Home Reference Evidence Evidence 58 claims reviewed · Accepted · Accepted with conditions · Rejected · Qualified · Advanced · Unresolved · Experimental
Two bounded evidence cases Reference REF03 Case lessons with scope and non-transferability limits. From CH12 .
Booking.com and experiment-linked machine learning. The problem was turning model development into business improvements. The supplied research describes a primary practitioner paper about successful deployed ML applications and their experimental validation. The useful mechanism is connecting each model to an outcome and a decision. The reported outcome is production precedent across roughly 150 successful applications, not a denominator of all attempts or a pooled profit return. Evidence is a primary technical account, with selection and scale limits. Learn to evaluate the decision effect; do not infer that modern agent orchestration, a small firm or every ML project will achieve the same result. Booking.com KDD paper . Source treatment: OS §11; RT §§7,50–51. Publication: 2019 (year precision); Tier 2 practitioner evidence; not independent commercial evidence.
Unilever and product-image production. The problem was the cost and speed of imagery production. The company described a digital-twin and AI workflow that reduced production time and cost. The mechanism is a changed asset-production process, not an autonomous commercial strategy. The company-reported outcome illustrates a potential expense or throughput benefit; it does not isolate pure model causality, demonstrate incremental sales or provide a transferable return estimate. Evidence is a company case. Learn to compare accepted assets and actual purchases; do not generalise it to media, retention or an integrated GTM platform. Unilever announcement . Source treatment: OS §11; RT §50. Publication: 18 March 2025; Tier 3 company-reported evidence; not independent.
These references are carried forward from the supplied research and Red Team review. Editorial & Evidence QA checked the existing primary pages and publication metadata on 9 October 2026. This was a bounded citation check, not independent validation of commercial outcomes or a fresh audit of the full Booking.com methods. Their stated scope is preserved. The excluded Uber amount, provider ROAS claims and universal uplift ranges do not enter any investment model.
Claim classifications Accepted means a principle or mechanism survived critical review, not that it guarantees a return. Accepted with conditions means prerequisites apply. Qualified means the original claim has been narrowed. Advanced needs substantial readiness. Experimental needs a capped comparison. Unresolved means the evidence cannot settle the claim. Rejected means the recommendation is excluded.
High, Medium and Low describe confidence in the stated target, not the chance of a positive ROI. For a Qualified or Rejected historical claim, confidence applies to the decision to narrow or reject it and to the supported limit—not to the original headline. Read the headline and classification together. An accounting identity or design principle is not empirical evidence of a commercial outcome. Follow the linked chapters, Plays and tools to see where each claim applies.
Research sources This section explains where the Playbook’s evidence comes from and what it can support. Several documents may draw on the same research; they do not count as independent confirmation of one another. External references were carried forward from that research with their original limitations. No new external research was added for this edition.
Code Source document Scope and evidence caution AGT AI Agent and Automation Operating Architecture Automation architecture; explicitly no proven commercial agent result. GRW AI Augmented GTM Growth and Performance Architecture Growth and measurement; conditional play mechanisms and illustrative calculations. OPS AI Augmented GTM Operating Model and Transformation Plan Operating model; derived design, not independent organisational evidence. OS The AI Augmented Go To Market Operating System Research synthesis; company filings and preprints are not automatically highest-tier evidence. RES AI Go To Market Systems Research Research dossier; strong claims require RT corrections. RT AI Augmented GTM Red Team Verdict Red Team; claim audit, investment challenge and simplification. RTB AI Augmented GTM Playbook Editor Brief Red Team summary; same workstream as RT, not corroboration. STR AI Augmented GTM Strategy Commercial strategy; twelve principles and commercial loop. SYS AI Augmented GTM Systems Architecture Systems architecture; proposed patterns, not an audited deployment.
Claims and their evidence Evidence C03 Confidence High Seven-stage GTM loop Confidence target Seven-stage GTM loop Qualification A mnemonic for accountable work, not evidence of AI superiority or required infrastructure. Evidence level Management model Source STR §7; SYS §18; RT §§9,65–66 Where this applies CH03 The GTM Growth System Evidence C04 Confidence High Connected decision systems are the dominant AI value pool — excluded Confidence target Confidence in rejecting the original claim. Supported boundary: No head-to-head portfolio evidence; content savings may dominate locally. Qualification No head-to-head portfolio evidence; content savings may dominate locally. Evidence level Unsupported generalisation Source RES §§1,14; OS §§1,4; RT §§7–12,50 Where this applies CH12 Reference and stewardship Evidence C05 Confidence High Routine media recovery and conversion uplift ranges — excluded Confidence target Confidence in rejecting the original claim. Supported boundary: No universal 20–40%, 25–50% or 10–25% planning assumptions. Qualification No universal 20–40%, 25–50% or 10–25% planning assumptions. Evidence level Selected cases extrapolated Source RES §14; RT §50 Where this applies CH12 Reference and stewardship Evidence C07 Confidence High Hours saved become financial savings — narrowed, not endorsed Confidence target Confidence in narrowing the original claim. Supported boundary: Capacity stays separate until expenditure disappears or useful redeployment is observed. Qualification Capacity stays separate until expenditure disappears or useful redeployment is observed. Evidence level Accounting distinction Source OS §14; OPS §§21,43; RT §14 Where this applies CH02 Customer economics · P10 Redesign and test one AI-assisted workflow · T2 Economics and benefits ledger Evidence C10 Confidence Medium Individual predicted cLTV for acquisition Confidence target Individual predicted cLTV for acquisition Qualification Temporal validation, reliable contribution and improved prospective decisions required. Evidence level Established method, local utility unknown Source GRW §§5,7,17; SYS §8; RT §§18,24 Where this applies CH08 Configure the system · P03 Acquire customers with useful downstream value Evidence C11 Confidence Medium Individual uplift and treatment optimisation Confidence target Individual uplift and treatment optimisation Qualification Treatment variation, adequate outcomes and prospective benefit; cohort rules are valid end states. Evidence level Causal method with demanding data Source GRW §§18,30–31; RT §§18,67,69 Where this applies CH08 Configure the system · P07 Prevent profitable addressable churn Evidence C12 Confidence High Precise next-dollar optimisation — narrowed, not endorsed Confidence target Confidence in narrowing the original claim. Supported boundary: Use the next feasible investment tranche, ranges, real budget rights and protected commitments. Qualification Use the next feasible investment tranche, ranges, real budget rights and protected commitments. Evidence level Economic principle, uncertain estimates Source STR §18; GRW §§49–51; RT §§19,23 Where this applies CH05 Prioritise and choose · P11 Run an accountable investment and decision review · T2 Economics and benefits ledger Evidence C13 Confidence High Attribution estimates causality — excluded Confidence target Confidence in rejecting the original claim. Supported boundary: Attribution assigns credit; causal claims need a defensible design and assumptions. Qualification Attribution assigns credit; causal claims need a defensible design and assumptions. Evidence level Contradicted by identification limits Source RES §§2,5; GRW §§42–45; RT §20 Where this applies P04 Validate incremental acquisition · CH07 Measure and learn Evidence C15 Confidence High Every decision should be experimentally validated — narrowed, not endorsed Confidence target Confidence in narrowing the original claim. Supported boundary: Repair obvious defects; use qualified observation or judgement when testing cannot inform the choice. Qualification Repair obvious defects; use qualified observation or judgement when testing cannot inform the choice. Evidence level Proportionality Source OS §17; RT §§21–22 Where this applies P02 Remove the dominant growth constraint · CH07 Measure and learn Evidence C16 Confidence High Bandits replace A/B tests — excluded Confidence target Confidence in rejecting the original claim. Supported boundary: Adaptive allocation is optional and does not create sample size or automatic causal validity. Qualification Adaptive allocation is optional and does not create sample size or automatic causal validity. Evidence level Different optimisation and inference purposes Source RES §17; GRW §38; RT §22 Where this applies CH07 Measure and learn Evidence C17 Confidence High Measurement triangulation Confidence target Measurement triangulation Qualification Reconcile estimands, periods and assumptions; do not average lenses into truth. Evidence level Complementary measurement methods Source GRW §§42–47; SYS §16; RT §§20–23 Where this applies T3 Decision experiment and evidence record · CH07 Measure and learn Evidence C18 Confidence High Customer quality and native value signals Confidence target Customer quality and native value signals Qualification Native functionality is an option, not a proven local uplift; use verified cohort signals. Evidence level Economic mechanism and documented capability Source STR §§12,16; GRW §§7,10,17; RT §24 Where this applies P03 Acquire customers with useful downstream value Evidence C19 Confidence High Activation means customer experienced value Confidence target Activation means customer experienced value Qualification Validate leading milestone against retained value; do not label the association causal. Evidence level Commercial mechanism Source STR §13; GRW §28 Where this applies CH06 Execute growth interventions · P06 Improve the first value experience Evidence C20 Confidence High Churn risk identifies preventable churn — narrowed, not endorsed Confidence target Confidence in narrowing the original claim. Supported boundary: Identify addressable causes and compare repair, education, offer and no treatment. Qualification Identify addressable causes and compare repair, education, offer and no treatment. Evidence level Prediction versus intervention Source STR §14; GRW §§18,31; RT §18 Where this applies P07 Prevent profitable addressable churn Evidence C22 Confidence High Creative features cause performance differences — narrowed, not endorsed Confidence target Confidence in narrowing the original claim. Supported boundary: Tags generate hypotheses; controlled concept tests assess effects. Qualification Tags generate hypotheses; controlled concept tests assess effects. Evidence level Confounded delivery and provider evidence Source GRW §§22–25; RT §§24,50 Where this applies P05 Learn which proposition and creative concept works Evidence C24 Confidence High Expansion, usage, referral and win-back Confidence target Expansion, usage, referral and win-back Qualification Use customer need and net account contribution; account for cannibalisation, fraud and repeat churn. Evidence level Commercial mechanisms Source STR §15; GRW §§32–36 Where this applies P08 Grow relevant customer and account value Evidence C25 Confidence Low AI search visibility produces commercial ROI Confidence target AI search visibility produces commercial ROI Qualification Monitor dated observations only; no standalone GEO return claim. Evidence level Descriptive monitoring; no general counterfactual Source GRW §14; RT §§53–54 Where this applies CH12 Reference and stewardship · P09 Turn customer and market evidence into a decision Evidence C27 Confidence High A default portfolio of five agents — excluded Confidence target Confidence in rejecting the original claim. Supported boundary: One selected workflow; zero required agents; at most one candidate without a separate capacity case. Qualification One selected workflow; zero required agents; at most one candidate without a separate capacity case. Evidence level No local comparative evidence Source AGT executive decision; STR §25; RT §§33–34 Where this applies CH09 AI and automation Evidence C29 Confidence High Creative Intelligence build now — narrowed, not endorsed Confidence target Confidence in narrowing the original claim. Supported boundary: Optional pilot after a baseline, never an automatic build instruction. Qualification Optional pilot after a baseline, never an automatic build instruction. Evidence level Internal sequencing conflict Source AGT §§38,74; RT §81 Where this applies P05 Learn which proposition and creative concept works Evidence C30 Confidence High Agent for audience, measurement or campaign publishing — excluded Confidence target Confidence in rejecting the original claim. Supported boundary: Keep the capability as SQL, tested statistical code or exact approved publishing. Qualification Keep the capability as SQL, tested statistical code or exact approved publishing. Evidence level Simpler exact mechanisms sufficient Source AGT §75; RT §34 Where this applies CH09 AI and automation Evidence C31 Confidence Low Multi-agent collaboration as default Confidence target Multi-agent collaboration as default Qualification Isolation or context benefit must beat a single workflow after handoff and maintenance costs. Evidence level Design rationale, no general commercial proof Source RES §7; SYS §14; AGT §§45–48; RT §35 Where this applies CH12 Reference and stewardship Evidence C32 Confidence High Autonomous strategic and material budget authority — excluded Confidence target Confidence in rejecting the original claim. Supported boundary: Human authority remains for strategy, major investment, pricing and novel claims. Qualification Human authority remains for strategy, major investment, pricing and novel claims. Evidence level Broad risk and missing economic evidence Source RES §§16–17; AGT §§28–32; RT §§36,40 Where this applies CH09 AI and automation Evidence C33 Confidence Medium Bounded dynamic tool use Confidence target Bounded dynamic tool use Qualification Dynamic path must beat fixed workflow; explicit scopes, evaluation, stop and support. Evidence level Technical feasibility, local value unknown Source SYS §12; AGT §§1,4,56–61; RT §§33,38 Where this applies CH09 AI and automation · T4 Action authority and operating contract Evidence C34 Confidence Low Removing human approval creates value Confidence target Removing human approval creates value Qualification Compare the same intervention with approval; quantify delay and review cost versus added exposure. Evidence level Unproven autonomy premium Source AGT §§28–32; RT §36 Where this applies CH09 AI and automation · T4 Action authority and operating contract Evidence C35 Confidence High Human approval is sufficient safety — narrowed, not endorsed Confidence target Confidence in narrowing the original claim. Supported boundary: Reviewer authority, capacity and inspectable evidence plus independent technical limits. Qualification Reviewer authority, capacity and inspectable evidence plus independent technical limits. Evidence level Control design with human limits Source AGT §§23–27; OPS §32; RT §37 Where this applies T4 Action authority and operating contract · CH10 Operate with human accountability Evidence C36 Confidence High Model self-confidence grants permission — excluded Confidence target Confidence in rejecting the original claim. Supported boundary: Use empirical evaluation by risk slice and enforced action policy. Qualification Use empirical evaluation by risk slice and enforced action policy. Evidence level Uncalibrated assertion Source AGT §31; RT §38 Where this applies CH09 AI and automation · T4 Action authority and operating contract Evidence C37 Confidence High External content is data and action authority is separate Confidence target External content is data and action authority is separate Qualification Validate scope, current state, destination and exact approved action outside the model. Evidence level Security and control pattern Source SYS §§20–22; AGT §§20–24; RT §39 Where this applies CH09 AI and automation · T4 Action authority and operating contract Evidence C38 Confidence High Receipts, idempotency and unknown execution states Confidence target Receipts, idempotency and unknown execution states Qualification Required for relevant machine writes; query indeterminate state before retrying. Evidence level Distributed execution pattern Source SYS §15; AGT §24 and contract pack Where this applies CH12 Reference and stewardship · T4 Action authority and operating contract Evidence C39 Confidence High A custom Control Plane is foundational — narrowed, not endorsed Confidence target Confidence in narrowing the original claim. Supported boundary: Keep common controls; use existing tools first; price a separate coordination case. Qualification Keep common controls; use existing tools first; price a separate coordination case. Evidence level Logical requirements versus physical build Source RES §5; STR §27; SYS §§19,29,43; RT §§25–27 Where this applies CH08 Configure the system Evidence C40 Confidence High Graph, vector and stream infrastructure as mandatory — excluded Confidence target Confidence in rejecting the original claim. Supported boundary: Optional components need query, retrieval or latency evidence and ownership economics. Qualification Optional components need query, retrieval or latency evidence and ownership economics. Evidence level Unjustified universal architecture Source RES §5; SYS §§4,9,29; RT §§29–30 Where this applies CH12 Reference and stewardship Evidence C41 Confidence Medium Real-time architecture Confidence target Real-time architecture Qualification Native time-critical checks first; batch slower decisions; measure value lost through delay. Evidence level Decision latency mechanism Source SYS §4; AGT §66; RT §30 Where this applies CH08 Configure the system Evidence C42 Confidence High Minimal decision and evidence memory Confidence target Minimal decision and evidence memory Qualification One record with scope, contradiction and expiry; measure useful reuse before expansion. Evidence level Useful recordkeeping, unpriced reuse Source STR §23; SYS §9; OPS §38; RT §28 Where this applies T3 Decision experiment and evidence record · CH07 Measure and learn Evidence C43 Confidence High First-party data or a graph guarantees a moat — excluded Confidence target Confidence in rejecting the original claim. Supported boundary: Exclusive evidence must improve repeat decisions and resist imitation before any moat claim. Qualification Exclusive evidence must improve repeat decisions and resist imitation before any moat claim. Evidence level No demonstrated durable advantage Source STR §11; RES §5; RT §§11,29 Where this applies CH12 Reference and stewardship Evidence C44 Confidence High Federated organisation as universal default — narrowed, not endorsed Confidence target Confidence in narrowing the original claim. Supported boundary: Begin in current functions; structural change needs recurring work, authority and net coordination value. Qualification Begin in current functions; structural change needs recurring work, authority and net coordination value. Evidence level Design without comparative evidence Source OPS §§10–11; RT §§41–44 Where this applies CH10 Operate with human accountability Evidence C45 Confidence Medium Pods, CoE and new job titles Confidence target Pods, CoE and new job titles Qualification Accountabilities required; new permanent teams and titles are not. Evidence level Workload-dependent organisational design Source STR §29; OPS §§14,20; RT §§42–44 Where this applies T5 Delivery and retirement plan · CH10 Operate with human accountability Evidence C47 Confidence High Roadmap dates and separate specialist workstreams — narrowed, not endorsed Confidence target Confidence in narrowing the original claim. Supported boundary: One capacity plan and named gates; dates are review windows, not proof deadlines. Qualification One capacity plan and named gates; dates are review windows, not proof deadlines. Evidence level Uncosted parallel capacity Source STR §35; SYS §33; AGT §70; OPS §§56,63–67; RT §48 Where this applies CH12 Reference and stewardship Evidence C48 Confidence High Maturity measures AI and autonomy intensity — excluded Confidence target Confidence in rejecting the original claim. Supported boundary: Readiness per capability and target play; no overall maturity or autonomy-level arithmetic. Qualification Readiness per capability and target play; no overall maturity or autonomy-level arithmetic. Evidence level Incompatible scales and goal confusion Source OS §13; STR §36; OPS §50; RT §§36,81 Where this applies T5 Delivery and retirement plan · CH04 Diagnose the constraint and readiness Evidence C50 Confidence High Provider cases establish general ROI — excluded Confidence target Confidence in rejecting the original claim. Supported boundary: Use scoped teaching examples; do not upgrade repeated/vendor evidence. Qualification Use scoped teaching examples; do not upgrade repeated/vendor evidence. Evidence level Evidence scope and commercial interest Source RES §11; OS §11; RT §§49–52 Where this applies CH12 Reference and stewardship Evidence C51 Confidence Medium Unilever production example Confidence target Unilever production example Qualification Illustrates production workflow economics, not audited causal sales or orchestration ROI. Evidence level Company-reported case Source OS §11; RT §50 Source date 2025-03-18 (day precision) Where this applies CH12 Reference and stewardship Citation Unilever reinvents product shoots with digital twins and AI · Unilever · 2025-03-18 (day; publisher HTML metadata) · Tier 3 · Independent: No · Checked 2026-10-09 Supports: Company-reported use of digital twins and AI in product-image production.
Does not support: Audited incremental sales, general savings or agent/orchestration ROI.
Evidence C52 Confidence High Booking.com ML example Confidence target Booking.com ML example Qualification Teaches experiment-linked deployment; not average success rate or modern-agent return. Evidence level Primary practitioner paper, selected successes Source OS §11; RT §§7,50–51 Source date 2019 (year precision) Where this applies CH12 Reference and stewardship Citation 150 successful Machine Learning models: 6 lessons learned at Booking.com · KDD 2019 / ACM conference programme · Lucas Bernardi,Themistoklis Mavridis,Pablo Estevez · 2019 (year) · Tier 2 · Independent: No · Checked 2026-10-09 Supports: Authors describe selected deployed ML applications and experiment-linked evaluation.
Does not support: Attempt-level success rates, average returns, SME transfer or modern-agent economics.
Evidence C53 Confidence Low Uber quantified savings example Confidence target Uber quantified savings example Qualification Exclude from publication pending primary methods; exclusion does not assert the claim is false. Evidence level Primary amount and design unverified Source RES §11; RT §50 Where this applies CH12 Reference and stewardship Evidence C54 Confidence High Avoided incentives and cLTV notation — narrowed, not endorsed Confidence target Confidence in narrowing the original claim. Supported boundary: Avoided cost increases profit once; discount and sum all relevant future cost consistently. Qualification Avoided cost increases profit once; discount and sum all relevant future cost consistently. Evidence level Arithmetic and cost-scope correction Source OS §14; GRW §5; RT §15 Where this applies CH02 Customer economics Evidence C55 Confidence High Retention worked example projection — narrowed, not endorsed Confidence target Confidence in narrowing the original claim. Supported boundary: 2-point effect implies 100 in 5,000 treated, or 200 in a separate 10,000-treatment rollout. Qualification 2-point effect implies 100 in 5,000 treated, or 200 in a separate 10,000-treatment rollout. Evidence level Arithmetic distinction; cost basis ambiguous Source GRW §§41,43 Where this applies CH07 Measure and learn Evidence C56 Confidence High Closed loop guarantees improvement — narrowed, not endorsed Confidence target Confidence in narrowing the original claim. Supported boundary: Outcomes must inform the next decision; objective, measurement, constraints, owner and stop remain explicit. Qualification Outcomes must inform the next decision; objective, measurement, constraints, owner and stop remain explicit. Evidence level Feedback can reinforce bias Source STR §7; SYS §18; AGT §69; RT §40 Where this applies T4 Action authority and operating contract · CH03 The GTM Growth System Evidence C57 Confidence High Legal compliance certified by model — excluded Confidence target Confidence in rejecting the original claim. Supported boundary: Model flags support qualified review; it is never the legal or policy authority. Qualification Model flags support qualified review; it is never the legal or policy authority. Evidence level Authority and verification limits Source RES §§7,12; SYS §8; AGT §§20–23; RT §§54,71 Where this applies CH09 AI and automation Evidence C58 Confidence High Priority score products imply investment rank — excluded Confidence target Confidence in rejecting the original claim. Supported boundary: Use qualitative screening and local value ranges with hard vetoes; no synthetic ROI score. Qualification Use qualitative screening and local value ranges with hard vetoes; no synthetic ROI score. Evidence level Ordinal scales lack currency meaning Source AGT §74; GRW §39; RT §71 Where this applies CH05 Prioritise and choose