Riley Blake Designs — 90-Day AI Transformation Plan (CTO / Chief AI Officer)

Horizon: First 90 days from full-time start (~late June / early July 2026)

v2 changes from critic review: (1) Outpost framing stripped — the Company Brain is built as a pure RBD-owned internal asset; Cole retains know-how/patterns but RBD owns the artifact, which removes the core conflict-of-interest exposure. (2) Buying copilot stays the first lighthouse but is now hard-gated on a backtesting methodology with an independent validator and a defined fallback. (3) Measurement contract moved to Phase 0 with leading indicators and an explicit 90-day-horizon caveat. (4) Team decision pulled to a Phase 0/1 parallel track. (5) Added change-management (who-loses), data-governance, and NetSuite-admin stakeholder sections.

Objective

Establish Cole as RBD's CTO/Chief AI Officer by standing up the operating model, foundation, and first proof points for org-wide AI — such that by Day 90 RBD has (1) a working Company Brain foundation owned by RBD, (2) 1–2 visible lighthouse wins demonstrating efficiency and revenue value, and (3) a governed, repeatable operating model for prioritizing and shipping AI across departments. Positioning is shown through shipped wins, not declared.

Executive Summary

The setup is unusually favorable. Cole arrives with a transformation mandate (authority to change workflows, require adoption), judged primarily on strategic positioning — which gives runway to build foundations properly — while leadership simultaneously wants efficiency, modernization, and revenue. Positioning must therefore be evidenced by concrete wins in those currencies.

This is not greenfield. As Alpine Analytica, Cole has already built RBD a live NetSuite→rb_warehouse warehouse, the Athos integration, Cotton/Batik PO automation, a reconciliation tool, static dashboards, a quilting calculator app, and the RBD Stash B2B portal. The job is to step inside, scale that into a governed internal capability, with Alpine retainers continuing in parallel.

Governance decision (resolved): The Company Brain and everything built under this plan is a pure RBD-owned internal asset — no Outpost product framing, no reusability-as-product requirement, no RBD-as-proof-case. Cole retains general know-how; RBD owns the artifacts. This removes the structural conflict the critic flagged. The residual governance item is narrower: the continuing Alpine retainers still bill RBD, so they need a clean written engagement letter, an alternate invoice approver, and Cole's recusal — handled in Phase 0.

The core tension to manage: "Strategic positioning" is the softest currency politically — easiest to second-guess with mixed literacy and a near-zero inherited team. The plan deliberately pairs foundation-building (Company Brain) with fast, visible lighthouse wins so momentum and proof accrue while the platform is built.

The wedge sequencing (all four departments bleed — this is prioritization, not discovery):

1. Forecasting / inventory (buying) — first lighthouse, hard-gated on backtesting. Highest value; built on existing PO tooling; but judgment-heavy on imperfect data, so it ships only after a validation gate clears.
2. Company Brain knowledge layer — the foundation. "Ask your data" on rb_warehouse, expanding to department knowledge. Drives modernization + literacy org-wide.
3. Marketing / content — fast parallel win + Phase-2 second lighthouse, and the fallback if the copilot gate slips. Reuses the existing focus-group + marketing-engine systems with a willing, sophisticated team (Mary's).
4. Order / sales ops — deferred to Q2. Highest operational/customer-facing risk; enter later via a low-risk RBD Stash retailer-assistant slice.
5. Design / catalog production — medium-term. Approach via the data side (SKU/catalog automation) before the creative act.

Decisions requested at the gate: wedge sequencing · hire-vs-contract (now a Phase 0/1 parallel track) · Alpine-retainer engagement governance · Q1 budget envelope.

What Already Exists (Inherited Stack)

Known data landmines (forecast credibility depends on these):

• tran_date = 1970-01-01 for ~60–70% of older rows; the IF(tran_date>'1970-01-01', tran_date, created_date) pattern is a workaround, not a fix. Phase 0 must validate that created_date actually approximates sale date (in NetSuite it is often record-creation time — if so, historical seasonality is systematically shifted).
• A $309K Q1 revenue miscount (PR #162) traced to this exact field.

Strategy

Three workstreams run in parallel, not in sequence — a CTO transformation fails if it's all platform (no visible wins) or all point-solutions (no durable capability).

A — Foundation: the Company Brain (RBD-owned). A governed knowledge + data layer on rb_warehouse, starting as a read-only NL analytics assistant ("ask your data") and expanding to department knowledge. Built and owned as an RBD internal asset.

B — Lighthouses: visible, leverage-able wins. Buying copilot first (gated), marketing content engine second. Each must be high-value, low political/operational risk, and built on existing assets.

C — Operating model & governance. AI use-case intake + prioritization, the agent-heavy delivery model, the (early) hire/contract decision, data classification + access controls, change management, and the Alpine-retainer engagement boundary.

Operating principles:

1. Foundation and wins ship together. Every 2-week block produces something a non-technical leader can see.
2. Build on what exists. Net-new infra only where existing assets genuinely block the use case.
3. Champions over mandates — but map who loses. Recruit the sophisticated pockets as partners; explicitly address the people whose standing the change threatens.

The Phased Plan

Phase 0 — Listen, Land & Lock the Contracts (Weeks 1–2)

Goal: Orient as an officer; lock access, baselines, governance, and the measurement contract before building.

• ☐ Internal listening tour — Purchasing (Sabrina), Design, Marketing (Mary), Sales reps, NetSuite admin (Bret?), ownership. Output: ranked pain inventory + a who-loses map per department (see Change Management).
• ☐ Data & access audit — confirm officer-level access (NetSuite read + who controls write, rb_warehouse, GA4, Meta Ads, Klaviyo API, Workspace). Model a Phase-1 contingency if access isn't fully cleared by Week 2 (don't assume the title clears IT turf/provisioning instantly). Output: access matrix + gap list + contingency.
• ☐ Measurement contract (locked before Phase 1) — for each baseline metric, specify the exact query/system, the owner who pulls it, and the identical Day-90 query. Define leading indicators (copilot recommendation-acceptance rate, buyer time-per-PO, Company Brain weekly active users) because outcome metrics like stockout/cancellation rates won't fully resolve in 90 days — only ~1 partial collection cycle closes. State this horizon caveat explicitly so the Day-90 review isn't a story.
• ☐ Data validation spike — verify the created_date-as-sale-date assumption; quantify how many rows have both date fields unusable; document the bias. Gates the forecasting backtest.
• ☐ Governance v0 (narrowed) — Alpine-retainer engagement letter (which SOWs continue, rates), alternate invoice approver, Cole recusal. Data classification (financial + customer PII) + a draft access policy for the Company Brain.
• ☐ Team track opened (parallel, not contingent) — start the hire/contract conversation now; don't make it depend on Phase 1 results (see Risk #2).
• ☐ Operating cadence — weekly leader-visible update; AI use-case intake doc opened.

Gate before Phase 1: wedge sequencing approved · access confirmed (or contingency active) · measurement contract locked · Alpine engagement governance signed.

Phase 1 — Foundation + First Lighthouse (Weeks 3–6)

Goal: Company Brain v0 live; AI buying copilot v1 — only if the backtest gate clears.

• ☐ Company Brain v0 — "ask your data" — read-only NL query over rb_warehouse (sales/inventory/customers by collection/designer/colorway/lifecycle), hardened against the date bug. First users: analytics + buyer. Owner: Cole + agents.
• ☐ Buying copilot v1 — HARD GATE. Before any output reaches Sabrina, a backtesting methodology must clear: backtest forecasts against historical seasons with known outcomes, on clean-data cohorts, to a predefined numeric tolerance (e.g., ±X% at closeout stage), validated by someone who is not the champion (Sabrina can't be the validator of the tool she's asked to adopt).
• If the gate clears: surface reorder recommendations inside the existing Cotton/Batik PO workflow, human-in-the-loop.
• If the gate does NOT clear by Week 3: copilot stays in backtesting/advisory-internal, and the visible Week-3–6 proof falls back to the marketing quick win below. No unvalidated forecast is ever shown to leadership.
• ☐ Marketing quick win (parallel + designated fallback) — point the focus-group + marketing-engine systems at one live RBD campaign/newsletter with Mary's team; renew the expired IG token. Low-risk, fast, champion-building.
• ☐ Wire the demand-forecast dashboard to live rb_warehouse — replace static CSV; demo surface.
• ☐ Sabrina change-management — frame the copilot as augmenting her judgment with an audit trail she controls, not replacing it (see Change Management).

Risk to watch: Data-quality credibility. The gate exists precisely so one bad forecast never reaches leadership.

Phase 2 — Expand & Embed (Weeks 7–10)

Goal: Extend the foundation across departments; harden the copilot; ship the marketing content engine; resolve the team model.

• ☐ Company Brain v1 — department knowledge — ingest PO methodology, sales-timing logic, collection lifecycle, process docs. Role-scoped access designed before ingest (a sales rep must not be able to pull a competitor retailer's full buying history — see Data Governance). Expand to Purchasing + Marketing.
• ☐ Buying copilot v2 — incorporate Phase 1 feedback; add backorder/closeout intelligence; move toward semi-automated PO drafting (human approves). Note: full automation needs a NetSuite/Sheets write path — acknowledged dependency, deferred to Q2; the advisory ceiling holds until then.
• ☐ Marketing content engine (second lighthouse) — repeatable RBD content pipeline (pillars → calendar → copy → focus-group QA) for Mary's team.
• ☐ Adoption & literacy — enablement for mixed-literacy departments; per-department champion designation; published "how to use the Company Brain" patterns.
• ☐ Team decision executed — hire/contract resolved and recruiting underway (conversation began Phase 0).
• ☐ RBD Stash retailer-assistant spike — scoped exploration only; safest entry into the deferred order-ops area.

Trigger logic: Don't expand the Company Brain to a new department until the prior one has ≥1 active weekly user. If the anchor department stalls at zero users, stop and run the objection matrix (don't push breadth to escape a depth problem).

Phase 3 — Prove & Institutionalize (Weeks 11–13)

Goal: Measure honestly, present, and lock the next-quarter operating model.

• ☐ Results review vs Phase-0 baselines — report leading indicators (recommendation-acceptance, time-per-PO, active users, marketing cadence) and clearly label outcome metrics as early/partial-cycle. Honesty here protects credibility with a skeptical CFO/board.
• ☐ Leadership presentation — show, don't tell — live Company Brain demo, forecast dashboard, before/after leading indicators, modernization narrative tied to efficiency + modernization + revenue.
• ☐ Operating model locked — AI use-case intake/prioritization ratified; agent-heavy delivery + new team roles defined; data governance finalized.
• ☐ Next-90 roadmap — Q2 preview: deeper order-ops automation, design/catalog data automation, retailer-facing AI, NetSuite write-integration where warranted, full outcome measurement once the collection cycle closes.

Day-90 definition of done: Company Brain in active use by ≥2 departments · buying copilot embedded in the live PO workflow (or transparently still in validation, with the marketing win carrying visible proof) · marketing engine on cadence · governed operating model documented · honest leading-indicator measurement presented.

Change Management — Who Loses (new)

"Champions over mandates" is half a strategy; the other half is the people whose standing the change threatens. Map and address each before the relevant phase:

• Sabrina (buyer) — her seniority rests on forecast judgment; "AI copilot" can read as "replacement audit trail." Frame: the copilot is her tool that documents and accelerates her judgment; she stays the decision-maker. Watch for her becoming the most sophisticated internal critic — that's a signal to engage, not override.
• Sales reps — carry collection/retailer knowledge in their heads; a Company Brain that democratizes it erodes their information advantage. Frame: it makes them faster in front of accounts, not redundant.
• Bret / NetSuite admin — may have owned the data layer; an AI layer over NetSuite can feel like lost control. Treat as the single most operationally important Phase-0 stakeholder; co-design access rather than route around him. Have a plan for the uncooperative/slow-walk scenario (escalation path to ownership via the mandate, used sparingly).
• Long-tenured staff — "modernization" is a threat word to people who are good at the old way. Make them visible winners (early access, their workflows improved first), not holdouts.

Data Governance (new)

The Company Brain sits over financial transactions + customer PII (Klaviyo) + business-process docs. Before end-user exposure:

• Data classification of rb_warehouse contents (financial / PII / internal).
• Role-scoped access designed before ingest, not retrofitted — define who can query what. Explicit control on B2B competitive exposure (one retailer's rep must not see another retailer's full purchase history).
• Access + query-audit policy documented and owned. Computational guardrails before user rollout.

Risks & Watchlist

1. Forecasting credibility (highest residual risk). Judgment-heavy forecasting on imperfect data; a wrong number in front of leadership sets the narrative back. Mitigation: the hard backtest gate + independent validator + marketing fallback.
2. Solo capacity / single-threaded critical path. Two hard workstreams in Weeks 3–6 with no human backstop, while Cole is also onboarding and navigating the org. Mitigation: team track opened in Phase 0 (not contingent on Phase 1); Alpine headcount budgeted for Phase 1.
3. Adoption stall / champion defection. Mixed literacy + threatened standing can mean quiet non-adoption even under a mandate. Mitigation: who-loses map, objection matrices, depth-before-breadth gate with a stall-stop rule.
4. Access friction. Officer title may not clear NetSuite/IT provisioning or turf by Week 2. Mitigation: Phase-0 audit + explicit Phase-1 contingency; read-first until write authority is explicit.
5. Measurement validity / horizon. 90 days ≈ one partial collection cycle; outcome metrics can't fully resolve. Mitigation: measurement contract + leading indicators + honest labeling at Day 90.
6. Data-governance gap. AI over financial + B2B customer data without classification/role-scoping. Mitigation: governance section above, before end-user exposure.
7. Alpine-retainer conflict (narrowed, not eliminated). Continuing retainers still bill RBD. Mitigation: signed engagement letter + alternate approver + recusal in Phase 0.
8. Scope sprawl across four departments. Mitigation: enforced wedge sequencing; explicit deferral of order-ops + design.

Measurement

Weekly leadership snapshot: visible artifact shipped this week · Company Brain active users (by dept) · copilot recommendation-acceptance + forecast-vs-actual variance (defined denominator) · marketing engine on cadence (Y/N).

Phase-gate metrics: P0 — baselines + measurement contract locked, access matrix complete · P1 — Company Brain v0 live, copilot through the backtest gate (or fallback active) · P2 — ≥2 departments active, team decision executed · P3 — leading-indicator impact vs baseline, honestly labeled.

Decisions Requested (at the gate)

1. Wedge sequencing — forecasting/inventory first (gated), Company Brain as parallel foundation, marketing as fast win + fallback + Phase-2 second lighthouse, order-ops + design deferred to Q2. Approver: Cole.
2. Hire vs contract (now a Phase 0/1 parallel track, not contingent) — recommend budgeting Alpine headcount for Phase 1 and committing to a hire by Phase 2, decided up front rather than gated on Phase 1 proof. Approver: Cole + ownership (budget).
3. Alpine-retainer engagement governance — approve the engagement letter + alternate-approver + recusal mechanism for the continuing retainers. Approver: Cole + ownership.
4. Q1 budget envelope — ceiling for tools/infra/API + the hire/contract spend. Approver: ownership.

(Former Decision: "Outpost build approach" removed — moot now that the build is a pure RBD-owned asset.)

Open Questions (for Cole / Leadership)

1. Start date + internal announcement — how/when is the CTO/CAIO role communicated to staff?
2. NetSuite/IT ownership — who controls NetSuite write access and the website platform? Is Bret the admin, and where does he report?
3. Data access confirmations — GA4, Meta Ads, Klaviyo API confirmed for you as officer?
4. Board/owner reporting cadence — who do you report results to, and how often?
5. Alpine retainer scope post-transition — which SOWs continue vs fold into the internal role?
6. First champion departments — confirm Purchasing (Sabrina) + Marketing (Mary).

Bottom Line

The conditions are rare: a real mandate, runway from a positioning-based scorecard, a substantial inherited stack, RBD as primary focus. The v1 failure modes the critic surfaced — an ungoverned Alpine/Outpost conflict, a high-variance forecasting bet on dirty data, and a measurement story that wouldn't survive scrutiny — are now addressed: the build is a pure RBD-owned asset, the copilot is hard-gated with a fallback, and measurement is contracted up front with honest leading indicators.

By Day 90: a Company Brain in active use, a buying copilot proving value (or transparently still validating, with marketing carrying the visible win), a marketing engine on cadence, a governed operating model, and a credible, defensible results story.

Recommend: run /grill-me on the decision tree, resolve the six open questions, then approve the four decisions to begin Phase 0.