Retention
The keep-customers brain (guidance-only): branches on business model FIRST (agency | saas_b2b | b2c — three different playbooks; a discount saves a B2C buyer and gets an agency fired), scores churn risk on the model's own signals, picks the right save move, schedules the timed sequence, and hands the draft to its peers to send. Picks up where sales-pipeline ends. No callable API — the agent's own LLM runs the decision tree in SKILL.md, composing @di-atomic/outbound-engine (sequence) + smartlead (send) + copy-engine (copy) + voice-builder (register) + @spideriq/publish-skills (SpiderFlow timing) + the OPVS coordination peers (state/recall). Backed by 70 metric-bearing real exemplars + 21 learnings + 4 enforcement scripts. Compliance-aware (CA ARL single-offer + regulated-product claim limits).
Every churn tool I evaluated sold me the same motion: spot the leaver, offer a discount, dun the failed cards. I nearly shipped that, then checked it against my own book of business and found it was wrong for two of my three customer types. A discount rescues a consumer who drifted off. Offered to an agency client who is quietly unhappy, the same discount tells them you knew the work was not worth the price. So retention branches on business model before anything else. Agency churn is a slow fade you score on relationship signals, not logins, and 51% of clients leave within a year of their champion. SaaS churn turns on activation and reason code, where frustration churn needs a conversation rather than a price cut. B2C churn is about recency, on a 60 to 90 day win-back window. Worked example: a six-month client, quiet for 14 days, champion departed. The agency branch returns an executive review, not a discount, and schedules it. Guidance-only. It decides and composes, then hands drafting to outbound-engine, sending to smartlead, copy to copy-engine, register to voice-builder, and timing to SpiderFlow.
Features5
Routes agency, SaaS and B2C down separate playbooks before proposing any save move.
Sequences dunning ahead of save offers: 20-40% of churn is an expired card.
Surfaces pause first, where roughly a quarter of would-be cancellers stay.
Four scripts catch agency discounts and referral asks aimed at unhappy customers.
Every exemplar carries a measured number and a business-model tag.
Videos1 video
DemoRetention · Demo video
Demo · SpiderMediaArticles about Retention2 articles
Retention isn't one playbook. It's three, and they disagree.
opvs.ai22 Jul 2026Retention isn't one playbook. It's three, and they disagree.
Every churn tool sells the same motion: spot the leaver, offer a discount, dun the failed cards. I nearly shipped that, then checked it against my own book of business and found it was wrong for two of my three customer types. A discount rescues a consumer and gets an agency fired.
Videos1 video
DemoRetention · Demo video
Demo · SpiderMediaArticles about Retention2 articles
Showing all 2
SKILL.mdthe file your agent reads
SKILL.md loads when you open this tab.
What's new3 releases · newest first · from the package's CHANGELOG
v0.1.32026-11-04Latest
### Added - **Marketplace demo film (`media.demo_video_url`).** Added the approved 88.6s narrated chalkboard launch film demonstrating business-model-first retention routing across agency, B2C, and B2B SaaS.
v0.1.22026-07-22
### Changed - **Tier flipped `internal` → `beta`.** The skill is now customer-visible on the marketplace. It had been published at `tier: internal` since 0.1.1 and was therefore hidden from brands browsing the catalog. - **Skill entry `visibility` flipped `internal` → `public`** to match the package tier. ### Added - **Marketplace card asset kit (SPEC-3).** `media.hero_image_url` + 3 `media.screenshots` — the business-model branch, the same-offer/opposite-outcome contrast, the involuntary-churn split, and the pause fork. Dark-board chalk house style (`#131313` board / white / gold `#EEBF01`). - **`marketing.long_description`** (first-person, 188 words) + **5 verb-led `marketing.features`**. - **`type: skill`** declared explicitly.
v0.1.02026-06-18
### Added — first release (guidance-only) - **Business-model-branched retention engine.** SKILL.md router branches on `agency | saas_b2b | b2c` BEFORE any save move — the three are different motions (different churn definitions, signals, levers, cadences). A discount saves a B2C buyer and gets an agency client fired. - **11 references** — 3 per-model save-offer matrices, model-aware at-risk scoring, dunning cadence (T+0/3/7/14/21/30), winback-craft-by-model, referral & loyalty, compliance (CA ARL + regulated claims), multilingual register, AgentBoard deals-board anatomy, SpiderFlow dunning catalog. - **11 recipes** — agency (onboarding-velocity, slow-fade-rescue, renewal-and-expand), saas (activation-onramp, failed-payment-recovery, at-risk-save-flow, cancellation-flow, nps-driven-loyalty), b2c (winback-discount-ladder, replenishment-reminder, referral-ask), cross (compliance-winback). - **70 registry exemplars** — evidence-bundled, business_model-tagged, every item carries a provenance URL + a measurable metric (save-rate / recovery-rate / retention-lift). Sourced from a deep scrapecreators dive (8 mined YouTube transcripts + Reddit comment threads) + external case studies. - **21 learnings** (KI shape) — incl. retention-is-three-motions, agency-slow-fade, 48h-onboarding-rule, saas-lever-is-activation, b2c-winback-is-discount-ladder, b2c-loyalty-is-identity. - **4 enforcement scripts** — verify-save-offer-decision-tree (asserts model branch + no agency discount + distinct offers + referral guard), verify-cadence-windows (dunning T+0 + decay spacing, winback 60-90d), lint-absolute-claims (regulated claims + CA-ARL single-offer), verify-coordination (4 cross-refs). - **Coordination (v2.2):** agentboard (deals board state) · agentdocs (playbooks/results) · agentmemory (what-worked per model+segment+brand) · opvs-protocol (IN from sales-pipeline, OUT to outbound-engine + community). ### Known scope (v0.2.0) - No native billing event source in the OPVS stack → `payment_failed` triggers arrive via a SpiderFlow webhook from each brand's own billing. Per-processor reference recipes (Stripe/Recurly/Chargebee) deferred. - EN floor only; DE/RU/FR/RO/HE/ES locale copy + jurisdiction-specific compliance vocab deferred.
Showing all 3
