A multi-sector industrial group ran three businesses that shared almost nothing: a heavy manufacturing arm, a field services division, and a consumer products unit. Each argued its AI needs were unique, and each was right about the last mile and wrong about the foundation. The engagement built one shared readiness posture, a common data, governance, and platform core, then let each division layer its own use cases on top. The result was a group that could move an AI capability from one division to another in weeks instead of rebuilding it, without forcing three very different businesses into one template.
Three businesses, one holding company, no shared floor
The group held three operating divisions under a lean corporate center. Heavy Industrial ran capital-intensive plants where the AI prize was predictive maintenance and yield. Field Services ran a dispatched workforce of about 4,000 technicians across a national footprint, where the prize was scheduling density and first-time-fix rates. Consumer Products ran fast-moving brands through retail and direct channels, where the prize was demand forecasting and marketing spend allocation. The divisions had almost nothing operationally in common, and each had used that fact to justify going it alone on AI. A plant engineer and a brand marketer really do live in different worlds, and the divisions were not wrong that their use cases diverged. The corporate center's suspicion was narrower and sharper: that beneath those genuinely different use cases sat a set of foundations that were identical, and that the group was paying three times to build them while capturing none of the leverage a group structure is supposed to provide.
The suspicion was correct. Each division was separately negotiating cloud and model vendor contracts, separately writing AI governance policy, and separately hiring scarce machine learning talent into three competing pools. Combined AI spend was near 9 million dollars a year with almost no reuse between divisions, and the corporate center could not answer a simple board question about what the group as a whole was getting for it. The engagement was scoped to answer a specific question: what should be shared across three genuinely different businesses, and what should stay local, so that the group captured leverage without forcing a false uniformity that none of the divisions would accept. The failure mode to avoid was the standard corporate reflex of mandating one platform, one methodology, and one roadmap for everyone, which would have been rejected by every division that could point to a real operational difference. The opposite failure, leaving everything local, was the status quo that had produced triple spend. The work was to find the exact line between the two.
Shared foundation, local last mile
We drew a hard line between the layers that benefit from being common and the layers that must stay divisional. The readiness assessment was run once per division against the same six dimensions, which let the corporate center see, on one page, where the divisions genuinely diverged and where they had simply chosen to. The table below shows how each layer was assigned.
| Capability layer | Heavy Industrial | Field Services | Consumer Products | Ownership |
|---|---|---|---|---|
| Cloud and compute contracts | Common | Common | Common | Shared group platform |
| Data governance and model risk policy | Common | Common | Common | Shared group platform |
| ML talent pool and hiring | Common | Common | Common | Shared, deployed to divisions |
| Data pipelines and feature engineering | Local | Local | Local | Divisional teams |
| Use case models | Predictive maintenance | Scheduling and dispatch | Demand forecasting | Divisional owners |
| Deployment and adoption | Local | Local | Local | Divisional operations |
The three shared layers, contracts, governance, and talent, were consolidated into a group platform function that served all three divisions on a transparent chargeback model, so each division saw its own consumption and no division subsidized another. The three local layers, pipelines, use case models, and adoption, stayed with the divisions that understood their own operations and their own frontline workers, because context cannot be centralized without losing the very thing that makes a use case land. This split let the corporate center negotiate one vendor contract instead of three, publish one governance policy every division inherited, and pool machine learning talent that rotated to wherever the demand was, while leaving the actual use cases and their adoption firmly with the people who ran the floor. The readiness assessment was the instrument that made the split defensible. Because all three divisions were scored on the same six dimensions, the leadership team could see that Heavy Industrial and Consumer Products had wildly different use case portfolios but nearly identical gaps in governance and MLOps. That was the evidence that settled the argument: the divergence the divisions felt so strongly was real at the use case layer and imaginary at the foundation layer, and the operating model was drawn to match.
What the shared posture produced
- Vendor and cloud spend consolidated into one group contract, cutting combined platform cost by about 2.1 million dollars a year against the three-way baseline.
- A predictive maintenance capability built in Heavy Industrial was ported to Field Services fleet equipment in six weeks, rather than rebuilt from zero.
- The shared talent pool cut the group's open machine learning roles from a chronic backlog of 11 to 3, because scarce hires served three divisions instead of one.
- One inherited governance policy and model inventory replaced three divergent ones, giving the group a single audit posture across all three businesses.
- Divisional use cases stayed local and kept their adoption ownership, so no division felt flattened into a template that ignored its operating reality, and buy-in held because the shared layers demonstrably saved money rather than adding a corporate tax.
What transferred beyond this engagement
- Different businesses can share a foundation without sharing a template. Separate the common layers from the local last mile and assign ownership deliberately.
- The layers worth sharing are the expensive, scarce ones: contracts, governance, and specialist talent. The layers worth keeping local are the ones that need operating context.
- Run the same readiness assessment across divisions. A common scale exposes where divergence is real and where it is just three teams reinventing the same wheel.
- Portability is the payoff of a shared foundation. When the plumbing is common, a capability built once can move across divisions in weeks.
- Leave adoption with the operators. Central teams can build platforms, but only the division that runs the floor can make a use case stick.
How to run this pattern yourself
- Map every division's AI spend and identify where contracts, governance, and talent are being duplicated across the group.
- Draw an explicit line between shared foundation layers and local last-mile layers, and assign clear ownership to each.
- Run one common readiness assessment across all divisions so divergence is visible on a single page.
- Consolidate the scarce, expensive layers into a group platform function while leaving pipelines, use cases, and adoption with the divisions.
- Design for portability from the start, so a capability proven in one division can move to another without a rebuild.