# Editorial Synthesis Report

The AI-Augmented GTM Playbook • Version 1.0 • 9 October 2026

## A Central thesis

**Build GTM around valuable customer interventions, sound economics, accountable decisions and reliable execution. Use AI where it demonstrably improves the quality or cost of that work. Scale only the additional value that survives comparison with the best feasible alternative.**

This thesis preserves STR’s commercial decision focus, GRW’s contribution economics, SYS and AGT’s separation of analysis from action, and OPS’s named accountability. It incorporates RT’s decisive restriction: the combined proposal is **promising but unproven** as an investment programme. Strong confidence in the operating principles does not establish a return from an integrated AI system. [STR §§1,17,46; GRW §§1–7; RT §§1–12,77–83]

The core reader promise is practical: understand the model, diagnose the constraint, prioritise an intervention, choose a Play, configure its prerequisites, execute, measure, retain useful evidence and improve the next decision. The minimum programme is one valuable decision, one owner and one feasible comparison. Zero agents is a valid permanent configuration.

## B Red Team verdict integration

| Original claim or default | Red Team challenge and evidence | Editorial verdict and product change |
|---|---|---|
| Connected AI decisioning is the largest value pool | No comparative portfolio evidence; production savings and better management are different mechanisms | **Qualify.** Local bottleneck and comparative economics determine value; remove the universal ranking from the opening. |
| Five agent-assisted capabilities should start together | AGT reports no proven local agent result; shared reviewers and engineers are uncosted | **Delete the default.** One workflow and at most one AI comparison, unless separate capacity and economics justify more. |
| A Control Plane coordinates the system | Shared policy may matter, but SYS distinguishes logical responsibility from deployable software | **Simplify and demote.** Existing tools implement the core; a custom service needs its own case. |
| Precise continuous allocation maximises returns | Response estimates, budget mobility and channel interactions are unresolved | **Qualify.** Human owners compare feasible investment tranches and ranges. |
| Predicted LTV and churn enable better treatment | Prediction does not identify incremental response; sparse data can make decisions worse | **Preserve conditionally.** Realised cohorts first; advanced forecasts and uplift need prospective validation. |
| Human approval contains AI risk | Reviewers may lack time, evidence or authority; allowed tools can still be misused | **Strengthen.** Inspectable actions, staffed review and deterministic limits remain necessary. |
| Connected federated teams are the default | OPS provides a coherent design, not evidence of superiority over repaired existing functions | **Demote.** Start in current functions; prove a persistent coordination need before changing structure. |
| More memory creates a learning advantage | No measured reuse premium; stale conclusions can become institutionalised | **Simplify.** One decision and evidence record, with scope, contradictions and review triggers. |
| Faster output creates profit | Review, rework and unrealised capacity can absorb the gain | **Qualify.** Measure accepted-work cost; recognise cash only through a real expense action. |
| Source cases establish general uplift | RT identifies provider interests, secondary citations and missing counterfactuals | **Delete benchmarks; narrow examples.** Keep method lessons, not return assumptions. |

Sources: RT §§13–54,77–83; RTB A–R. These changes affect the minimum system, the Play selection rule, the roadmap, the evidence labels and every funding gate.

### Editorial resolutions

These reconcile source positions. They are editorial choices, not newly validated strategy.

| ID | Contradiction and source positions | Narrowest defensible resolution | Confidence |
|---|---|---|---|
| ER01 | RES and OS promote connected-system superiority; later specialists and RT withhold universal proof | Keep commercial decision discipline; test process, AI and coordination separately using A/B/C/D alternatives | High |
| ER02 | RES specifies a stack; SYS describes optional logical capabilities; RT asks for a concrete small deployment | One reference architecture with minimum and optional expansions; no new application implied by a box | High |
| ER03 | STR central intelligence, native platform optimisation and OPS embedded ownership appear to compete | Share definitions and cross-team constraints; retain craft, local execution and named budget authority | High |
| ER04 | AGT §38 says Creative Intelligence “Build now”; §74 says “Pilot” | Optional pilot after a baseline, only when annotation or synthesis is a bottleneck | High |
| ER05 | OS M1 means MMM; STR and OPS M1 mean standard economics; SYS and OPS gate codes differ | Retire numeric maturity and gate codes. Use Basic, Connected and Advanced configurations, dimension-specific readiness, and named gates | High |
| ER06 | Separate 90-day plans collectively require several pilots and workstreams | One capacity plan, one intervention and one decision date. Foundation work belongs to that intervention | High |
| ER07 | OS subtracts avoided incentive expense; GRW’s cLTV trailing cost term is ambiguous | Avoided cost adds value once. Sum and discount future contribution and any excluded future costs consistently; show cost scope | High |
| ER08 | GRW §41 uses 5,000 treated and 5,000 controls, then values 200 retained accounts per 10,000 assigned | A 2-point effect implies 100 additional retained accounts among 5,000 treated. A 10,000-treatment rollout projects 200 only under transport assumptions and a new cost basis. Do not carry the original net value across denominators | High on arithmetic; original cost basis unresolved |
| ER09 | User brief invites many frameworks; RT consolidates into five artefacts | Provide specialised views within five working records. The core loop remains the conceptual model; views are not new registers | High |
| ER10 | “Observed, attributed, modelled, incremental, causal” can look like a single evidence ladder | Separate what was estimated from how it was identified. Modelled effects can be causal under justified assumptions; recorded outcomes alone cannot | High |
| ER11 | Universal default Play rankings conflict with constraint-led prioritisation | Publish an editorial access order with qualitative screening; require local reranking and veto checks before funding | High |
| ER12 | A broad future roadmap implies inevitable autonomy | Make each phase a review window. Remaining simple, maintaining human approval or retiring a workflow are legitimate advanced outcomes | High |

## C Canonical model

**Sense → Understand → Decide → Activate → Measure → Learn → Optimise.**

Retain STR §7 because it traces a complete commercial intervention and survives RT when interpreted as management work. “Optimise” means revise, repeat, scale or stop within constraints. It does not mean a machine continuously changes strategy. “Learn” requires a reviewed conclusion with scope, not a saved AI summary.

| Stage | Economic connection | Minimum output |
|---|---|---|
| Sense | Recognise a material customer problem or loss | Dated, attributable signal |
| Understand | Distinguish the binding constraint from symptoms | Supported diagnosis and uncertainty |
| Decide | Allocate real money and capacity to a feasible alternative | Owner, rationale, limits and decision date |
| Activate | Change the customer experience or operating work | Actual intervention and delivery record |
| Measure | Estimate what changed and reconcile costs | Scoped outcome and economic readout |
| Learn | Preserve evidence that can change another decision | Reviewed finding, applicability and expiry |
| Optimise | Improve the next feasible allocation or intervention | Continue, change, replicate or stop |

The physical minimum is existing CRM, commerce or product records; relevant Finance data; a controlled export or query; a brief and approval; manual or native execution; an outcome comparison; and a small linked record. It is not seven software products. [SYS §§3,29; RT §66]

## D Canonical principles

1. Fund a customer or commercial intervention with a named owner and an available action.
2. Compare investments on contribution and cash over an explicit horizon, with customer and strategic constraints.
3. Diagnose the binding constraint before optimising a downstream metric.
4. Compare AI with a repaired conventional workflow and integration with the simpler AI-assisted workflow.
5. Keep observation, prediction, attribution and causal effect distinct.
6. Test material uncertainties only when the evidence can inform a decision at proportionate cost.
7. Choose the least complex reliable mechanism; preserve native delivery where it fits.
8. Keep human authority for strategy, material investment, pricing, novel claims and ambiguous customer consequences.
9. Enforce exact action limits outside probabilistic models and budget meaningful human review.
10. Treat time saved as capacity until an expense or useful output changes.
11. Retain scoped evidence that people reuse; revise or retire stale findings and old work.
12. Add infrastructure, permanent roles and autonomy only when their additional value pays for continuing ownership and exit.

These consolidate STR §46, AGT §5, OPS §3 and RT §83. Principles 4, 10 and 12 make the Red Team’s funding boundaries especially explicit.

## E Canonical frameworks

| Framework | Classification | Decision enabled | Consolidation |
|---|---|---|---|
| GTM Growth System | Canonical conceptual model | Where is the intervention failing? | All sensing, decision and learning loops use the same seven stages |
| T1 Intervention and investment card | Canonical working artefact | Is this problem and intervention worth acting on? | System canvas, constraint diagnostic, opportunity brief, Play priority |
| T2 Economics and benefits ledger | Canonical working artefact | Does the option beat alternatives and become economic value? | CAC/LTV, tranche comparison, value realisation and ownership cost |
| T3 Decision experiment and evidence record | Canonical working artefact | What do we know, what did we decide and what should change? | Experiment brief, performance review, measurement plan and memory |
| T4 Action authority and operating contract | Canonical working artefact | Who or what may act, under which boundaries? | Work allocation, decision rights, agent card, tool contract and autonomy checks |
| T5 Delivery and retirement plan | Canonical working artefact | Can this be delivered, maintained and closed? | Readiness, capability gaps, build/buy, roadmap, role design and retirement |
| Rule / Model / AI / Agent / Human router | Supporting view within T4 | What method fits this task? | Retires multiple competing automation taxonomies |
| Named gates | Supporting view within T5 | Continue, fix first or stop? | Replaces SYS and OPS gate codes |
| Detailed statistical, model and API contracts | Appendix | Implement a justified advanced component | Refer to source specialists rather than reproduce their whole catalogues |
| Multiplicative priority scores, autonomous maturity ladder, multiple flywheels | Delete | No additional reliable decision | Removed rather than renamed |

## F Concepts deleted

The following do not reappear as recommended practice: universal uplift and return ranges; the compulsory graph/vector/streaming stack; the five-agent launch portfolio; agents for ordinary arithmetic, assignment, audiences and exact publishing; strategy super-agents; autonomous material allocation as a destination; model self-confidence as permission; automatic compliance certification; blanket bandit replacement; technology ownership as a guaranteed moat; attributed profit presented as incremental; and arbitrary score products presented as investment evidence.

The quantified Uber example is excluded pending primary verification. Provider cases are not upgraded because another paper cites them. The OS avoided-incentive sign error is corrected. None of these deletions asserts that all components are useless in every context. A narrower component may return only with an explicit new case and evidence. [RT §§50,71,80]

## G Concepts qualified

Customer-value acquisition starts with reconciled cohorts. Retention targets addressable causes, not predicted risk alone. Creative tags support hypotheses, not causal rules. Marginal economics uses feasible tranches and uncertainty. Measurement is proportionate to stakes and sample. AI synthesis requires source checking and a comparison of accepted work. Memory requires reuse and expiry. Structural change requires recurring workload and authority. Closed loops require trusted feedback and can remain human governed.

## H Concepts moved to advanced

Predictive cLTV, individual uplift, MMM, cross-channel scenario services, custom coordination software, specialised retrieval, knowledge graphs, new streaming infrastructure, multi-agent designs, persistent pods and dedicated AI roles sit outside the minimum. Each has a prerequisite and a separate cost case. Bounded tool-using agents are advanced; their incremental commercial value remains locally unproven. Autonomous portfolio policy and general GEO ROI remain experimental or unresolved, not advanced best practice. [RT §§67–70]

## I Unresolved issues

| Question | Why it matters | Current position and evidence gap | Recommended validation | Impact if wrong |
|---|---|---|---|---|
| Does AI add value over process repair? | Determines C−B investment | Unknown locally; task examples cannot settle it | Same-task comparison, blinded quality review, full cost | Positive total ROI can conceal a worse choice |
| Does integration add value? | Determines shared-platform spend | Unknown D−C premium | Measure recurring handoff loss and test simpler coordination | Permanent cost without changed decisions |
| Can a useful effect be detected? | Determines honest claims and timing | No company volumes, variance or lag | Method owner assesses power or identification before launch | Long, uninformative tests |
| What are actual cohort economics? | Determines acquisition and retention limits | No company cost scope or outcome joins | Finance reconciliation and short-horizon cohorts | Scale destroys contribution or liquidity |
| Are outcomes addressable? | Risk scores do not create treatments | No local treatment evidence | Customer research and eligible treatment comparison | Accurate prediction still funds waste |
| Can budgets and delivery capacity move? | Recommendations need execution | Proposed rights are not observed authority | Execute one small authorised change | Reporting grows; intervention does not |
| Are native tools sufficient? | Determines new build need | Installed editions and contracts unknown | End-to-end native acceptance test | Duplicate technology and support |
| What does five-year ownership cost? | Determines long-term value | No quotes, operator roster or exit price | Cost cash, staff capacity, review, change and retirement | Benefit overwhelmed by fixed cost |
| Can reviewers perform meaningful checks? | Determines reliability of human gates | Capacity and error evidence absent | Sample review time, quality and queue load | Rubber stamping or delays |
| How should long-term demand be valued? | Protects future growth | No universal horizon or causal allocation | Finance and brand scenarios; preserve strategic floors | Short-term extraction weakens demand |
| Does memory or autonomy earn its extra cost? | Determines advanced additions | Reuse, delay value and residual loss unknown | Track cited reuse; compare same-action approval options | Complexity is funded on speculation |
| What did the source retention cost cover? | Prevents denominator error | GRW §41 cost basis is ambiguous | Clarify treated versus rollout cost before reusing source net figure | Material overstatement of return |

**Additional Research Required:** no external research is needed to state the conservative core. Before publishing excluded quantified cases, obtain the original population, period, intervention, comparator, method, cost and independence evidence. Before local investment, answer the organisation-specific questions above; do not fill them with industry averages.

## J Final product architecture

The Playbook has a compact core and reusable implementation modules:

1. Executive orientation and reader routes.
2. Customer economics and comparative investment.
3. The GTM Growth System and decision architecture.
4. Diagnose constraints and readiness.
5. Prioritise and select a Play.
6. Execute growth interventions across the customer lifecycle.
7. Measure and preserve useful learning.
8. Configure the minimum system and justify advanced additions.
9. Use AI and automation where they earn their cost.
10. Operate with human accountability.
11. Implement through named gates.
12. Reference, evidence, vocabulary and open issues.

The separate Play Library contains 12 fully specified Plays. The Toolkit Inventory supplies five copyable working artefacts with specialised views. Six prompts support grounded synthesis, diagnosis, creative learning and briefing. The Implementation Guide gives dependency-led review windows from the first 90 days through the optional advanced state. A chapter-by-chapter Blueprint and the 58-claim Source and Claim Register support future editing.

### Coherence gate before manuscript drafting

The thesis, minimum system, twelve principles, terminology, five working artefacts, twelve Plays and named gates use the same constraints. No rejected claim supplies a recommended workflow or financial assumption. Frameworks enable a decision, execution or measurement; advanced modules have entry conditions. ER08 remains visibly qualified. The manuscript can now be drafted against this specification.