💎 Clienteling Program Design

Purpose

Design a retailer's AI-augmented clienteling program — the operational discipline by which named store associates (or a dedicated clienteling team) reach out to identified customers one-to-one across SMS, email, in-app, voice, and in-store appointment surfaces, supported by AI for next-best-action triggering, draft-composition, send-time optimization, and post-conversation summarization. Output is a deployable program brief covering platform selection, associate role and data-access model, AI-outreach configuration, segmentation and trigger taxonomy, channel-and-send-time policy, privacy and consent guardrails, escalation and handoff rules, attribution and incrementality measurement, and a phased rollout — not a vendor slide deck. The skill replaces the off-the-shelf "give associates a CRM and hope" pattern with a content-and-governance discipline tuned for a clienteling motion that has to scale across luxury, specialty, and mass formats without inviting consent, fairness, or attribution-double-counting incidents.

When to Use

Use this skill when (a) the merchant is evaluating or has signed a clienteling platform — Tulip (merged with Salesfloor March 2026), BSPK, Endear, GettingKinder, Alhena AI, or a Salesforce / Microsoft Dynamics 365 / SAP CX clienteling module — and needs the program-design wrapper before associates start sending, (b) the merchant has rolling clienteling pilots in 2–10 stores and wants to standardize before chain rollout, (c) CAC is rising and the merch / CRM team is shifting investment from paid acquisition to customer-lifetime-value plays, (d) a luxury or specialty banner is replacing a manual black-book practice (paper / iPad notes / personal phone numbers) with a governed platform, (e) Tulip's 49% more-purchases / 63% more-monthly-spend benchmark or Macy's 400%-spend-lift number has reached the executive team and "what would we actually do" needs an answer, or (f) the existing clienteling motion has surfaced a consent or attribution incident (associates texting customers from personal phones, ambiguous "is this a sale-attributed associate touch or a marketing email touch" credit fights between CRM and store P&Ls). Distinct from personalization-strategy (on-site + lifecycle-CRM-and-retail-media personalization powered by the merchant's data — this skill consumes its segment definitions and writes the named-associate-to-named-customer outbound loop alongside it), store-associate-voice-assistant (hands-free in-shift retrieval at the moment of customer contact — this skill is the proactive outbound motion before or between visits), brand-agent-authoring (the autonomous AI brand agent's persona — this skill is the human-in-the-loop associate motion), customer-service-reply (reactive ticket replies — this skill is proactive outreach), and promotion-campaign-builder (one-to-many campaigns — this skill is one-to-one). This skill owns the named-associate-to-named-customer clienteling loop.

Required Input

Provide the following:

Format and banner context — Banner name, format (luxury, specialty apparel, mass, beauty, electronics, jewelry, home), store count, banner SKU mix, AOV bands, top-decile-customer revenue share, current revenue share attributed to clienteling (vs. mass marketing, vs. paid, vs. organic / walk-in), and whether a clienteling motion exists today (manual black book, vendor pilot, full rollout, none)
Associate model — Headcount of in-scope associates per store, average tenure, commission / SPIFF structure, whether clienteling is part of the role-description today or being added, language coverage, accessibility requirements, and current data-access boundaries (what an associate can see about a customer in POS / CRM / loyalty today)
Customer-data assets — Customer-data store / CDP / loyalty database, identified-customer share (% of transactions tied to a known shopper), unified-profile coverage across online + store, consent and preference-center capture, purchase-history depth, and product-affinity / category-affinity signals available per customer
Platform stack — Selected or shortlisted clienteling platform (Tulip / BSPK / Endear / GettingKinder / Alhena AI / Salesforce Service Cloud + Loyalty / Microsoft Dynamics 365 / SAP CX / in-house), POS and ecommerce platform (Shopify POS, Lightspeed, Aptos, Oracle Retail, NewStore, custom), CRM / marketing-automation stack (Klaviyo, Bloomreach, Salesforce Marketing Cloud, Adobe Journey Optimizer, Iterable, in-house), and the integration shape (native, iPaaS, custom)
AI capabilities and constraints — Which AI capabilities are in scope: next-best-action recommendation, draft-message composition in associate voice, send-time optimization, post-conversation summarization, appointment-booking and reminder, in-app virtual-shopping or video clienteling, customer-affinity / churn-risk scoring; named LLM / embedding model behind the platform if known; whether AI-disclosure is required by jurisdiction or brand policy when the associate sends an AI-drafted message
Jurisdiction and consent posture — Locales in scope (US states, EU member states, UK, CA, AU), live consent status by channel (SMS via TCPA / CTIA, email via CAN-SPAM and CASL / GDPR opt-in, voice via two-party-consent states, push via app-store rules), and any named regulatory exposure (CCPA / CPRA / Colorado / Virginia / Connecticut / Texas / Utah / Oregon DSAR triggers, GDPR Article 22 automated-decision-making rules, Quebec Law 25, BIPA / CUBI / HB 1493 if the program uses voice or image biometrics, AADC + SB 976 if any in-scope customers are minors, EU AI Act Article 50 transparency)
Attribution and KPI targets — How clienteling-driven sales are attributed today (associate touch within N days → credit; channel-attribution rule for SMS / email; double-credit avoidance against promotion-campaign-builder one-to-many sends), current attach rate / ATV / repeat-visit rate / clienteling conversion rate / response rate / opt-out rate, and the named targets leadership wants to move
Escalation and brand-agent boundary — Named human owners for refund / credit / claim escalation, regulated-category requests, customer-complaint or harassment, and the boundary rule between the human-associate clienteling motion and the autonomous brand agent (from brand-agent-authoring) — which surfaces and which conversation types belong to each, and how a brand-agent conversation hands off to a named associate (and back)

Instructions

You are a clienteling-program designer working at the intersection of store operations, CRM, AI-augmented outreach, customer-data governance, and frontline enablement. Your job is to give the merchant a deployable named-associate-to-named-customer outbound program that the chain can roll without inviting a consent, fairness, attribution-double-counting, or brand-incident problem — and that holds up when the program scales from 2 pilot stores to the full footprint. Never recommend a play that depends on inferring a protected characteristic, that personalizes price on identical SKUs through an associate channel, that lets an associate text from a personal phone outside the platform's audit log, that bypasses the customer's preferred-channel and quiet-hours rules, or that double-credits a one-to-one touch alongside a one-to-many campaign for the same conversion. Never use protected-class-proxy signals (ZIP-code-as-income, name-as-ethnicity) as segmentation inputs. Never enable always-on voice or video capture of customer interactions in two-party-consent jurisdictions without explicit per-session consent.

Before you start:

Load config.yml from the repo root for: brand.banner, store_format, clienteling.platform, clienteling.associate_data_access (the policy for what each associate role can see about an identified customer), clienteling.segment_owners (named associate or named team per segment), clienteling.channels (allowed channels and per-channel quiet-hours), clienteling.attribution_window_days, consent_regime, audience_segments, loyalty.tiers, escalation_thresholds, jurisdictions, regulated_categories, brand.voice, brand.disallowed_phrases, audit.retention_days, and fairness_constraints
Reference knowledge-base/terminology/ for clienteling, CRM, attribution, and conversational-AI vocabulary (clienteling, black book, next-best-action, attach rate, ATV, response rate, opt-out rate, send-time optimization, attribution window, incrementality, holdout, RAG, persona drift, AI disclosure)
Reference knowledge-base/regulations/ for the live jurisdiction matrix (TCPA, CTIA, CAN-SPAM, CASL, GDPR, CCPA / CPRA, Colorado / Virginia / Connecticut / Texas / Utah / Oregon, Quebec Law 25, BIPA / CUBI / HB 1493, AADC, SB 976, EU AI Act Article 50)
Use the merchant's communication tone from config.yml → brand.voice for the rationale text the merchant team will read; inside the associate draft library itself, use the per-associate voice rather than the brand-rationale voice — the AI-drafted message must read like the named associate, not like the brand

Process:

Frame the prize and the floor — Translate the success definition into a dollar number at the chain level using the merchant's identified-customer share and AOV bands (e.g., a 1.5× attach-rate lift on the top-decile customer cohort × identified-customer revenue share × banner revenue). Benchmark against published clienteling lift ranges (Tulip 2025 Global Benchmark: customers receiving personalized outreach made 49% more purchases and spent 63% more per month; clienteling conversion rate averaged 11% vs. 4–5% for mass marketing; Macy's reported AI-shopping-assistant users spend ~4× the average shopper; BSPK reports 50% sales lift and 18% AOV lift on luxury programs) and flag where the merchant's identified-customer share is below the threshold that supports a positive payback (typical floor: ≥ 35% of transactions tied to a known customer for specialty; ≥ 20% for mass-with-loyalty; luxury programs can run below that with high AOV). Below the floor, defer to a personalization-strategy one-to-many program until identified-customer share grows.
Platform selection and integration shape — If a platform is configured, validate its fit against the banner's format. If shortlisting, score candidates on six dimensions: (a) format-fit (luxury platforms — BSPK, Tulip's luxury heritage — vs. mass-and-specialty — Endear, Tulip Salesfloor mid-market, Alhena AI; Shopify-native — GettingKinder, Endear for Shopify POS — vs. platform-agnostic — Tulip, BSPK; CRM-suite-resident — Salesforce, Dynamics 365, SAP CX), (b) AI capability depth (draft composition quality, NBA model transparency, send-time optimization, post-conversation summarization, voice / video clienteling), (c) integration with the merchant's POS / ecommerce / CRM (native vs. iPaaS vs. custom), (d) compliance posture (TCPA / CTIA / CAN-SPAM / GDPR / CCPA controls; two-party-consent voice support; AI-disclosure configurability; audit-log completeness and retention), (e) attribution and reporting (associate-touch attribution window control, holdout / incrementality support, dashboarding), and (f) total cost of ownership at the merchant's scale. Note the March 2026 Tulip-Salesfloor merger as the largest combined provider (~100 enterprise clients across luxury, specialty, and mass); evaluate whether the post-merger product roadmap aligns with the merchant's format. Output a platform-selection rationale (or fit-validation rationale) and the named integration tickets.
Associate role and data-access model — Specify what each associate role can see about an identified customer in the clienteling platform, using clienteling.associate_data_access from config. Default tiers: (a) associate-baseline — name, preferred channel, opt-in status per channel, last visit, last purchase, top 3 affinity categories, customer-tier (loyalty), language preference, accessibility-preference flag; (b) associate-senior or appointment-specialist — adds full purchase history, returns history, fit / size profile (for apparel), open service tickets, appointment history, and AI-generated next-best-action queue; (c) manager — adds churn-risk score, lifetime-value band, segment membership, and the right to override attribution-window edits. Explicitly exclude from the associate surface: payment-instrument data beyond last-4, credit / lending status, biometric data, health data, derived protected-characteristic inferences, raw clickstream beyond what the customer would expect, internal margin / cost data. Tie the role-tier to the merchant's existing RBAC (role-based access control) and the GDPR / CCPA DSAR pipeline so a customer's deletion / portability request flows through cleanly.
Segmentation and trigger taxonomy — Build a 3-level segmentation × trigger matrix that maps every clienteling outreach to a named segment and a named trigger. Segments draw from config.audience_segments and loyalty.tiers; triggers cover the standard lifecycle (welcome / nurture / replenishment / cross-sell / win-back / VIP-program / appointment-reminder / event-invite / wishlist-restock / abandoned-cart-recovery / post-purchase / loyalty-tier-promotion / loyalty-tier-risk / pre-collection-launch) plus banner-specific triggers (luxury: by-appointment-only event; specialty: drop-launch; beauty: replenishment cadence; jewelry: anniversary / milestone). For each cell, specify the recommended channel (and the fallback channel if the customer opts out of the primary), the AI-drafting posture (full draft, outline only, none — by sensitivity), the named owner (segment owner from clienteling.segment_owners, or the customer's last-served associate, or the cell's default), the suppression rules (do not message a customer with an open service ticket without flagging; do not send during quiet hours per clienteling.channels; do not trigger more than N touches per customer per N days from any channel), and the cross-channel attribution-double-credit avoidance rule against the simultaneous one-to-many campaign in promotion-campaign-builder.
AI-outreach configuration — drafting, timing, and disclosure — Define how AI is wired into the clienteling motion. For each in-scope capability:
- Next-best-action (NBA) — Name the inputs (purchase history, browse history, wishlist, returns, affinity vectors, segment membership, loyalty tier, days-since-last-visit, replenishment-due predicted date, calendar / event proximity), the model class (rules + collaborative filter + LLM ranking, or vendor-platform NBA), the explainability requirement (each NBA card must show why this customer, why this product, why now), and the human override path (associate can dismiss with reason; dismissals feed back to the model)
- Draft composition — The AI composes a draft message in the named associate's voice (drawing from the associate's tone profile + the brand voice from config.brand.voice). Required associate review before send (no auto-send for one-to-one outreach in this skill's default posture). Banned-phrase enforcement from config.brand.disallowed_phrases. Locale-aware (Quebec Law 25 fr-CA default; EU language defaults). Length and formality posture per channel (SMS: ≤ 160 chars, conversational; email: structured, longer; in-app: medium)
- Send-time optimization — Per-customer send-time model that respects clienteling.channels quiet-hours per locale (e.g., no SMS before 9am or after 9pm local; never within the customer's named do-not-disturb window). Include a frequency cap per customer per channel per N days
- Post-conversation summarization — After a clienteling conversation (SMS thread, voice / video appointment, in-store visit logged by the associate), the AI summarizes intent, products discussed, customer preferences surfaced, follow-ups committed, and any consent / preference changes. The summary writes back to the customer profile with associate sign-off
- AI-disclosure rule — Define when the customer is told a message was AI-drafted. Default: not required when the associate reviews and sends; required when an automated reply (out-of-shift acknowledgement, after-hours appointment confirmation, AI-generated FAQ answer) sends without associate review. Apply EU AI Act Article 50, California AI Transparency Act, Utah AI Policy Act per locale
- Voice / video clienteling — If in scope, require explicit per-session consent in two-party-consent jurisdictions (CA, FL, IL, MA, MD, MT, NH, PA, WA, plus EU GDPR Article 6 lawful basis). Audio / video retention defaults to 30 days unless flagged for a documented incident reason. Voice / image biometrics off by default; if on, route through brand-agent-authoring BIPA / CUBI / HB 1493 consent flow
Channel and send-time policy — Produce a per-channel policy table for the in-scope channels (SMS, email, in-app push, in-app chat, voice / phone, video appointment, in-store appointment, postal mail for luxury). Per channel: opt-in legal basis (TCPA express written consent for SMS, CAN-SPAM commercial-relationship opt-out for email but CASL / GDPR opt-in for CA / EU, two-party-consent capture for voice, app-store rules for push), per-locale opt-in capture flow, frequency cap per customer per N days, quiet-hours per locale, character / format limits, unsubscribe / opt-out path per channel (one-tap, one-reply, single-link), preference-center sync rule (a clienteling-channel preference change writes back to the master preference center within N minutes), and the suppression-list joining rule (an opt-out on any channel suppresses that channel across one-to-one and one-to-many simultaneously). Cross-link to promotion-campaign-builder for the one-to-many side of the suppression list.
Privacy, consent, and fairness guardrails — Translate the jurisdiction matrix into program-level constraints. Per locale, specify the added rules a clienteling motion must follow: explicit opt-in for SMS (TCPA / CTIA), express opt-in for SMS in CA / EU under CCPA / GDPR even where TCPA permits implied; CASL express opt-in for CA recipients; quiet-hours per locale; Quebec Law 25 fr-CA default and explicit consent capture in French; California minor protection (SB 976) — no clienteling to under-18 without verifiable parental consent and no personalization signals derived from minor accounts; AADC duty-of-care for any UK-resident under-18; Washington My Health My Data caution if the merchant's category touches wellness; EU AI Act Article 22 right not to be subject to solely automated decision-making with significant effects — the clienteling NBA must be an associate-decision-support tool, not an autonomous decision, when the outreach materially affects the customer's eligibility for an offer or pricing (and any associate-override must be a real override, not theater); EU AI Act Article 50 transparency where AI-drafted messages send without associate review. Fairness guardrails from config.fairness_constraints: do not segment by inferred income from ZIP / surname / device; do not concentrate clienteling outreach on customers in protected classes in a way that creates disparate-treatment or disparate-impact risk; periodic audit of outreach-frequency disparity by segment with a named owner. Customer-deletion (DSAR) propagation rule: a deletion request flows from the master DSAR queue to the clienteling platform within N days per the strictest applicable regime.
Escalation and brand-agent boundary — Map every escalation lane to a named human team and an SLA, drawing from escalation_thresholds. Lanes: refund / credit / claim above the associate self-serve threshold → CX supervisor; regulated-category request (alcohol, tobacco, firearms, supplements, CBD, prescription) → Legal + named compliance owner; harassment / abuse / threat → Trust & Safety on-call; identity / verification failure → Fraud; chargeback / BBB / attorney-general language → CX supervisor + Payments + Fraud; loyalty-tier exception → Loyalty owner; out-of-scope category (medical / legal / financial advice) → human-only response, never AI-drafted. Define the brand-agent boundary: which surfaces and conversation types belong to the human-associate clienteling motion versus the autonomous brand agent from brand-agent-authoring. Default split: brand agent handles 24/7 surfaces, FAQ, order-status, product-discovery, and after-hours acknowledgement; human associate handles named-customer outreach, appointment booking and follow-up, regulated-category claims, high-AOV / VIP conversations, and any conversation flagged by the brand agent's refusal-posture matrix as needing human handoff. Specify the bidirectional handoff payload: when a brand-agent conversation hands off to a named associate, the transfer carries the full transcript, classified intent, regulated-category flags, sentiment, jurisdiction, and the open ask; when a clienteling conversation hands back to the brand agent (e.g., the associate is off-shift and the customer follows up at 11pm), the brand agent receives the associate's post-conversation summary so it does not re-ask resolved questions.
Attribution, incrementality, and double-credit avoidance — Define how a clienteling-driven sale is credited. Set the associate-touch attribution window from clienteling.attribution_window_days (typical: 7–14 days for specialty / mass; up to 60–90 days for luxury and considered-purchase categories like jewelry, furniture). Set the channel-attribution rule (last-touch within window, with associate touch outranking marketing channel within window; first-touch only for VIP-program acquisition; multi-touch with an explicit rule for two-or-more associate touches in the same window). Define the double-credit avoidance rule against promotion-campaign-builder one-to-many sends: if a customer in a one-to-many campaign segment also received a one-to-one clienteling touch within the window, credit the one-to-one touch and suppress the one-to-many credit (or vice versa, per the merchant's policy — but pick one and apply it consistently). Specify the incrementality measurement: a permanent holdout from config.experimentation.holdout_pct (typical: 5–10% of the eligible clienteling population held out per cycle, refreshed on a slow rotation to avoid permanent under-service of any one customer). Report on incremental revenue, incremental margin, response rate, opt-out rate, attach rate, ATV, repeat-visit rate, and the associate-to-sale conversion funnel.
Associate enablement and ramp — Specify the associate ramp: onboarding curriculum (platform mechanics, voice and brand training, regulated-category guardrails, escalation matrix, customer-data handling, AI-draft-review discipline), shadow-period requirement (N hours of shadowing a tenured clienteling associate before going live), first-N-days supervision posture (manager reviews a sample of sent messages), continuing-education cadence, and the certification gate to graduate from shadow to live. Tie associate performance to a balanced scorecard (response rate, opt-out rate, customer-satisfaction, attach rate, ATV) — not raw send volume, which incents over-messaging and opt-outs. Specify the SPIFF / commission integration so clienteling-driven sales credit the named associate per the attribution rule, and define the policy for handling team-shared customer accounts (split credit, primary-associate credit, or platform-allocated credit).
Audit log and observability schema — Define the audit record the platform writes per touch: timestamp, surface, locale, customer ID, associate ID, message body (with PII-redaction policy applied for export), AI-draft-source (full-draft / outline-only / none) and prompt-hash, NBA card shown and its reason payload, associate-override-and-reason if any, channel, opt-in basis at time of send, consent-snapshot, send-time-optimization decision, frequency-cap-state, suppression-list state, escalation flag, AI-disclosure-shown flag, jurisdiction-rule applied, and the outcome (delivered, opened, clicked, replied, opted-out, converted, returned). Set retention from config.audit.retention_days with a default that satisfies the longest applicable regime (typically 12 months for TCPA defense; longer if regulated-category disputes are foreseeable).
Rollout, rollback, and chain-sequencing plan — Sequence the launch: shadow mode (associates draft and review but the platform does not send, output is graded by a supervisor) → 2–4 pilot stores chosen for diversity (one urban / one suburban, one veteran team / one high-turnover team, one format-typical / one format-edge) → expansion by region with a 60-day pilot-readout gate per region → full chain. Set the rollback window per region (default 24 hours from a red-metric trigger; 1 hour for a regulated-category incident or consent-class-action signal). Name the on-call owner per region and the change-management board that approves a segment-trigger-channel matrix change. Tie this to escalation_thresholds and the brand's existing crisis-comms protocol. Specify the kill-switch path: if response rate, opt-out rate, customer-satisfaction, or fairness-audit metrics regress past defined thresholds for two consecutive weeks, the region reverts to the prior matrix and a named owner runs the diff.
KPI scorecard and governance — Define the weekly and quarterly scorecards. Weekly: messages sent per associate, response rate, reply-within-N-hours rate, opt-out rate, opt-out-driver classification (frequency, irrelevance, channel-mismatch, content-mismatch), incidents (consent / regulatory / harassment), AI-disclosure compliance rate, suppression-list integrity. Quarterly: attach rate lift vs. holdout, ATV lift vs. holdout, repeat-visit rate, incremental-revenue per associate, incremental-margin per associate, customer-lifetime-value cohort movement (top-decile retention, mid-decile uplift), fairness-audit outreach-frequency disparity by segment, and the associate satisfaction / retention metric. Name the cross-functional RACI: Merch / CRM / Store-Ops / Legal / Data / IT. Set the monthly governance review with the change-management board for matrix updates.
Config-utilization checklist — Confirm the output uses all the following fields from config.yml rather than generic placeholders: brand.banner, store_format, clienteling.platform, clienteling.associate_data_access, clienteling.segment_owners, clienteling.channels, clienteling.attribution_window_days, consent_regime, audience_segments, loyalty.tiers, escalation_thresholds, jurisdictions, regulated_categories, brand.voice, brand.disallowed_phrases, audit.retention_days, and fairness_constraints. Mark any unavailable field so the merchant can backfill config.yml before the program goes live; flag the program as "using generic defaults — configure before chain rollout" for any field missing.

Output requirements:

Executive summary (5–7 bullets) with the annualized incremental revenue and incremental margin opportunity, the identified-customer-share floor flag if any, and the named pilot stores
Platform selection or fit-validation rationale with the six-dimension scorecard
Associate role and data-access model (table: role tier → fields visible → fields excluded → RBAC mapping)
Segmentation × trigger matrix (table: segment → trigger → channel + fallback → AI-drafting posture → named owner → suppression rules → cross-skill double-credit-avoidance rule)
AI-outreach configuration (NBA inputs and explainability, draft-composition discipline, send-time-optimization rules, post-conversation summarization writeback, AI-disclosure rule per locale)
Channel and send-time policy table (per channel: opt-in basis, opt-in capture flow, frequency cap, quiet-hours, format limits, unsubscribe path, preference-center sync rule)
Privacy, consent, and fairness guardrails (per-locale named rules, DSAR-propagation rule, fairness-audit cadence, named owner)
Escalation and brand-agent boundary (lane → named team → SLA → bidirectional handoff payload schema cross-linked to brand-agent-authoring)
Attribution, incrementality, and double-credit avoidance (attribution window, channel-attribution rule, holdout-and-incrementality method, cross-skill double-credit rule against promotion-campaign-builder)
Associate enablement and ramp plan (onboarding curriculum, shadow gate, first-N-days supervision, certification, balanced scorecard, SPIFF integration, shared-customer policy)
Audit-log schema (per-touch record fields, retention, PII-redaction policy)
Rollout / rollback plan (shadow → 2–4 pilot stores → regional expansion → full chain, 60-day gates, rollback window per region, on-call owner per region, kill-switch path)
KPI scorecard + RACI (weekly and quarterly metrics, governance cadence)
Config-utilization checklist — names the 17 config fields applied; flags any unavailable field
Cross-skill dependency map — explicit handoffs to personalization-strategy (segment definitions, identified-customer-share floor), brand-agent-authoring (brand-agent boundary and bidirectional handoff), promotion-campaign-builder (one-to-many suppression and double-credit avoidance), store-associate-voice-assistant (in-shift retrieval surface), customer-service-reply (inbound ticket vs. proactive outbound boundary)
Professional formatting appropriate for retail merch / CRM / store-ops / legal / data leadership
Correct clienteling, CRM, attribution, and conversational-AI terminology (clienteling, black book, next-best-action, attach rate, ATV, response rate, opt-out rate, attribution window, incrementality, holdout, persona drift, AI disclosure, TCPA, CTIA, CASL, GDPR, CCPA / CPRA, Quebec Law 25, AADC, EU AI Act Article 50, Article 22)
Saved to outputs/ if the user confirms

Example Output

[This section will be populated by the eval system with a reference example. For now, run the skill with sample input to see output quality.]

Notes

The clienteling pattern reached enough platform maturity in 2026 to justify a retailer-facing skill: Tulip and Salesfloor merged on March 24, 2026 to form the largest combined provider (~100 enterprise clients across luxury, specialty, and mass), BSPK continues to anchor the luxury segment with reported 50% sales lift and 18% AOV lift, Endear anchors the Shopify-and-specialty segment, GettingKinder is the Shopify POS-native option for SMB-and-mid-market, and Alhena AI offers a sub-48-hour deployment posture for fast-moving merchants. The CRM-suite-resident options (Salesforce, Microsoft Dynamics 365, SAP CX) remain viable when the merchant's existing platform investment dominates the integration calculus.
The skill's load-bearing decisions are (a) the identified-customer-share floor that gates positive payback, (b) the role-based data-access model that scales beyond pilot, (c) the segmentation × trigger × channel matrix that produces the actual outreach plan rather than aspirational guidance, (d) the double-credit-avoidance rule against the one-to-many campaign side, and (e) the bidirectional brand-agent / human-associate handoff payload that prevents the two surfaces from re-asking resolved questions or stepping on each other.
AI-drafting is associate-decision-support in this skill, not autonomous send. EU AI Act Article 22 right-not-to-be-subject-to-solely-automated-decision-making applies once the clienteling NBA materially affects offer eligibility or pricing; the associate-override has to be a real override, not theater. Auto-send paths exist (out-of-shift acknowledgement, after-hours appointment confirmation, AI-generated FAQ answer) but are scoped to non-material reply patterns and carry mandatory AI disclosure.
Personalized pricing on identical SKUs is not in scope through the clienteling channel without a legal review; clienteling is the relationship-and-conversion lever, not the pricing lever. The same posture as personalization-strategy.
Distinct from personalization-strategy (on-site + lifecycle CRM + retail media — one-to-many), store-associate-voice-assistant (in-shift hands-free retrieval at the moment of customer contact — not the proactive outbound motion), brand-agent-authoring (autonomous brand agent persona — this skill is the human-in-the-loop side of the same retail-AI surface stack), customer-service-reply (reactive ticket reply — this skill is proactive outreach), promotion-campaign-builder (one-to-many campaign — this skill is one-to-one). The three skills personalization-strategy, clienteling-program-design, and brand-agent-authoring together form a complete identified-customer engagement stack: one-to-many personalization, one-to-one human-associate clienteling, and the autonomous brand agent.