A cross-Asia retail group converts four stalled pilots into a US$28M Center of Excellence in nine months

Context

Your retail group operates 1,800 stores across nine Asian markets under three banner brands, with 42,000 employees and approximately US$3.6B in annual revenue. The CEO had committed publicly to "AI as an operating capability" but four prior pilots in the prior 24 months had stalled in handoff: marketing analytics in Hong Kong, demand forecasting in Korea, store-staffing optimisation in Indonesia, and a returns-fraud model in Japan. Each pilot had produced a model that worked. None had produced organisational capability that lasted past the consultant departure. The Group CFO blocked further pilot funding until the group could show a coherent capability buildout, not another point-solution attempt.

Challenge

Three constraints. First, the group operated a federated structure: each country president held P&L and could veto group-level initiatives. Second, AI talent was scarce and expensive across the region; competing with tech firms on compensation alone was not viable. Third, the group had no internal definition of what "AI capability" meant, so success criteria had been negotiated separately for each prior pilot.

Approach

We ran a 6-phase capability-buildout model: define, structure, hire, train, sequence, hand-over. Define (4 weeks) ran a working session with the CEO, CFO, country presidents, and the four prior pilot leads to produce a one-page Group AI Capability Charter covering scope, governance, funding model, and the test for "capability achieved" (not "model deployed"). The charter was adopted at the next Group Executive Committee meeting.

Structure (4 weeks) designed a hub-and-spoke Center of Excellence: a 12-person group team owning platform, governance, hiring playbooks, and a country-engagement model; a 4-6 person spoke in each of the three largest markets owning local delivery. The country presidents co-signed the operating model.

Hire (12 weeks, run in parallel from week 6) executed a structured talent search across the region using AIMenta's hiring playbook: defined role rubrics, structured interview loops with calibrated scoring, and offer benchmarks indexed against tech-firm comparables minus 18% (offset by mission and stability). The group made 22 hires (14 hub, 8 spoke) at a 78% offer-acceptance rate against a 60% regional benchmark.

Train (run from week 8 as a structured 12-week internal cohort program) brought 240 mid-level managers through an applied AI curriculum: framing AI use cases, evaluating vendor pitches, reading model evaluations, and managing AI-augmented teams. The cohort produced 38 internally-generated AI use cases by graduation, of which 14 entered the new sequenced pipeline.

Sequence (run from week 14) replaced the prior pilot funding model with a stage-gated portfolio: discovery, prove, scale, and capability-handed-over. Funding flowed to the next stage only when the prior stage hit defined criteria. Hand-over (parallel from week 20) was structured into the operating model itself: every project carried a named country-team handover counterpart from kick-off.

Results

Within nine months the CoE had three projects in the scale stage (demand forecasting in Korea, store-staffing in Indonesia, returns-fraud in Japan) all of which had stalled under the prior model. Combined annualised gross benefit across the three: approximately US$28M, against a CoE annual run rate of US$6.4M, a 4.4x payback inside the first year. Six additional projects entered the prove stage from the cohort-generated pipeline. Average time from country-president request to deployed capability fell from "indefinite" to 19 weeks.

Voluntary attrition on the AI hires sat at 6% in year one against a regional tech-talent attrition benchmark of 22%, attributed in exit-interview data to mission clarity and the explicit stage-gated career path. The CFO unblocked further funding at the year-one review and authorised expansion to the remaining six markets in year two.

Lessons

Defining what "capability" meant before any further pilot funding flowed was the bridge that the prior 24 months had been missing entirely. The hub-and-spoke structure with country president co-signature converted a federated structure from a blocker into an asset. Hiring at -18% to tech comparables, offset by mission clarity and a defined career path, produced a 4x lower attrition rate than the regional benchmark.