Appendix B — Practitioner's Toolkit + Further Reading · AI Human Interaction Guide

How to use this appendix

The guide cites these instruments and frameworks inline where they matter; this appendix is the consolidated reference so a practitioner can find them without re-reading the parts to locate citations. The instruments in §B.1 are the measurement set — what to instrument, with what scoring rubric. The frameworks in §B.2 are the methodological frame — the decision discipline the instruments operate under. The further-reading anchors in §B.3 are the next step beyond what the guide treats.

Most of the instruments in §B.1 are PeopleAnalyst-developed IP that has been deployed in named consulting engagements over fifteen-plus years; the underlying methodology is well-tested even when individual citations are internal-substrate-shaped rather than peer-reviewed. The frameworks in §B.2 include both PeopleAnalyst-developed frameworks and field-standard frameworks (the diffusion canon's structural-network framing; Edmondson's psychological safety; Burt's structural holes) the guide pairs with the PeopleAnalyst material.

B.1 Instruments

The measurement instruments practitioners can implement to operationalize the guide's arguments.

The 12-factor AI-readiness instrument

PeopleAnalyst's organizational AI-readiness self-assessment. Twelve dimensions, scored individually, that together describe the substrate an organization brings to AI adoption.¹ The dimensions:

AI-Driven Workforce Planning and Talent Strategy
Change Fatigue and Resistance
HR and Organizational Design for AI Integration
Human-AI Collaboration and Quality Assurance
Management Buy-In
Organizational Restructuring and Role Redesign
Technology, Management, and HR Capability Building
Transparency and Understanding of AI Decisions
Trust in AI
Viewing AI as a Strategic Partner
Workflow Mapping and Process Redesign
Workforce and Job Analysis

How to deploy: Each dimension is scored 1–5 against a behaviorally-anchored rubric. Aggregate produces a readiness score; pattern across the twelve names the binding constraints on adoption.

The network-readiness reframe. Per Part II §2.3 and Part VII §7.2, each dimension reframes to ask where in the trust graph this property lives rather than whether the organization, as an aggregate, has this property. The reframe is what makes the instrument useful for AI-adoption specifically (vs. generic organizational-readiness work). The full per-dimension reframes are at manifests/self-assessments-2026-05-12.json in the source repository.

Where the guide treats it: Part II §2.3 (the instrument itself); Part II §2.5 (its workforce-design implications); Part VII §7.2-§7.3 (the network-readiness reframe in operational form).

CAMS — Capability, Alignment, Motivation, Support

The four-factor activation framework underneath the People Analytics Toolbox's measurement spine.² Per Mike's Rapid Collaborative Impact essay: four conditions must exist for an employee or team to produce at or above expectations.

Capability — knowledge, skills, abilities; what people bring to the work.
Alignment — knowing what they're expected to accomplish and how they're performing.
Motivation — preferences, commitment, engagement; willingness to do the work.
Support — tools, resources, manager and peer relationships, absence of negative consequences.

If any one of the four is missing, reliable performance becomes difficult-to-impossible. The framework is multiplicative, not additive: high scores on three factors do not compensate for a low score on the fourth.

How to deploy: The CAMS index is an 8-item survey on a 0–10 agreement scale (one team-perspective + one individual-perspective item per factor). Total ranges 0–80 per respondent. From the index:

Activated = CAMS index ≥ 70
At-Risk = CAMS index < 60
The middle range (60–69) is Watch — neither activated nor at-risk; trending in one direction.

Where the guide treats it: Part II §2.5 (organizational-design implications); referenced in the practitioner-tool layer through Part II's workforce-design section. The CAMS framework predates the guide and is treated in fuller detail in the Rapid Collaborative Impact magazine piece at peopleanalyst.com/magazine/rapid-collaborative-impact.

NAV — Net Activated Value

The financial-translation layer for the CAMS index. Converts workforce activation status into dollar outcomes for executive consumption.³

Formula stack:

Net Activated Percent = (workforce − at-risk in workforce) ÷ total headcount
ELV (Employee Lifetime Value) = HCROI × annual cost × lifetime tenure (computed per segment)
NAV = Net Activated Percent × ELV (aggregated to organizational level)

How to deploy: NAV is the leadership-facing translation of the CAMS index. It answers the question "what's the dollar consequence of our current activation status?" without requiring the executive audience to engage with the CAMS index directly. The People Analytics Toolbox's forecasting spoke can run scenario analysis on NAV under different intervention assumptions.

Where the guide treats it: Part II §2.5; referenced in Part VII §7.3's operational-discipline section as the financial-translation layer that makes adoption-investment decisions executive-legible.

The Three A's framework

The lifecycle measurement spine: Attraction, Activation, Attrition.⁴ The three measurement domains that together cover the workforce lifecycle and define the load-bearing measurement set for any people-analytics function.

Attraction — the pre-employment relationship; sourcing pipelines, candidate quality, time-to-hire, offer-acceptance, employer brand.
Activation — the employed-and-producing-value relationship; CAMS index, NAV, engagement, performance, capability-build trajectory.
Attrition — the exit relationship; turnover rate, voluntary vs. involuntary split, regrettable attrition, retention forecasting, post-exit cohort analysis.

The framework's value is in the completeness claim: most people-analytics functions instrument fragments (often just attrition forecasting, the easiest to defend). The Three A's framing argues the load-bearing measurement set is all three, not any one. A function instrumenting only attrition can predict turnover but cannot diagnose its causes or design interventions that prevent it.

Where the guide treats it: Part II §2.5; this framework predates the guide and is treated in fuller detail in PeopleAnalyst's consulting materials and the Rapid Collaborative Impact magazine piece.

Penwright Measurement Framework

The six-dimension capability instrument underneath the Penwright Authorship Research Program at vela.⁵ Six dimensions of writer capability that change over time and that AI tools may improve, degrade, or leave unchanged:

Structural rhetoric — what the writer can deploy at the paragraph/section level.
Lexical control — vocabulary breadth and precision; appropriate-register selection.
Argumentative coherence — claim-evidence-warrant connectivity across extended prose.
Voice and signature — distinctive prose patterns that identify the writer.
Discipline — willingness to revise, cut, restate; tolerance for difficulty.
Verification fluency — ability to source-check, fact-check, calibrate.

How to deploy: Repeated measurement at session-start of writers using AI tools, tracked longitudinally. The instrument operationalizes the longitudinal test the Penwright Research Program treats as primary: is the writer better with the tool, than without it, in six months? — distinct from session-output quality, which can be high even when the longitudinal capability trajectory is degrading.

The instrument generalizes beyond writing-tool research: any AI deployment where the user's underlying capability change matters (vs. just the AI's output quality) can adapt the six-dimension structure to its domain.

Where the guide treats it: Part V §5.4 (the Penwright Research Program); Part V §5.5 (cross-domain product-feature implications); Part V §5.6 Position 4 (longitudinal-capability-instrumentation as a design constraint).

B.2 Frameworks

The methodological frames the instruments operate under. These are decision-shaping rather than measurement-shaping.

The Principal-Issues Thesis

PeopleAnalyst's load-bearing methodological argument: every domain has a principal-issues set — the small number of measurement objects that drive most of the variance — and most domains are stuck because they have not named theirs.⁶ The thesis generalizes beyond people analytics; the magazine principal-issues derives its name from this thesis.

The discipline the thesis enforces: rather than measuring everything that could be measured (the conventional analytical reflex), the practitioner identifies which measurements actually matter for the decisions at hand and instruments those first, with rigor. The non-principal measurements are documented but not instrumented until/unless the principal set is complete.

For people analytics specifically, the principal-issues set is operationalized through Three A's + CAMS + ELV + NAV (Section B.1 above). For AI adoption, Part II §2.3-§2.5 and Part VII §7.2-§7.3 name the principal set: the 12-factor instrument scored at network position; the trust-graph topology map; the seed cohort identified via psychometric + network signal; the cluster-saturation operational discipline.

Where the guide treats it: Implicit throughout; explicit in Part II §2.4 and Part VII §7.3 as the methodological discipline the operational moves enforce.

The Four-S synthesis

The four legs of disciplined people analytics: Behavioral Science + Statistics + Systems + Strategy.⁷ The framework's claim is that all four are necessary; none is sufficient. Most people-analytics functions are stuck because they have one or two but not four:

Behavioral Science — the questions to ask, how to interpret answers; the discipline that prevents data from being misread. Without it, the function produces analytical artifacts that don't translate to action.
Statistics — the calibration of signal versus noise; the discipline that prevents over-claiming. Without it, the function produces confident-looking outputs whose actual reliability is low.
Systems — the engineering capability that makes the work scale past a heroic individual. Without it, the function depends on the one analyst and breaks when they leave.
Strategy — the framing that makes the analytical work matter to the decisions the organization is actually making. Without it, the function produces analyses nobody acts on.

The four-S synthesis is the underlying explanation for why most non-Google organizations cannot reproduce Google's people-analytics outputs by copying the outputs (a slide; a model; an attrition predictor): they don't have all four legs of the underlying capability.

Where the guide treats it: Implicit in Part II's discussion of why workforce-AI adoption fails; explicit in PeopleAnalyst's consulting framing and the Rapid Collaborative Impact magazine piece.

Rapid Collaborative Impact (RCI)

The meta-methodology PeopleAnalyst applies across domains: identify the principal-issues set; ship the load-bearing measurement set fast with the four-S synthesis; prove value before requesting more.⁷ RCI generalizes the principal-issues thesis from people-analytics specifically to any domain where rigorous measurement could be applied but isn't.

How to deploy: RCI is a process methodology, not a measurement instrument. Engagement-shape: identify the principal-issues set in the first sprint; deploy the load-bearing minimum; iterate against decision-grade outcomes; scale only after the principal-issues set is operational and demonstrating value.

The methodology is anti-perfectionist by design — it ships the principal-issues set fast and treats methodology refinement as a follow-on rather than a prerequisite. The trade-off is intentional: rapid value-demonstration buys the credibility that funds deeper methodology investment.

Where the guide treats it: Implicit in the deployment-discipline framing across Part II §2.6 and Part VII §7.3.

Lean People Analytics

RCI applied to HR in resource-constrained environments — single-analyst teams; zero-budget settings; organizations that don't have Google's capability stack and never will.⁸ Lean People Analytics ships the principal-issues set (Three A's, CAMS, ELV, NAV) at minimum-viable-fidelity with the tools the analyst has available, and proves value before requesting more.

The methodology's existence proves a load-bearing claim: most of the field's cannot do people analytics condition is not actually a capability gap — it's a methodology gap. The capability to ship Lean People Analytics exists in every organization above ~500 employees; what's missing is the discipline that the principal-issues set is enough, and the executable framework that names what the principal-issues set actually is.

Where the guide treats it: Referenced in Part II §2.5 as the deployment pattern that scales people-analytics work to mid-market organizations; treated in fuller detail in PeopleAnalyst's consulting materials.

Full-Stack People Analytics Systems

The architectural framing that treats people-analytics as a software stack rather than a reporting team.⁹ The stack:

Substrate layer — data sources (HRIS, ATS, LMS, performance, engagement, exit, plus contextual data)
Normalization layer — canonical-field resolution; identity unification; cohort definition
Privacy layer — min-N gates; anonymization; access controls; protected-feedback substrate
Statistical layer — confidence intervals; effect sizes; multiple-comparison corrections
Decision-support layer — VOI analysis; structured decision frameworks; aligned-chance trees
Surface layer — dashboards, reports, narrative analyses, consulting deliverables

The framing matters because a function configured as a reporting team is naturally one-layer-deep (the surface layer); a function configured as a stack has all six layers and can serve decision-grade outputs reliably across the organization. Most stuck people-analytics functions are stuck because they're trying to do six-layer work with one-layer staffing.

The People Analytics Toolbox at peopleanalyst.com/research/pa-platform is the production implementation of this stack architecture; the segmentation-studio + data-anonymizer + calculus + forecasting spokes correspond directly to the normalization, privacy, statistical, and decision-support layers.

Where the guide treats it: Part II §2.5 (architectural framing); Part IV §4.4 (decision-support layer specifically); Part VII §7.3 (toolkit-level deployment).

The Network-Mediated Adoption framework

The four-stage adoption discipline introduced in Part VII §7.3. The structural alternative to the conventional consulting playbook (executive sponsor → uniform training → KPI dashboard).

Map the trust graph — organizational network analysis (ONA) measuring who actually influences whom, where the structural holes are, which sub-networks would receive an adoption signal and which would not. The deliverable is a topology map, not an aggregate readiness score.
Identify the seed cohort by network position + psychometric signal — the innovator + early-adopter ~16% (per Rogers's diffusion curve), selected by where they sit in the trust graph and their disposition toward adopting new practice when not required to.
Saturate the seed cohort's tie clusters — instead of expanding to the next ring after a successful pilot, accelerate adoption inside the seed cohort's immediate tie clusters until local density crosses Centola's complex-contagion threshold.
Sustain through mop-up — adoption is steady-state, not event; instrument for regression detection and re-seed sub-networks that drift back to pre-adoption patterns.

Where the guide treats it: Part VII §7.2-§7.3 (the framework as advocated); cross-referenced extensively across Parts II, III, IV, V, VI as the structural correction to the failure patterns each part documents.

B.3 Further-reading anchors — per topic cluster

The reading list, organized by which part of the guide the reading anchors. Each entry: author, title, one-line takeaway, citation. For full bibliographies see each part's ### Bibliography section.

Foundations (Part I)

Boden, Margaret. Artificial Intelligence: A Very Short Introduction. Oxford 2018. The cleanest 150-page introduction to the field in print.
Russell, Stuart, and Peter Norvig. Artificial Intelligence: A Modern Approach. 4th ed., Pearson 2020. The canonical undergraduate textbook; ~1,100 pages; mathematically rigorous.
Brynjolfsson, Erik, Danielle Li, and Lindsey R. Raymond. Generative AI at Work. NBER Working Paper 31161, 2023. The foundational empirical study of GenAI's productivity effects on knowledge work; the cognitive redistribution finding.
Shumailov, Ilia, et al. AI Models Collapse When Trained on Recursively Generated Data. Nature, 2024. The model-collapse paper; substrate-degradation evidence that grounds Part V §5.2.3.

Workforce-AI adoption (Part II)

Pereira, Elisa, Alvin Wang Graylin, and Erik Brynjolfsson. The Enterprise AI Playbook: Lessons from 51 Successful Deployments. Stanford Digital Economy Lab, April 2026. The 51-deployment study underneath the 95% failure-rate stylized fact.
MIT NANDA. The GenAI Divide: State of AI in Business 2025. August 2025. The convergent 95% finding from a different methodology — three independent reads of the same pattern.
Garvin, David A. How Google Sold Its Engineers on Management. Harvard Business Review, December 2013. The canonical Project Oxygen write-up.
Edmondson, Amy C. The Fearless Organization. Wiley, 2018. Psychological safety as the team-effectiveness load-bearing variable; the practitioner-facing synthesis of her 1999 ASQ paper.

Customer-experience and marketing AI (Part III)

Sharma, Mrinank, et al. Sycophancy in AI Assistants. 2024. The sycophancy-as-RLHF-artifact paper; explains why reasoning-personalization in customer-facing AI is structurally fragile.
Lee, Hao-Ping, et al. Confidence in Generative AI and Critical Thinking. Microsoft Research / CHI 2025. The 319-knowledge-worker / 936-task study; trust in AI inversely predicts critical-thinking effort.
The NICE × HBR webinar Using AI to Build Strong Connections With Customers — captured in the editorial substrate at content/research/ai-encyclopedia/03-encyclopedia-body.md pages 41-54. Strategic-capability framing for customer-experience AI.

Product, operations, and decision support (Part IV)

Bainbridge, Lisanne. Ironies of Automation. Automatica 19, no. 6 (1983): 775-779. The foundational paper on the human-in-the-loop architectural paradox; cited everywhere agentic-AI work is rigorous.
Howard, Ronald A., and Ali E. Abbas. Foundations of Decision Analysis. Pearson, 2016. The canonical text on Value-of-Information analysis and structured-decision frameworks underneath Part IV §4.4.
Laban, Philippe, et al. LLMs Get Lost in Multi-Turn Conversation. 2025. The ~39%-multi-turn-degradation finding; foundational for agentic-AI reliability arithmetic.

The research frontier (Part V) — the AHI program reading list

The AHI program's topic reviews are at peopleanalyst.com/research/ai-human-interaction/ and each closes with a primary-source reading list. Anchor entries:

Chen, et al. Persona Drift in LLMs Across Extended Sessions. 2024. The counter-intuitive finding that larger models drift more, not less.
Glickman, Moshe, and Tali Sharot. Human-AI Bias Amplification: A Bidirectional Loop. Nature Human Behaviour, 2024. The bidirectional-amplification result; one of the strongest empirical anchors for Part V §5.2.5.
METR. Experienced Open-Source Developers Slower with AI Tools on Familiar Repositories. 2025. The cognitive-redistribution / skill-erosion evidence from the experienced-developer side.
The AHI program's six topic-review syntheses. Long-context emergence; calibration of personalization; cognitive apprenticeship and ZPD; distributed cognition; longitudinal cognitive effects; institutional economics of AI. Each is a ~10-15 page review of the peer-reviewed literature on the corresponding concern from Part V §5.2.

Governance, privacy, compliance (Part VI)

Edwards, Lilian, and Michael Veale. Slave to the Algorithm? Why a Right to an Explanation Is Probably Not the Remedy You Are Looking For. Duke Law and Technology Review, 2017. The substrate-opacity problem treated from a regulatory perspective.
Doshi-Velez, Finale, and Been Kim. Towards a Rigorous Science of Interpretable Machine Learning. 2017. The methodology paper that defines what interpretable means in a way that scales beyond hand-waving.
The EU AI Act. Official Journal of the European Union, 2024. Risk-tiered regulation of AI systems; the operative European regulatory framework through 2027.
AHI program review: Calibration of Personalization — the paternalism-vs-autonomy spectrum operationalized.

Network-mediated adoption (Part VII) — the diffusion canon

Rogers, Everett M. Diffusion of Innovations. 5th edition. Free Press, 2003. The canonical adoption-curve framework; sixty years of replication.
Granovetter, Mark S. The Strength of Weak Ties. American Journal of Sociology 78, no. 6 (1973): 1360-1380. Most-cited sociology paper of the last five decades.
Burt, Ronald S. Structural Holes: The Social Structure of Competition. Harvard University Press, 1992. The structural-position-and-managerial-value demonstration.
Centola, Damon. The Spread of Behavior in an Online Social Network Experiment. Science 329, no. 5996 (2010): 1194-1197; and How Behavior Spreads: The Science of Complex Contagions. Princeton University Press, 2018. The empirical anchor for the cluster-saturation operational discipline.
Christakis, Nicholas A., and James H. Fowler. Connected: The Surprising Power of Our Social Networks and How They Shape Our Lives. Little, Brown, 2009. The popularization that brought network effects into the practitioner conversation.
Watts, Duncan J. Six Degrees: The Science of a Connected Age. Norton, 2003. Small-world networks and their adoption implications.
Cross, Rob, and Andrew Parker. The Hidden Power of Social Networks: Understanding How Work Really Gets Done in Organizations. Harvard Business School Press, 2004. The practitioner reference for organizational network analysis (ONA).

Cross-domain diffusion confirmation

The Namesake research surface at peopleanalyst.com/research/namesake/. Cross-domain quantitative-diffusion confirmation of the cluster-local-topology requirement, on U.S. first-name adoption data.
The Sociological Science Vol. 12 country-scale RCT (October 2025). Field-experiment replication of Centola's clustered-topology result at country-scale.

West, Michael. 12 Factors For Successful AI Adoption. PeopleAnalyst internal substrate, captured 2026-05-14, content/research/ai-encyclopedia/03-encyclopedia-body.md pages 114-141. Per-dimension self-assessments at content/research/ai-encyclopedia/manifests/self-assessments-2026-05-12.json. Network-readiness reframes per dimension at the same manifest path. ↩
West, Michael. Why People Analytics Is Stuck — and How to Unstick It. Magazine essay at peopleanalyst.com/magazine/rapid-collaborative-impact. The CAMS framework, NAV financial-translation layer, and the principal-issues thesis are treated together in this essay; the CAMS index methodology + 8-item survey instrument are documented in the practitioner appendix of PeopleAnalyst's consulting materials. ↩
NAV (Net Activated Value) formula stack is documented in the Rapid Collaborative Impact essay (see ²) and operationalized in the People Analytics Toolbox's forecasting spoke for scenario analysis. ↩
West, Michael. The Three A's framework (Attraction, Activation, Attrition) is documented in PeopleAnalyst's consulting materials and the Rapid Collaborative Impact essay. The framework predates the published essays; it was operationalized in named consulting engagements at Merck, PetSmart, and Google during the 2008-2018 period. ↩
The Penwright Measurement Framework is the AHI program's primary capability instrument, documented across the AHI program review files at content/research/ai-human-interaction/ and the Penwright research-program documents (paper-01-technical, paper-04-measurement, etc.) at the same path. The longitudinal-test methodology (better with the tool, than without it, in six months) is documented in Part V §5.4 of this guide. ↩
West, Michael. Why People Analytics Is Stuck — and How to Unstick It. See ². The principal-issues thesis is the central methodological claim of the essay and the magazine the same name (principal-issues) derives from. ↩
West, Michael. The four-S synthesis (Behavioral Science + Statistics + Systems + Strategy) and Rapid Collaborative Impact (RCI) are treated together in the Rapid Collaborative Impact essay (see ²). The synthesis predates the essay and is operationalized in PeopleAnalyst's consulting engagements. ↩ ↩²
West, Michael. Lean People Analytics For Dummies. Wiley, 2019. The full-length treatment of Lean People Analytics methodology. Mike's 2019 trade book. ↩
The Full-Stack People Analytics Systems framing is documented in PeopleAnalyst's consulting materials and operationalized in production at the People Analytics Toolbox (peopleanalyst.com/research/pa-platform/). The corresponding spokes (segmentation-studio, data-anonymizer, calculus, forecasting, conductor, et al.) implement specific layers of the stack architecture. ↩

Practitioner's Toolkit + Further Reading