# MANUAL — CRM Automation & Orchestration > Master's-degree artifact for the **crm-automator** agent. > Discipline: the DATA / TRIGGERS / ORCHESTRATION layer of the growth stack. > Sibling masters: copywriting, funnel-psychology, email-marketing. This master turns their artifacts into a running machine. --- ## 1. Mandate & Mental Model ### What this master owns The crm-automator owns the **single source of truth** on every contact and the **machine-to-machine wiring** that makes the rest of the growth stack fire. Concretely: - **The customer master** — one authoritative record per person/account, with identity resolution across 8+ platforms so a human is never two rows. - **Lead scoring & routing** — the model that decides who is worth a human's time versus an automated touch, and the rules that send them there. - **Lifecycle-stage automation** — one canonical stage property (subscriber → evangelist) with every automation keyed to a stage transition. - **Speed-to-lead SLAs** — the monitored clock between `lead_created` and `first_touch`, with automated first-touch and human escalation. - **Enrichment** — the waterfall that turns a bare email into a scored, researched, routable contact. - **Orchestration** — the wiring where *every funnel event becomes a CRM property change, and every property change can fire a sequence or a task.* The other three masters produce **artifacts** (a landing page, a funnel, an email sequence). This master produces the **nervous system** that connects them: the events, the properties, the triggers, and the hand-off contracts to the humans and agents who close. ### The core operating theory Four propositions, each traceable to a named originator (fully sourced in §2): 1. **The relationship is the unit of competition, not the transaction.** (Peppers & Rogers, *The One to One Future*, 1993.) Chase share-of-customer over a lifetime, not share-of-market. Every interaction must write back a learned attribute so the system gets smarter about each contact — the Learning Relationship that raises switching costs. 2. **One authoritative record is the asset; fragmented data destroys value.** (Tom Siebel, Siebel Systems, 1993.) The single source of truth is not a nicety — it is the entire premise of the discipline. 3. **Response is the only truth; every touch carries a trackable action.** (Lester Wunderman, who coined "direct marketing," 1961/1967.) Awareness is not the job. A named individual who responds, logged in a database, is the job. This is the philosophical parent of "every funnel event becomes a CRM property change." 4. **Orchestration is a platform, not a feature.** (Marc Benioff, Salesforce Flow lineage, 1999→.) The moat is the wiring between systems, not any single system. The property-change-fires-a-trigger primitive is what the master builds; the Sovereign CRM's machine-to-machine layer is the self-hosted analog of Salesforce Flow. **The one-sentence mandate:** *own the record, time the trigger, honor the hand-off — and never let a scored metric become a lie.* --- ## 2. The Science The intellectual bedrock. Every framework here maps to a concrete property, trigger, or SLA the master owns. Names are attributed to primary sources; where the field commonly mis-attributes something, the correction is noted. ### 2.1 The foundational legends **Don Peppers & Martha Rogers — IDIC and the Learning Relationship.** *The One to One Future* (Currency/Doubleday, 1993); *Enterprise One to One* (1997); *The One to One Fieldbook* (1999). Thesis: marketing economics invert — mass marketing chases **share of market** (one product to many strangers); one-to-one chases **share of customer** (many products to one known customer over a lifetime). Their named framework is **IDIC**: **I**dentify customers individually and addressably → **D**ifferentiate by value and by need → **I**nteract cheaply at scale → **C**ustomize the message/product. This *is* the CRM data lifecycle: identity resolution → segmentation/scoring → engagement → personalization. The **Learning Relationship**: each interaction makes the vendor smarter, so a competitor would have to re-learn from zero — a data-moat argument before the phrase existed. → *For 19Keys, the Hot 1000 / Super 10000 lists are IDIC's "Differentiate by value" step made operational.* **Robert Kestnbaum — the true origin of RFM.** RFM (Recency, Frequency, Monetary) originated with database-marketing pioneer **Robert Kestnbaum in the 1960s**, alongside early customer-lifetime-value modeling (his collaborators included a young **Robert Blattberg**). Common mis-attribution: crediting Bult & Wansbeek with "inventing" RFM — they gave it the rigorous optimization treatment three decades later, they did not invent the construct. Why it matters: RFM is the *cheapest durable scoring model that exists* — three columns already in Supabase (last-order date, order count, total spend), no ML required. It is the correct v1 score before anything fancier. **Lester Wunderman — "response as the only truth."** Coined **"direct marketing"** in a 1961 speech, popularized in a landmark 1967 MIT address; founded Wunderman. Memoir: *Being Direct* (1996). Pioneered loyalty/continuity mechanisms (Columbia Record Club, the American Express member program, the toll-free 1-800 response number, subscription cards) and the "**gold box**" activation device — a printed call-to-action that lifted response by giving the reader a job to do (a physical micro-conversion trigger). Principle: **every touch must carry a trackable call to action, and the database of responders is the asset.** **Jan Roelf Bult & Tom Wansbeek — RFM given decision-theoretic rigor.** "**Optimal Selection for Direct Mail**," *Marketing Science* 14(4): 378–394, 1995 (DOI 10.1287/mksc.14.4.378). Thesis: don't mail everyone — solve for the profit-maximizing cutoff by **equating marginal cost with marginal return**. Touch a household only when expected response value exceeds the cost of the touch. Principle: **scoring is not the goal; the action threshold is.** → *For a real CRM of ~4,591 buyers / $611K, the Bult–Wansbeek question is exact and answerable: which segment clears the marginal-cost bar for a human closer's time versus an automated email? That cutoff is a CRM property + routing rule this master owns.* **Tom Siebel — invented the CRM software category.** Siebel Systems, founded **1993** (Siebel was Oracle's top salesperson); held ~half the CRM market by ~2002; acquired by Oracle in **2006** (~$5.8B). Productized "CRM" — turned Peppers/Rogers theory into a purchasable enterprise category. Contribution: the enterprise **customer master** — the canonical account/contact/opportunity/activity data model that every CRM since (Salesforce, and by lineage the Supabase schema) still mirrors. Principle: **the single source of truth.** **Marc Benioff — CRM as cloud platform and ecosystem.** Salesforce, founded **1999** ("No Software"). *Behind the Cloud* (2009). Two ideas: multi-tenant SaaS delivery, then the bigger one — CRM as a **platform** (Force.com / AppExchange, 2005–06). "**Customer 360**" (unified profile) and the workflow/automation layer (Process Builder → **Flow**) that turns *property changes into fired actions*. Principle: **orchestration as a platform, not a feature.** **Paul Greenberg — the field's chronicler and definer.** *CRM at the Speed of Light* (1st ed. 2001 → 4th ed. 2009), "the bible of CRM." Definition: CRM is **"a philosophy and a business strategy, supported by a system and a technology, designed to improve human interactions in a business environment"** — not merely software. Gave the durable **operational / analytical / collaborative** CRM taxonomy and chronicled the shift to Social CRM. Guardrail: **automation without a relationship strategy just industrializes spam.** **Aaron Ross — Predictable Revenue (role specialization + pipeline math).** *Predictable Revenue* (2011), from building outbound at Salesforce (~$100M added recurring). "**Cold Calling 2.0**": stop making one rep do everything — **specialize into distinct roles**: SDRs/Prospectors (source & qualify), AEs/Closers (close), Account Managers/Farmers (retain & expand), plus inbound response reps. Named model "**Seeds, Nets, Spears**" (three lead sources: seeds = word-of-mouth/organic, nets = marketing/inbound, spears = targeted outbound). Pipeline arithmetic: revenue = #leads × qualification-rate × win-rate × ACV. Principle: **specialization + measurable hand-off contracts.** → *Direct parent of the 19Keys execution tier: outreach-operator (spears), funnel-closer (closer), owner-action-chaser (farmer/ops). The critical artifact is the hand-off SLA between stages — a CRM contract, not a personality.* **Mark Roberge — scoring by regression.** *The Sales Acceleration Formula* (2015), from scaling HubSpot sales 0→$100M; MIT-engineer-turned-CRO. Thesis: run sales like an **engineering system** — hire, train, and score with data and regression, not gut. Built a hiring scorecard by regressing rep attributes against quota attainment, and a **lead-scoring model** by regressing which contact attributes/behaviors predicted closing. Principle: **lead scores should be empirically fit to closed-won outcomes, not hand-assigned points.** v1 can be RFM/heuristic; v2 should be "which properties actually predicted a buyer in the $611K history." Roberge also supplies the **pipeline-hygiene discipline** (weekly stale-deal / next-step-empty / stage-age review) behind the pipeline-hygiene-cadence playbook. ### 2.2 The underlying science **A. Speed-to-lead — the empirical crown jewel.** Two distinct studies; keep them straight: 1. **2007 MIT / InsideSales.com Lead Response Management study** (Dr. **James Oldroyd**, MIT Sloan; 6 companies, ~15,000 leads, ~100,000 dials). Contacting a web lead within **5 minutes vs. 30 minutes** raises the odds of *making contact* **~100×** and the odds of *qualifying* the lead **~21×**. Best call windows ~**8–9am and 4–5pm**; best first-contact days Wednesday/Thursday. 2. **2011 HBR "The Short Life of Online Sales Leads"** (Oldroyd, **Kristina McElheran**, **David Elkington**; *HBR* 89(3), March 2011; audit of 2,241 US firms). Only **37%** responded within 1 hour; **23% never responded at all**; average first response **42 hours**. Firms responding within an hour were **~7× more likely to qualify** than those waiting one more hour, and **~60×** more than firms waiting 24h+. The durable law: online-lead value decays on the order of *minutes*, and the median company is catastrophically slow. → *The speed-to-lead SLA is a hard, monitored CRM property: timestamp `lead_created`, timestamp `first_touch`, alarm on breach (target sub-5-min automated ack, sub-1-hour human). "$611K / 4,591 buyers / 0 emails ever sent" is the 42-hour problem taken to infinity — this is where the leaked money is.* **B. RFM — the durable segmentation baseline.** Recency, Frequency, Monetary (origin Kestnbaum 1960s; optimization Bult & Wansbeek 1995). Survives 60 years because it needs no training data and **Recency is consistently the strongest single predictor** of next-purchase probability in transactional data. Operationalization — **Arthur Middleton Hughes**, *Strategic Database Marketing*: sort the file into quintiles on each dimension → 5×5×5 = 125 cells, act per cell. A ready-made segment table for Supabase and the input to lifecycle routing. **C. Probabilistic CLV — Peter Fader and "Buy Till You Die."** **Peter Fader** (Wharton), *Customer Centricity* (2011): not all customers are equal — find your best and build the firm around them (the analytic backbone of Hot 1000 / Super 10000). Model lineage: **Pareto/NBD** (Schmittlein, Morrison & Colombo, 1987 — the original "Counting Your Customers"; latent purchase rate + latent dropout; powerful but painful) → **BG/NBD** (**Fader, Hardie & Lee, 2005**, "'Counting Your Customers' the Easy Way," *Marketing Science* 24(2): 275–284; same story, spreadsheet-fittable; the practical standard), paired with the **Gamma-Gamma** model for monetary value. What it adds over RFM: RFM *ranks*, BTYD *predicts* — P(alive), expected future transactions, expected residual CLV per contact. A forward-looking property that drives how much of a closer's time a contact is worth (ties back to the Bult–Wansbeek marginal cutoff). Tooling: open-source `lifetimes` / PyMC-Marketing run BG/NBD + Gamma-Gamma directly on a Supabase transactions export — a build-don't-spend path. **D. The Demand Waterfall — stage-conversion benchmarks.** **SiriusDecisions Demand Waterfall** (~2006; SiriusDecisions acquired by **Forrester** in 2018; superseded internally by the 2017 Demand Unit Waterfall). Leads flow through named, measurable stages: **Inquiry → MQL → SQL/SAL (Sales Accepted) → SQO (Sales Qualified Opportunity) → Closed/Won**, each with a **conversion rate** and a **hand-off definition**. This *is* the lifecycle-stage model: every stage is a CRM property value; every transition is a trigger and a hand-off contract. Map: funnel-page event → Inquiry; scoring threshold → MQL; outreach-operator accepts → SAL; funnel-closer works → SQO. **E. BJ Fogg — B = MAP (the psychology of trigger timing).** **Behavior = Motivation × Ability × Prompt**, all three present *at the same moment* (Stanford Behavior Design Lab; *Tiny Habits*, 2019; earlier stated B=MAT). A **Prompt only works at a moment of sufficient Motivation and Ability.** Fire too early and you burn the trigger; lower the ability cost (make the next step trivially easy) so a smaller prompt converts. → *Triggers must be timed to ability/motivation windows, not raw events. Speed-to-lead is B=MAP: the moment after a form-fill is peak motivation — the prompt must arrive before it decays. Design every step to be maximally easy (one-click reply, pre-filled booking) to raise the Ability term.* **F. Cialdini — Commitment & Consistency (micro-conversion staging).** *Influence: The Psychology of Persuasion* (Robert Cialdini, 1984). The load-bearing principle for lifecycle design is **Commitment & Consistency**: people act consistently with prior voluntary commitments, especially small ones. Named tactic **foot-in-the-door**: a small yes makes the next larger yes more likely. → *Design the lifecycle as an escalating ladder of micro-conversions (open → click → free artifact → booked call → purchase), each a small commitment. Maps to 19Keys' own doctrine: sell the outcome artifact (Builder Passport) as the small yes before the month/kicker.* Also relevant: **Reciprocity** (lead with the free artifact) and **Social Proof** (buyer counts in outreach). **G. Goodhart's Law — why scored metrics get gamed.** "**When a measure becomes a target, it ceases to be a good measure.**" (Economist **Charles Goodhart**, 1975; sharpened by **Marilyn Strathern**, 1997; cousin **Campbell's Law**.) Every scoring/SLA system *will* be gamed: optimize "MQLs generated" → marketing loosens the MQL definition; optimize "speed-to-lead" → reps auto-dial to stop the clock then hang up; optimize "activities logged" → busywork inflation. Countermeasures (this master's responsibility as metrics owner): 1. **Pair every throughput metric with a quality counter-metric** (MQL volume ↔ MQL→SQO conversion; speed-to-lead ↔ connect-rate). 2. **Score on outcomes, not proxies** where possible (Roberge: fit to closed-won). 3. **Hold definitions in the CRM, not in reps' heads** — a stage transition requires objective, system-verified criteria so it can't be self-declared. 4. **Rotate/audit** what you optimize on so the system can't calcify around one gameable number. ### 2.3 Crosswalk — science → the primitive it dictates | Foundation / law | The CRM primitive it dictates | |---|---| | Peppers & Rogers IDIC | Identity resolution + write-back-on-every-interaction (8-platform sync) | | Kestnbaum / Bult–Wansbeek RFM + marginal cutoff | v1 lead/customer score + the routing *threshold* (human vs. automated) | | Wunderman "response as truth" | Every funnel event → a tracked CRM property change | | Siebel single source of truth | The Supabase customer master schema | | Benioff platform/Flow | The property-change-fires-a-trigger orchestration engine | | Aaron Ross role specialization | Hand-off SLA contracts to outreach-operator / funnel-closer / owner-action-chaser | | Roberge regression scoring | v2 lead score fit to the $611K closed-won history | | Oldroyd speed-to-lead | The monitored first-touch SLA (sub-5-min ack) — the biggest leak | | SiriusDecisions Waterfall | The lifecycle-stage model + per-stage conversion metric | | Fogg B=MAP | Trigger *timing* (peak-motivation window) + easy next step | | Cialdini commitment | Escalating micro-conversion ladder in sequences | | Goodhart | Counter-metrics + outcome-based scoring + system-verified stage definitions | --- ## 3. The Architecture The canonical structure of a best-in-class CRM automation system, component by component, with the WHY behind each. This is the reference architecture the master builds toward — a self-hosted analog of the Salesforce/HubSpot stack. ### 3.1 The customer master (system of record) **What:** one authoritative row per person, with a stable internal ID, and per-account grouping. **Why:** Siebel's single-source-of-truth thesis — fragmented data destroys value. Everything downstream (scoring, routing, journeys) reads this one record. **Components:** canonical identity key; contact + account objects; an activity/event log (append-only); a properties layer (scores, stages, RFM cells, enriched fields). The record must be *write-back-enabled* per Peppers & Rogers — every interaction deposits a learned attribute. ### 3.2 The event spine (track / identify) **What:** a standardized event schema — `identify` (who is this) and `track` (what did they do) — carrying every funnel-page, email, and product signal. **Why:** Wunderman's "response as truth" + the modern CDP pattern (Segment/Customer.io): email, tasks, and ads all react to the *same* event stream, so you wire once and everything downstream subscribes. **Rule:** no event without a schema; every funnel event maps to a named event type and a resulting property change. ### 3.3 Identity resolution **What:** the logic that merges rows across 8+ platforms into one person (email, phone, handle, device). **Why:** IDIC's "Identify" step — a human must never be two records or scoring and de-dupe both break. **Components:** deterministic match keys (email/phone), a merge-survivorship rule (which value wins on conflict), and a dedupe pass on every inbound sync. ### 3.4 The scoring layer **What:** two scores per contact — **explicit/fit** (firmographic/demographic, "should we talk to them") and **implicit/behavior** (engagement, "are they warming up") — plus, where data allows, a **BTYD CLV** property. **Why:** Marketo/SiriusDecisions dual-axis model; behavior scores **decay** (Fogg motivation is perishable); Roberge says fit the weights to closed-won. **Components:** fit criteria table, behavior criteria table with a **decay rule**, negative-scoring entries, an **MQL threshold**, and a Goodhart counter-metric. ### 3.5 The lifecycle-stage property (the spine of automation) **What:** ONE canonical, ordered stage property (subscriber → lead → MQL → SAL → SQO → customer → evangelist), each with objective entry criteria. **Why:** SiriusDecisions Waterfall + the HubSpot model — every automation is keyed to a stage *transition*, and **no automation exists without a stage-change contract.** This is what prevents a chaos of overlapping automations. **Components:** stage list, per-stage entry criteria, transition triggers, per-transition metric. ### 3.6 The trigger/automation engine **What:** the Benioff/Flow primitive — a property change or named event fires a sequence or a task. **Why:** orchestration is the platform. **Every automation must be specified** (see §8 rubric): name, trigger, entry conditions, action steps, mandatory EXIT/suppression condition, failure path, idempotency note, owner, kill switch. ### 3.7 The routing layer **What:** rules that send a qualified contact to the right destination — an automated sequence or a specific human/agent — with an SLA timer. **Why:** Bult–Wansbeek marginal cutoff (is this contact worth a human's time?) + Aaron Ross role specialization (which role gets it?). **Components:** threshold logic, assignment rules, SLA timer, escalation path. ### 3.8 The enrichment waterfall **What:** a cascade of cheap→expensive data providers that fills missing fields before scoring/routing. **Why:** the Clay school (Amin/Nowoslawski) — don't pay the expensive provider until the cheap one misses; branch personalized outbound on the enriched fields. **Components:** provider cascade with stop-on-hit, cost ceiling per contact, field-write-back. ### 3.9 The SLA/monitoring layer **What:** the clocks — chiefly speed-to-lead (`lead_created` → `first_touch`) — with breach alarms. **Why:** Oldroyd. The single highest-ROI automation in the stack. **Components:** timestamp pair, threshold, alarm/escalation, dashboard metric. ### 3.10 The hygiene & audit layer **What:** scheduled sweeps for stale deals, empty next-steps, stage-age, and pipeline coverage. **Why:** Roberge's engineering discipline — the CRM decays without maintenance. **Components:** query set, thresholds, a human review cadence, ranked action output. ### 3.11 The integration fabric **What:** the documented syncs between funnel pages, email engine, product, ads, and the CRM. **Why:** the moat is the wiring. **Components (per integration):** systems + direction, field-mapping table, auth via **env var / vault reference (never inline secrets)**, sync frequency, dedupe rule, retry/failure behavior, rollback. --- ## 4. The Schools / Styles The distinct named doctrines from the top practitioners AND top companies. A master knows which one it is running and does not mix their metrics. ### The five schools currently winning | School | Core belief | Signature primitive | Who carries it | |---|---|---|---| | **The Sales Machine** (metrics-driven demand→SLA) | Sales is an engineering problem; every stage has a conversion rate and an owner | Marketing→Sales SLA + regression-hired reps | **Mark Roberge**, **Jason Lemkin** | | **Revenue Architecture** (systems thinking) | The funnel is a *bowtie*; post-sale (retention/expansion) is half the machine; one shared vocabulary across GTM | Bowtie data model + **SPICED** qualification | **Jacco van der Kooij** (Winning by Design) | | **Signal-based / Waterfall outbound** (data engineering) | Stop spraying; enrich every record through many providers, let AI research each account, trigger on *signals* not lists | Enrichment waterfall + AI research agent (Claygent) | **Kareem Amin**, **Eric Nowoslawski**, **Guillaume Cabane** | | **PLG / Product-Qualified** (usage as the lead source) | The product generates the best leads; a PQL beats an MQL | PQL scoring on product events | **Kyle Poyar** (OpenView / Growth Unhinged) | | **Dark-funnel / Demand-creation** (attribution skepticism) | Multi-touch attribution is a lie; measure with self-reported attribution + incrementality; create demand, don't just capture it | Self-reported attribution on the form | **Chris Walker** (Refine Labs), **Jon Miller** (ABM lineage) | **Per-school detail:** **The Sales Machine — Mark Roberge / Jason Lemkin.** *Thesis:* treat GTM as a repeatable formula with a hard Marketing→Sales SLA (marketing owes N qualified leads, sales owes a response-time and work-rate in return). *Wins when:* you have volume and a definable ICP and need predictability. *Fails when:* the deal is bespoke/relationship-led or data is too thin to regress. *For 19Keys:* the SLA contract between the funnel and the closers is exactly this school. **Revenue Architecture — Jacco van der Kooij (Winning by Design).** *Thesis:* the funnel is a **bowtie** — acquisition is only the left half; onboarding, retention, and expansion are the right half and where recurring value compounds. One vocabulary (**SPICED**: Situation, Pain, Impact, Critical event, Decision) shared across all GTM roles. *Wins when:* recurring revenue, expansion motion, high LTV. *Fails when:* one-shot transactional sale with no expansion surface. *For 19Keys:* the farmer role (owner-action-chaser) and the winback engine live on the right side of the bowtie. **Signal-based / Waterfall outbound — Kareem Amin & Eric Nowoslawski (Clay lineage), Guillaume Cabane.** *Thesis:* never send a bare list to a sequence — enrich each record through a cascade of providers, have an AI agent research the account, and trigger only on real signals (job change, funding, product usage). *Wins when:* outbound to a definable account universe with public signals. *Fails when:* signals are absent or the audience is broad consumer. *For 19Keys:* this is the clay-enrichment-waterfall playbook feeding outreach-operator. **PLG / Product-Qualified — Kyle Poyar.** *Thesis:* the product is the best lead source; a **PQL** (usage threshold crossed) outperforms an **MQL** (form filled). Combine usage + firmographic + intent. *Wins when:* there is a product with usage telemetry (a free tier, a tool, an app). *Fails when:* there is no product surface generating events. *For 19Keys:* ZIION funnel tools and the Cognitive Wealth Key assessment emit usage events that can be PQL signals. **Dark-funnel / Demand-creation — Chris Walker, Jon Miller.** *Thesis:* multi-touch attribution over-credits the last-click capture channel and misses demand *creation*; measure with self-reported attribution ("how did you hear about us?" on the form) and incrementality tests. *Wins when:* brand/organic/community is a real driver and attribution is lying to you. *Fails when:* you need deterministic channel ROI for paid optimization. *For 19Keys:* a self-reported-attribution field on every funnel form is the cheap, honest instrument. **The 2026 convergence.** The winning stack is now **event-first + signal-triggered + AI-enriched**: a CDP-style event spine feeds a graph-shaped CRM; signals (funnel event, product usage, intent) fire enrichment + sequences; humans only touch pre-qualified, pre-researched contacts inside a strict speed-to-lead SLA. This is the architecture the Sovereign CRM builds toward. ### The platform lineage (companies, not just people) - **Siebel** — invented the customer master data model. - **Salesforce (Benioff)** — SaaS delivery + the platform/Flow orchestration primitive + Customer 360. - **HubSpot** — the canonical **lifecycle-stage** property model and the "no automation without a stage contract" discipline (Roberge's home turf). - **Marketo** — the dual explicit/implicit **lead-scoring** model and the demand-waterfall operationalization. - **Segment / Customer.io** — the CDP **track/identify** event schema and journey builders. - **Clay** — the **enrichment waterfall** + AI research agent pattern. --- ## 5. Decision Frameworks How the master chooses which approach/style for a given situation. ### 5.1 Which scoring model? - **No behavioral data yet, transactional history exists** → **RFM v1** (Kestnbaum). Three Supabase columns, ship today. - **Rich behavior + a closed-won history to fit against** → **Roberge regression v2** (fit weights to what actually predicted a buyer in the $611K history). - **Need forward-looking value per contact (who deserves a closer's time)** → **BG/NBD + Gamma-Gamma CLV** (Fader) on the transactions export. - **B2B with public firmographic signals** → **explicit/fit + implicit/behavior dual score** (Marketo), MQL threshold fires routing. ### 5.2 Which lead source / role model? Use Aaron Ross's **Seeds, Nets, Spears** to classify the source, then route to the matching role: - **Seeds** (organic/referral) → warm, low-friction nurture; often straight to a closer. - **Nets** (inbound/marketing) → speed-to-lead SLA + scoring gate → MQL → SAL. - **Spears** (targeted outbound) → enrichment waterfall → outreach-operator. ### 5.3 Which trigger basis — form-fill or product signal? - **There is a product surface emitting usage events** → **PLG/PQL** (Poyar): fire on behavior thresholds, not form fills. - **No product telemetry, capture is form-based** → lifecycle-stage + lead-scoring on form/email behavior. ### 5.4 By audience awareness (Schwartz, borrowed from the copywriting master) Eugene Schwartz's five awareness stages and market sophistication govern *what the trigger says*, and therefore *which sequence a stage transition fires*: - **Unaware / Problem-aware** → lead with the free artifact (Cialdini reciprocity); long nurture ladder of micro-conversions. - **Solution / Product-aware** → shorter path to the booked call / offer. - **Most-aware** → straight to the offer; suppress nurture (a mandatory EXIT condition). ### 5.5 By price point (Bult–Wansbeek marginal cutoff) - **High-value contact/segment (clears the marginal-cost bar for human time)** → route to a human closer (funnel-closer) with a tight SLA. - **Below the cutoff** → automated sequence only; a human touch would lose money on that contact. ### 5.6 By attribution reliability - **Paid channels you must optimize deterministically** → keep multi-touch tracking. - **Brand/organic/community is a real driver** → add Chris Walker's self-reported-attribution field; trust it over last-click. --- ## 6. The Multiple Ways to Run It (Playbooks) Nine repeatable builds. Each maps to one file under `playbooks/`. Each is a distinct machine with its own trigger, artifact, and success metric. ### 6.1 `lead-scoring-routing` — the Marketo / SiriusDecisions school Build explicit (fit) + implicit (behavior) scores; set an MQL threshold; write routing rules with an SLA. Scores **decay**; thresholds get revisited quarterly. Deliverable: a **lead-scoring-model** doc (see §8 rubric). *Origin:* Marketo dual-axis model + SiriusDecisions Demand Waterfall; scoring-to-closed-won fitting per Roberge. ### 6.2 `speed-to-lead-sla` — the Oldroyd play Every inbound lead gets an automated first-touch within **5 minutes** and a human task with an escalation timer. The highest-ROI automation in the stack. Deliverable: an **automation-spec** + a **journey-map** stage with minutes stated. *Origin:* 2007 MIT/InsideSales study (100× / 21× / 5-min) and 2011 HBR (42-hour / 7× / never-responded). *Fogg note:* the automated ack must land in the peak-motivation window and offer a maximally-easy next step (one-click booking). ### 6.3 `predictable-revenue-pod` — Aaron Ross Separate roles (prospector, closer, farmer) with the CRM **enforcing** the hand-off stages and pipeline-coverage math per role. Deliverable: a **journey-map** with per-stage hand-off contracts + a **pipeline-audit** with coverage ratios. *Origin:* *Predictable Revenue* role split + Seeds/Nets/Spears. *For 19Keys:* prospector = outreach-operator, closer = funnel-closer, farmer = owner-action-chaser. ### 6.4 `rfm-winback-engine` — Bult & Wansbeek applied (RFM base by Kestnbaum/Hughes) Score the buyer base on recency/frequency/monetary, auto-assign segments (5×5×5 or simplified tiers), trigger tiered winback and VIP tracks. Built for the **4,591-buyer** list. Deliverable: a **segmentation-plan** with numeric R/F/M boundaries + an **automation-spec** per track. *Origin:* RFM (Kestnbaum), quintile operationalization (Arthur M. Hughes), profit-optimal cutoff (Bult–Wansbeek). *Layer:* the right half of van der Kooij's bowtie (retention/reactivation). ### 6.5 `clay-enrichment-waterfall` — Kareem Amin / Eric Nowoslawski Cascade cheap→expensive enrichment providers per contact (stop on first hit, cost ceiling), then branch personalized outbound on the enriched fields. Deliverable: an **integration-runbook** (provider cascade, field map, vault-referenced auth) + an **automation-spec** for the branch. *Origin:* the Clay signal-based/waterfall school; Cabane's signal-triggering. ### 6.6 `plg-product-signal-crm` — Kyle Poyar Pipe product usage events into the CRM as **PQL** signals; sales touches fire on behavior thresholds, not form fills. Deliverable: a **journey-map** keyed to product events + an **automation-spec** with the threshold. *Origin:* Poyar's PLG / product-qualified-lead school. *For 19Keys:* ZIION tools + Cognitive Wealth Key emit the events. ### 6.7 `lifecycle-stage-orchestration` — the HubSpot model One canonical lifecycle property (subscriber → evangelist), every automation keyed to stage transitions, **no automation without a stage-change contract**. Deliverable: a **journey-map** (named ordered stages with entry criteria) that indexes every automation-spec touching the journey. *Origin:* HubSpot lifecycle model; Roberge's discipline; SiriusDecisions stage definitions. ### 6.8 `cdp-event-journeys` — the Segment / Customer.io pattern Standardize a `track`/`identify` event schema; feed journey builders so email, tasks, and ads all react to the same event stream. Deliverable: an **integration-runbook** (event schema + field map) + **journey-map**. *Origin:* the CDP event-spine pattern; the 2026 event-first convergence. ### 6.9 `pipeline-hygiene-cadence` — Mark Roberge discipline Weekly automated pipeline audits (stale-deal flags, next-step-empty flags, stage-age alerts) feeding a human review cadence. Deliverable: a **pipeline-audit** with sourced numbers + a ranked owner-assigned action list. *Origin:* Roberge's engineering-system discipline from *The Sales Acceleration Formula*. --- ## 7. Metrics & Instrumentation What good looks like numerically, the benchmarks, and how to know it's working. ### 7.1 Speed-to-lead (the crown-jewel metric) - **Automated first-touch:** target **< 5 minutes** (Oldroyd 100× contact / 21× qualify vs 30 min). - **Human first-touch:** target **< 1 hour** (HBR 7× qualify vs waiting one more hour; 60× vs 24h+). - **Benchmark to beat:** the median firm's **42-hour** average and **23% never-respond** rate — beating these is the fastest ROI in the discipline. - **Instrument:** `first_touch − lead_created` distribution (track the p50 and p90, not just the mean — the mean hides slow-tail leaks). ### 7.2 Stage-conversion rates (the waterfall) Per SiriusDecisions, measure a conversion rate on every transition: Inquiry→MQL, MQL→SAL, SAL→SQO, SQO→Won. Watch **trend vs prior period**, not absolute numbers. A dropping MQL→SQO with rising MQL volume is the classic Goodhart tell (definition got loosened). ### 7.3 Scoring health - **MQL→SQO conversion** is the quality counter-metric for MQL volume. - **Score decay** must be visible (behavior scores that never fall are broken). - **Negative-signal fire rate** — the model must actually be de-scoring some contacts, or it isn't discriminating. ### 7.4 Pipeline hygiene (Roberge) - **Stale-deal count** (age past a stated threshold). - **Next-step-empty count.** - **Pipeline coverage ratio** vs target (typically 3–4× for the period, but state your own target). - **Stage-age** outliers. ### 7.5 RFM / CLV health - **Segment migration** — contacts moving up/down R/F/M tiers period over period (the winback engine's proof of life). - **P(alive)** distribution from BG/NBD — a rising dead-tail is the winback trigger population. ### 7.6 Data quality (the foundation) - **Duplicate rate** (identity resolution working?) — target near zero after dedupe. - **Enrichment fill rate** per field. - **Sync failure/retry counts** per integration. ### 7.7 The Goodhart guardrail (mandatory pairing) | Throughput metric | Paired quality counter-metric | |---|---| | MQLs generated | MQL→SQO conversion | | Speed-to-lead (clock stopped) | Connect / meaningful-conversation rate | | Activities logged | Stage-progression rate | | Emails sent | Reply / positive-reply rate | --- ## 8. Failure Modes & the "No Bullshit" Bar The rubric the verify gate enforces, and the common ways practitioners fake competence. ### 8.1 How practitioners fake competence (the tells) - **A "scoring model" with no decay** — behavior points that only ever accumulate. A contact who engaged once in 2023 outranks a hot lead from today. Fails. - **A score with no negative signals** — everything adds points, nothing subtracts. It's a vanity number, not a discriminator. Fails. - **A threshold with no action** — "MQL = 50 points" with no statement of what firing at 50 *does*. A number with no consequence. Fails. - **An automation with no exit** — a sequence that can loop, double-fire, or email a contact who already converted. The single most common real-world CRM disaster. Fails. - **Prose "segments"** — "engaged power users" with no machine-evaluable rule. Can't be built. Fails. - **Untagged numbers in an audit** — "we have ~200 stale deals" with no query behind it. Invented confidence. Fails. - **Inline secrets in a runbook** — an API key pasted into the doc. Security failure AND a competence tell. Hard fail. - **A metric with no counter-metric** — a Goodhart trap shipped as a feature. ### 8.2 The verify-gate rubrics (per deliverable type) **`automation-spec`** — every automation defined as: name, trigger (named event or property change), entry conditions, action steps, **EXIT/suppression condition (mandatory — a spec with no exit fails)**, failure path (what happens on step error), idempotency note (what prevents double-fire), owner, and kill switch. *Linter fails on:* missing exit, missing failure path, missing idempotency note, or any step that emails a contact without referencing a compliance-checked template ID. **`lead-scoring-model`** — must contain: an explicit-fit criteria table with point values; an implicit-behavior criteria table with point values and a **DECAY rule** (no decay = fail); an MQL threshold **paired with the routing action it fires**; **negative-scoring entries (≥ 2 — a model with no negatives fails)**; a review-cadence line; and a **Goodhart note** (one named way the score could be gamed + a counter-metric). *Fails on:* missing decay, missing negatives, or missing threshold-action pairing. **`segmentation-plan`** — segments defined with: name; **exact machine-evaluable rule (property logic, not prose — prose-only rules fail)**; size estimate with data source; intended treatment per segment; and a mutual-exclusivity/overlap declaration. Must include an **RFM base layer with R, F, M boundaries stated numerically** and a **sunset/dead segment**. *Fails on:* rule-less segments or an absent RFM layer. **`journey-map`** — must specify: lifecycle stages (named, ordered, with **entry criteria per stage**); the trigger inventory (every automation touching the journey listed with its spec-file reference); **per-stage SLA where a human is involved (the speed-to-lead stage must state minutes)**; **conflict rules** (what happens when two automations target one contact); and instrumentation (a per-transition metric). *Fails on:* stages without entry criteria, missing SLA on the inbound stage, or missing conflict rules. **`pipeline-audit`** — reports, each number **tagged with a query/source**: stale-deal count (age threshold stated), next-step-empty count, stage-conversion rates vs prior period, pipeline-coverage ratio vs target, and top-5 at-risk deals with reason codes. **Every number requires a source tag (SQL / report name); untagged numbers fail.** Must end with a ranked action list, each action assigned to a named owner or agent (funnel-closer, outreach-operator, human). **`integration-runbook`** — per integration: systems and direction of sync; **field-mapping table** (source field → destination field, transform noted); auth method — **SECRET VALUES BANNED: any string matching a key/token pattern fails the linter hard; must reference an env var or vault name**; sync frequency/trigger; **dedupe rule**; failure/retry behavior; and rollback procedure. *Fails on:* missing field map, inline secrets, or an absent dedupe rule. ### 8.3 The governing bar Nothing ships that a machine cannot execute and a reviewer cannot verify. Every automation has an exit. Every score decays and can go negative. Every audit number has a query. Every secret is a reference. Every metric has a counter-metric. If any of these is missing, it is not done — it is a draft that *looks* done, which is exactly the failure mode this gate exists to catch. --- ## 9. Bridge to the 19Keys Stack How this master plugs into the real assets. Concrete, not hand-wavy. ### 9.1 The Sovereign CRM (Supabase `YOUR_SUPABASE_PROJECT_REF`) This is the customer master (§3.1) — the single source of truth across 8+ platforms. The master owns its schema and the property layer on top of it: - **Identity resolution** runs on every inbound sync from the 8+ platforms; dedupe rule is mandatory (§8 integration-runbook rubric). - **The scoring properties** (fit, behavior-with-decay, RFM cell, BTYD CLV) live as columns/derived tables here. - **The lifecycle-stage property** is the single canonical `lifecycle_stage` on the contact — every automation keys off its transitions. - **Hot 1000 / Super 10000 lists** are IDIC "Differentiate by value" made operational: they are the top segments of the RFM/CLV layer, and they get materially different cadence and offers (Peppers & Rogers). The rfm-winback-engine playbook (§6.4) is built directly on the **4,591-buyer / $611K** base. - **CLV** can be computed with open-source `lifetimes` / PyMC-Marketing on a Supabase transactions export (build-don't-spend), giving P(alive) and residual-value per contact to drive the human-vs-automated routing cutoff. ### 9.2 The ZIION funnels Every ZIION funnel page is an **event source** feeding the event spine (§3.2). Per Wunderman, every funnel event becomes a CRM property change: - Funnel-page submit → `track` event → `lifecycle_stage` set to Inquiry → scoring runs → threshold may fire MQL → routing. - Add Chris Walker's **self-reported-attribution field** on every ZIION form (cheap, honest instrument). - ZIION tools + the Cognitive Wealth Key assessment emit **usage events** → the plg-product-signal-crm playbook (§6.6) turns these into PQL signals. ### 9.3 The email engine (Resend + `outreach_worker`) The email engine is the primary *action* fired by triggers. The master owns the plumbing, not the copy (that's the email-marketing / copywriting masters): - Every automation step that emails a contact **must reference a compliance-checked template ID** (§8 automation-spec rubric) — the master enforces this, the copy masters supply the templates. - `outreach_worker` is the send/queue plumbing; the master specifies *when* it fires (trigger), *to whom* (routing/segment), and the mandatory **exit/suppression** (don't email a converted or unsubscribed contact). - **CAN-SPAM note:** the real blocker on record is a missing `OUTREACH_POSTAL_ADDR`. The master treats compliance state as a hard entry condition — an automation that would send without a compliant template + physical-address footer fails the gate. This is why "$611K / 4,591 buyers / 0 emails ever sent" is a *governance* problem the master unblocks, not just a wiring problem. - **Secrets** (Resend key, etc.) are referenced by env var / vault name in every integration-runbook — never inline (§8 hard fail). ### 9.4 The execution-tier agents (the hand-off contracts) Per Aaron Ross role specialization, the master does NOT close — it enforces the **hand-off SLA contracts** that route work to the human/agent tier: - **outreach-operator** = the prospector (Ross "spears"). Receives enriched, scored, pre-researched contacts from the clay-enrichment-waterfall (§6.5). Hand-off contract: only contacts above the fit threshold, with enrichment fields populated. - **funnel-closer** = the closer. Receives SQO-stage contacts under a tight speed-to-lead SLA (§6.2). Hand-off contract: `lifecycle_stage = SQO` + a stated minutes SLA + a populated next-step. - **owner-action-chaser** = the farmer / ops (right half of van der Kooij's bowtie). Receives retention/winback and owner-action tasks. Hand-off contract: stale-deal and next-step-empty flags from the pipeline-hygiene-cadence (§6.9), each tagged with a source query and a reason code. - Every hand-off is a **stage-change contract** (HubSpot discipline) with objective, system-verified criteria — so no agent can self-declare a stage and game the metric (Goodhart countermeasure §2.2.G). ### 9.5 The wiring, end to end `ZIION funnel event → event spine (track/identify) → Sovereign CRM property change → identity resolution + enrichment waterfall → scoring (fit + decaying behavior + CLV) → lifecycle_stage transition → trigger engine → { Resend/outreach_worker automated sequence } OR { routing → execution-tier agent with SLA } → pipeline-hygiene audit → owner-action-chaser.` Every arrow is a specified automation (§8), every property change is logged, every human touch is timed, and every scored number has a counter-metric. --- ### Appendix — Sources - The Short Life of Online Sales Leads — *HBR*, March 2011 (Oldroyd, McElheran, Elkington). - MIT / InsideSales.com Lead Response Management Study, 2007 (James Oldroyd). - Bult & Wansbeek, "Optimal Selection for Direct Mail," *Marketing Science* 14(4):378–394, 1995. - Fader, Hardie & Lee, "'Counting Your Customers' the Easy Way" (BG/NBD), *Marketing Science* 24(2):275–284, 2005. - Schmittlein, Morrison & Colombo, "Counting Your Customers" (Pareto/NBD), 1987. - Peppers & Rogers, *The One to One Future*, 1993. · Cialdini, *Influence*, 1984. · Fogg, *Tiny Habits*, 2019. · Ross, *Predictable Revenue*, 2011. · Roberge, *The Sales Acceleration Formula*, 2015. · Greenberg, *CRM at the Speed of Light*, 2001→. · Fader, *Customer Centricity*, 2011. · Hughes, *Strategic Database Marketing*. **Attribution corrections carried from research:** RFM origin = **Robert Kestnbaum (1960s)**, not Bult & Wansbeek (who gave the optimization treatment). Speed-to-lead numbers are two studies, un-conflated: the **2007 MIT/InsideSales** study (100× / 21× / 5-min) vs. the **2011 HBR** article (42-hour / 7× / never-responded). Fogg's model is the current **B=MAP** form.