The platform is built. The race is towards unlocking value.
Consumer products and manufacturing leaders have already made moves most industries are still debating. Billions are committed to strategic cloud and generative AI platforms, and the workloads are arriving. Whatever argument there was about infrastructure is over.
The demand signal has shifted underneath it. Today’s K-shaped consumer requires precision: winning premium and value shoppers at the same time, market by market, which volume-first playbooks cannot do at speed. Connected equipment has changed the data equation as well. Smart machines, sensors, and digital shelves now stream trillions of records a year across plants, distribution, and retail, and most of that signal never reaches a commercial decision.
The returns have not followed. MIT found in 2025 that 95% of enterprise GenAI pilots show no measurable P&L impact. Only 5% reach production with real returns, against an estimated $30 to $40 billion in enterprise spend. The constraint is no longer ambition or infrastructure. It is converting both into results, function by function.
Three patterns separate the 5% from everyone else
MIT’s research is blunt about the cause: model quality is not what fails.
The first pattern is tools that sit on top of the workflow. Generic assistants get layered onto unchanged processes, never learn the work, and produce high usage with no transformation.
The second is budget that follows the demo. Spend concentrates in front-office pilots that present well, while the returns sit in embedded, governed workflows deeper in the operation.
The third is data treated as exhaust. AI gets pointed at sources with no owners, quality standards, or SLAs, so outputs cannot be trusted and nothing scales past the pilot.
Value gets created where governed data, custom-built capability, and changed ways of working meet. That intersection has to be engineered into the operating model. No vendor sells it.
Four moves that convert investment into returns
The companies pulling ahead are converging on four connected moves. Prioritized use cases feed a governed delivery model, custom builds prove value, and adoption turns that value into new ways of working, which surfaces the next wave of use cases.
01. Identify the right use cases
Start where value, data, and the business agree. A five-stage workshop takes teams from literacy to a committed portfolio:
- AI literacy: align on what AI can and cannot do, using examples that reset misconceptions
- Business landscape: map goals and pain points in each function’s own language
- Applications deep dive: match ML, LLM, and agentic patterns to the challenges that surfaced
- Solution design: build targeted solutions against specific, named problems
- Prioritization: score value, feasibility, and data readiness, then commit with named owners
In asset-intensive businesses, the strongest candidates cluster around predictive maintenance and equipment health, AI-driven pricing and promotional benchmarking, selling-story automation for account teams, inventory and assortment optimization, and chat over governed operational and sales data.
One piece of portfolio discipline holds up in practice: roughly 50 to 60% of investment in operational efficiency, 20 to 30% in customer and growth, and the remainder in innovation bets. Quick wins fund the harder builds.
02. Stand up the operating model
Leading consumer companies are settling on central standards with local delivery. An AI hub owns standards, risk, platform patterns, and shared assets, feeding sales and marketing, field operations, supply chain, and corporate functions.
Three things make that work: a single intake funnel from idea to funded build, with value gates executives trust and operating units can clear in weeks; guardrails for human oversight, data lineage, model monitoring, and cost, defined once and applied everywhere; and formal business intake and support on the platform, so capability outlives any single project. Governance built this way is what lets teams ship faster.
03. Build custom capability
Advantage comes from AI engineered on an organization’s own data and its own cloud platform. That requires the full value chain: strategy and prioritization, data warehousing and source integration, curated data products, model development and training, app development, and product management through rollout. Every agent and model sits on governed, curated data products, including master data cleaned through an AI-assisted data quality toolkit.
04. Change how work gets done
Deploying tools is the easy part. Value shows up when decisions get made differently, so adoption belongs in the design from day one. Embed AI where decisions actually happen, in selling stories, service triage, or promo planning, and retire the old path so the new way is the only way. Roll out by persona, with role-based views aligned to how each stakeholder decides. In global organizations, build fluency through a deliberate change strategy rather than leaving it to organic uptake.
Then measure what the CFO measures: cycle time, margin, retention, and renewal rates, tracked from a baseline. Deployment counts and adoption dashboards are inputs to that, not the goal.
Case in point: a global beverage system
We started with the critical stakeholder teams, working through which insights were actually worth having and, more to the point, what decision each one would unlock. An insight nobody acts on is overhead.
From there we outlined how those insights would be integrated into strategic decisioning, so the AI capability landed inside operational processes rather than beside them.
We then developed the enabling capabilities, improving visibility into the data and surfacing prioritized insights automatically so teams stopped hunting for them. Speed to insight improved 80% after tuning the production data agent’s retrieval and response logic, with higher answer accuracy.
Last came rollout. Training and working sessions gave users a clear path to getting insights out of the tools, and a feedback loop back into the build team turned what they found into rapid enhancements.
Across this industry the platform decisions are largely made. What separates the 5% from the 95% now is execution, function by function, move by move.
