Rox $1.2B by REJECTING SDR-replacement). Hybrid beats autonomous; Tier-3 stays human in all cases.load-bearing
T16	AGENTS WORK THE LEAST-LEVERAGED 5%: because ~95% of value creation is pre-contact (buyer-state + brand), agents deployed almost entirely in the seller funnel optimize the visible tail. BEAR: they polish the harvested ~5% while brand/shortlist go unaddressed, degrading coefficients via signal-exhaustion. BULL: agents are the only scalable way to nurture the 95% out-of-market into future shortlist position. Which dominates is unproven; today's tooling makes the bear case empirically closer.load-bearing
T17	MAXIMAL-BUT-TAGGED: a complete map is only safe if its legend is honest. v3.0's coverage is bought with falsifiability; the measurement_status tag on every construct is the sole thing preventing the model from explaining any outcome after the fact.load-bearing
T18	REALIZED < DESIGNED: realized growth is min(GTM physics, organizational alignment bandwidth). In most firms the binding constraint is alignment bandwidth, not lead quality — so the org_cognition layer often dominates the equation layer.load-bearing
T19	REFLEXIVITY, NOT DRIFT: a tactic's coefficient is highest before imitation and decays toward the market mean as it is copied. The 'half-life of a GTM strategy' is an endogenous reflexive effect, measurable as effect-size decay, not exogenous noise.load-bearing
T20	THE SUBSTRATE IS NOT NEUTRAL: what the CRM measures, the org optimizes (Goodhart). Observability mutates the system observed; therefore the measurement protocols the model depends on are themselves interventions.load-bearing
T21	PHI-VECTOR > PHI-SCALAR, BUT ONLY IF MEASURED: decomposing PMF into sub-factors adds rigor only where each sub-factor is independently observable. Today only retention pull is; the rest remain hypotheses. Decomposition without measurement multiplies the ghost, it does not exorcise it.load-bearing
T22	TOPOLOGY VS DRIFT: the dangerous regime change is the one that rewrites the graph (AI search reshaping discoverability), not the one that shifts a coefficient. Misclassifying topology change as drift is fatal precisely in the highest-stakes periods.load-bearing
T23	AGENTS ARE LEVERAGE THEN POLLUTION: agentic GTM is alpha to the first mover and equilibrium degradation at scale — homogenized messaging, inflated noise floor, anti-agent defenses. Marginal GTM alpha from agents decays as adoption rises.load-bearing
T24	POWER LAW BREAKS THE AVERAGE: under revenue/rep/channel concentration, mean-coefficient reasoning and the naive Bottleneck Theorem mislead. Optimization must be tail-aware; a few units carry the outcome.load-bearing
T25	NAME THE OBJECTIVE OR OPTIMIZE INCOHERENTLY: growth, efficiency, survival, positioning, and optionality conflict. Without an explicit regime-conditional weight vector, the system silently optimizes whatever the comp plan rewards.load-bearing
T26	CREATION HAS DIFFERENT PHYSICS: in category-creation mode the funnel/Phi machinery is largely inapplicable; outcomes come from narrative and reality-construction. The model is strong on playing the game and weak on making it — and says so.
T27	THE WALK IS THE ONLY INSTANCE: the MRR conservation law is the subscription case of a family. Usage, transaction, and marketplace models need different stocks and break the ACV-driven motion math. The economics tier is revenue-model-conditional.load-bearing
T28	POROUS FOR HUMANS, GATED FOR AGENTS: the tier boundary is normative, not descriptive — execution constantly forces strategy revision, and that feedback loop is the real engine. The safety property is preserved by letting agents PROPOSE Tier-3 changes that humans ratify, never enact unilaterally.load-bearing
T8.1	[AGENTIC, C6+ EXTENSION] The behavioral tier an agent executes is governed by TWO separable levers, not one: how much of the chain it runs (autonomy alpha) and how mature the governance around it is (G). Raising alpha without raising G does not merely fail to help — it can raise effective cost (E-A4).load-bearing	architecture
T29	[AGENTIC, C6+ EXTENSION] Governance maturity is not a safety tax on autonomy; it is a PRECONDITION for autonomy to lower cost at all. Under low G the fatigue term rho makes review a poor investment and cost-minimizing alpha drifts toward full autonomy despite worse reliability; under high G, review pays for itself and cost-minimizing alpha drifts toward heavy oversight at a LOWER absolute cost floor. Which side of the G threshold a vendor sits on decides the strategy, not the nominal autonomy level.load-bearing	architecture
T30	[AGENTIC, C6+ EXTENSION] Agent adoption carries a category externality (E-A7) that no firm-level governance internalizes. 'Agents are leverage then pollution' (T23) is not just diminishing returns to one firm's adoption — it is degradation of a SHARED resource (buyer attention / signal-to-noise) with AGGREGATE category adoption kappa, so first-mover alpha is structurally time-limited regardless of any single vendor's execution quality.load-bearing	architecture

All 32 thesis pages →

Failure modes

Identity/correlation conflation. Treating a regime or correlational edge with the confidence of an identity. The original sin. Mitigation: every edge carries epistemic_type; counterfactuals are gated by type.

Optimizing the confounded proxy. Chasing NRR/engagement to 'cause' revenue when both are driven by Phi. Mitigation: never run a counterfactual on a correlational edge.

Ratio over payback. Celebrating 5:1 LTV:CAC while running out of cash. Mitigation: payback is the binding constraint at venture stage.

Comfortable-stage optimization. Improving the funnel stage you're already good at instead of the bottleneck. Mitigation: bottleneck theorem; always intervene at the min-conversion stage.

Brand over-cut. Cutting demand-creation because last-touch shows no ROI. Mitigation: model the convolution kernel; use self-reported attribution.

Tactics-before-fit. Scaling GTM machinery below threshold Phi. Mitigation: PMF gate before motion scale-up. The most expensive failure mode.

Static-model rot. Trusting fitted coefficients after the regime has moved (channel saturated, competitor copied you, macro shifted). Mitigation: regime-tier coefficients carry half-lives and are re-estimated; treat drift as a bug (LangChain's CI-eval principle).

Agent overreach. Letting an execution agent silently drift Tier-3 configuration (ICP/motion/pricing) or act without human-in-the-loop on relationship-destroying touches. Mitigation: the agent_interface boundary — execute Tier 1, read Tiers 2-3, never rewrite them.

Ungoverned agent spend. Agent execution cost (high, usage-based) silently inflates CAC. Mitigation: Agent FinOps tracking per task/account; the FinOps->CAC edge is real.

Compounding-error blast radius. A small early agent misjudgment cascades through a multi-step chain into a relationship-damaging send. Mitigation: human-in-the-loop checkpoints + data-quality gates + escalation paths interrupt the chain.

POC abandonment. ~30% of GenAI projects die after POC on poor data quality / unclear value. Mitigation: phased rollout starting with the lowest-risk highest-signal loop, proven against a manual control group.

Treating agents as automation. Buying a 'standalone agent' expecting OS-level breadth/governance, or wiring agents as if-this-then-that. Mitigation: the maturity ladder — match the rung to the need; agents reason, automation triggers.

Identity laundering (the v1.1 original sin, named). Tagging LTV/NRR/CAC-payback as 'identities' when they are assumption-dependent estimators. Inflates certainty and propagates upward into a circular PMF thesis. Mitigation: only the MRR walk is identity; everything else carries its hidden_assumption.

Regime-blind benchmarking. Applying a ZIRP-era benchmark (e.g. 24-mo payback is fine) in a high-rate regime. Mitigation: Psi gates which objective weighting and payback target apply.

Funnel-causal fallacy. Reading a Tier-1 conversion coefficient as seller skill and scaling headcount to 'double pipeline', when it is a mixture over buyer-state. Causes geometric decay into a fixed in-market pool. Mitigation: condition on B; de-confounded estimand.

Brand-as-flow error. Cutting brand spend because attribution shows low last-touch credit, treating a stock as a flow. Mitigation: B_r is a stock with slow decay; judge via share-of-search/CEP breadth, not last-touch.

Phi circularity. Proving PMF with the NRR/retention it is supposed to predict. Mitigation: fix Phi ex-ante (week-8-12 retention floor or t-1 third-party satisfaction); run the nested regression -- or admit it is unrun and label T7 a heuristic.

Agentic over-claim (cost-per-activity vs cost-per-opportunity). Citing collapsed cost-per-email as a CAC win. Cost-per-qualified-opportunity net of deliverability decay and operator maintenance is unproven. Mitigation: demand a holdout; treat uncontrolled ROI multiples as marketing math.

Optimizing the visible 5%. Pointing agents entirely at the seller funnel while ~95% of the decision is pre-contact. Mitigation: ask whether the agent moves B_r or B, not just closes meetings faster.

Maximal-map mistaken for surveyed territory. Reading v3.0's comprehensive coverage as uniform reliability. Mitigation: obey the measurement_status tag — never feed an unmeasurable_hypothesis as a number into a decision.

Automated the wrong playbook. Agents executing a stale Tier-3 config flawlessly because the execution->strategy feedback loop was severed. Mitigation: cross_tier_coupling — agents propose, humans ratify, loop stays live.

HITL rubber-stamp rot. Reviewer alert fatigue turns human approvals into noise, corrupting the learning signal. Mitigation: down-weight approvals when approve-rate and downstream quality diverge (learning_layer).

Agent equilibrium degradation. Everyone's agents homogenize outreach and inflate the noise floor, decaying marginal alpha. Mitigation: track reply-rate vs market send volume; differentiate or retreat from saturated channels.

Ontology drift unnoticed. Stage definitions inflate under quota pressure; the same label now means something weaker. Mitigation: monitor conversion-distribution shift under a fixed label (ontology_instability).

Average-case blindness under power law. Mean LTV / mean r_i reasoning when a few units carry the base. Mitigation: tail-aware metrics; treat top-decile concentration as tail risk, not strength.

Objective incoherence. Optimizing conflicting goals at once (short-term CAC vs brand, ACV vs PLG loops) with no stated weights. Mitigation: make the regime-conditional objective vector explicit (objective_function).

Category myopia. Optimizing the funnel inside a game a competitor is busy redefining. Mitigation: watch branded-category-search share; know whether you are in creation or optimization mode.

Open tensions

The maximal version owes an explicit accounting of what completeness cost. These are not resolved.

ADD vs SUBTRACT: v3.0 chose completeness; the disciplined critique (doc 10) chose parsimony. By that standard v3.0 is LESS falsifiable than the minimal core. The measurement_status tags are a mitigation, not a resolution.
MAP vs ENGINE: v3.0 is a far more complete MAP, but adding seventeen layers moved it further from a runnable ENGINE, not closer. Only the MRR-walk slice (gtm_engine_slice.py) executes. The ratio of spec-to-machinery got worse, exactly as the harshest critiques predicted.
PHI: decomposed but still mostly unmeasurable. Eight sub-factors, one (retention pull) currently measurable. The ghost-variable risk is reduced in principle and unchanged in practice.
LEARNING: specified, not built. The update rule answers the 'silicon BDR' critique on paper; it still needs experimental traffic most ventures cannot spare, and full causal-graph revision is unimplemented.
COVERAGE vs USABILITY: the operator layer (Q) is the bet that this collapses to a few levers. If it does not, v3.0 is a monograph operators admire and route around — the precise fate the critiques warned of.

Measurement-gap register

Marks the boundary between honest-but-unparameterized and operational. These parameters the rebuilt model NEEDS but CANNOT currently measure from public data. All four construction reports converge on this list.

Parameter	Consequence if unmeasured
Ex-ante Phi panel	T7's nested regression (b3 sign = multiplicative vs additive) cannot be run; T7 stays a localized heuristic until a private aggregator (Bessemer, SaaS Capital) collects independent t-1 satisfaction + subsequent growth + GTM spend.
Switching-cost S at account level	Regime boundary (multiplicative vs additive) can't be set directly; proxied via GRR>=95% + low-satisfaction confounder.
Brand alpha (build rate) and delta (decay) for SaaS	Brand-stock dynamics are directional only; flagged retrospective-only / forecasting liability.
Sustained k>=1 in B2B	No public example exists; the divergent PLG closed form is a non-binding limiting case.
Noise-floor theta as a stable constant	25-40% is a reasoned assumption (Gross/Taleb), not a fitted 1 - out-of-sample-R^2 for B2B SaaS.
Controlled CAC under agentic GTM	No public holdout exists; whether agents lower cost-per-qualified-opportunity net of decay+maintenance is unproven.
dCAC_agentic (E-A8)	No public holdout exists; whether an agentic motion lowers cost-per-qualified-opportunity net of deliverability decay + maintenance is unproven. Single most consequential hole for ROI-based GTM messaging.
gamma_max(G), rho(G) (E-A2)	T29's sign-flip result cannot be calibrated for any real deployment; the optimal-autonomy decision stays illustrative.
theta, delta_theta (E-A7, category noise floor)	T23/T30's category-externality claim is not yet even directionally instrumented.
lambda (E-A6, HITL learning-rate)	Cannot validate whether T9's data-collection framing is operationally real or aspirational.
B_agent (E-A9, buying-committee agent-evaluator share)	Cannot test whether T12's ~95% pre-contact primacy holds when the buying committee itself includes AI evaluators (Gap 7).
kappa_gov(G) (E-A4, failure-cost absorption)	E-A4's cost-floor estimate lacks its governance-absorption input; secondary to calibrating gamma_max/rho first.

Cluster hubs

Seven cluster hubs — each a focused slice of the model with vocabulary, equations, and cross-links.

C1 · Fundamentals

Fundamentals & definitions

GTM, ICP, PMF, positioning, TAM/SAM/SOM, pricing

Hub →

C2 · Motions

Motions & channels

PLG, ABM, outbound, inbound, motion inequality

Hub →

C3 · Funnel

Funnel & lifecycle

Bowtie, MEDDIC, SPICED, funnel metrics

Hub →

C4 · Economics

Economics & metrics

MRR, CAC, LTV, NRR, Rule of 40, magic number

Hub →

C5 · Buyer-state

Buyer-state & demand

95-5 rule, brand as stock, demand gen vs lead gen

Hub →

C6 · Agentic

Agentic GTM & GTM OS

SRAL tiers, HITL design, what agents can own

Hub →

C7 · Regimes

Regimes & strategy half-lives

Psi scalar, reflexivity, topology vs drift

Hub →

Reference & arguments

A1 · Reference

Glossary

219 go-to-market terms, each a citable definition.

Browse →

A2 · Arguments

Theses

32 contested claims with mechanism, evidence for/against, and operator implication.

Browse →

A3 · Comparisons

Comparisons

12 structured comparisons — PLG vs SLG, ABM vs demand-gen, R40 vs Rule of X, and more.

Browse →

C4 · Metrics

Metrics

12 metrics with formulas, benchmark ranges by stage, and naive vs corrected versions.

Browse →

A9 · Research

Research syntheses

6 cross-study syntheses reconciling conflicting evidence and arriving at a position.

Browse →

A10 · Teardowns

Company teardowns

Snowflake, Stripe, OpenAI, Figma, Notion, and more — GTM mapped against the model.

Browse →

A11 · Coined term

The parity polygon

Canonical definition of the coined GTM frame — a uniformly category-average profile with no attributable edge.

Read →

GTM Basics · Fundamentals

GTM Basics

Go-to-market from first principles — ICP, motions, funnel, and economics, for anyone learning how GTM works.

Learn →

Tools & interactive

A4 · Calculators

GTM Calculators

CAC payback, LTV:CAC, Rule of 40/X, and motion selector. Browser-native, no data sent.

Open →

A4 · Simulator

MRR Walk Simulator

12-month MRR identity projection with compounding NRR analysis and Phi check.

Open →

A5 · Diagnostic

GTM Diagnostic

30-question readiness scorecard across 6 dimensions. Results are private.

Take the diagnostic →

Data & research

A6 · Evidence

Evidence database

20 agentic GTM implementation cases with methodology quality scores.

Browse →

A6 · Dataset

Benchmark dataset

12 metrics by stage (Seed→BIC). Bessemer, SaaS Capital, KeyBanc. CC BY 4.0.

View →

A7 · Annual Index

Agentic GTM Index 2026

First systematic census of agentic GTM in production — 20 cases.

View →

Playbooks

A8 · Playbook · HowTo schema

90-Day GTM Plan

Open →

A8 · Playbook · HowTo schema

Minimum Viable GTM

Open →

A8 · Playbook · HowTo schema

Motion Decision Tree

Open →

A8 · Playbook · HowTo schema

GTM OS Checklist

Open →

A8 · Playbook · HowTo schema

API, data & infrastructure

REST API · SoftwareApplication schema

GTM World Model API

3 endpoints: terms, theses, full model JSON. No auth. CORS enabled. CC BY 4.0.

API docs →

Newsletter

GTM Currency

Monthly research digest: model updates, evidence, Currency Delta changelog, AI citation tracking.

Subscribe →

Certification (planned Q4 2026)

GTM World Model Certification

Practitioner (L1), Architect (L2), Researcher (L3). Self-assessment available now.

Learn more →

Downloadable artifacts

The model is open. Take the data and the engine.

world_model.json vocabulary.json engine_slice.py

Version history

Version	Date	Change
v3.2	2026-05-31	Tier 0 (buyer-state, brand stock), regime scalar, agent layer recalibration, measurement-gap register.

How to cite this

@misc{singh_gtm_world_model_2026,
  author = {Singh, Shalvi},
  title  = {GTM World Model v3.2},
  year   = {2026},
  url     = {https://shalvisingh.com/gtm}
}

Singh, Shalvi. "GTM World Model v3.2." shalvisingh.com, 2026. https://shalvisingh.com/gtm