From Pilot to P&L: How Business Context Turns AI Spend Into AI Returns

AI is now 1400+ days old. Personal use of AI across conversations, shopping recommendations or even  personal consultations is fast becoming second nature. In sharp contrast, AI’s impact in business is on a jagged  frontier, some peaks but many troughs. In fact, PwC’s 29th global CEO survey reveals only 12% of CEOs managing to extract meaningful (> 2%) benefits through AI deployment.

A regional sales head at a global beverage company asks an AI assistant why volumes fell in the Northeast last quarter. Within seconds she gets a neat chart and a confident summary: lower promotional spend, softer convenience-store sales. Every number is correct, and the answer is still useless. The AI does not know the region pulled back on promotions on purpose, to protect margin after a price increase. It has described what happened without understanding why, so it cannot say whether  anything needs to change.

This is the gap between "Absolved Intelligence", where AI is handed a task and left to get on with it, and "Accompanied Intelligence", where AI works alongside people with the same understanding of the business that they have. Closing that gap is as much a managerial shift as a technical one. Conversations with leaders across consumer goods, retail and manufacturing keep pointing to the same conclusion: scaling AI takes more than better models or bigger token budgets. It rests on business context.

The missing system of record

Enterprise systems record transactions, customer interactions and employee information. They are less effective at capturing why a distributor received additional credit, why marketing budgets shifted between brands, or why inventory was increased despite a working-capital constraint. That’s easier said than done given such context sits in the minds of employees, stakeholders, CXOs and thus inaccessible to AI.

We therefore, need a new system of record for capturing this proverbial ‘secret sauce’. What is needed is a complementary foundation for organisational decision-making: an Intelligence Warehouse (IW). Rather than replacing any system of records , it would connect their information with business definitions, performance  measures and decision histories.

Three layers of business context

The IW is built business first - it starts from how the organisation sets goals and makes decisions, and only then brings in the data that supports them. It does this in  three connected layers.

First, the ontology: a structured map of business entities and their relationships—products, brands, customers, channels, geographies and pricing. Drawing on internal and external data, it establishes a shared vocabulary and defines what the system can see.

Second, the performance framework: the organisation’s specific definitions of success. These might include sales by value and volume, market share, on-time-in-full delivery, salesforce effectiveness, return on advertising spend (ROAS) and granular portability. Crucially, it captures how these measures are calculated and related. Higher sales, for instance, need not mean higher profit.

Third, the decision layer: a record of what was decided, why, under which constraints and with what outcome. Most of this judgement has never been written down.  An experienced supply planner may build extra safety stock for certain South Asian markets every year before the monsoon, because she knows port delays will spike. No policy document says so. The IW draws this kind of judgement out through structured conversations with the people who make the calls, then encodes it as decision circuits that can be reused, audited and  improved.

These three layers help bring context in data at enterprise scale

From “what” to “why” to “so what”

In a demonstration for a large consumer goods company, we compared a general-purpose LLM setup with an IW-based approach using similar organisational data.

The questions became progressively more demanding: What happened to sales across quick-commerce platforms? Why did advertising returns differ? And how should marketing expenditure be reallocated?

In that demonstration, the general-purpose setup’s answers became less useful as the questions moved towards recommendations. The IW-based approach could draw on richer, explicitly structured business context.

This was an illustration, not a controlled benchmark.

The distinction was not simply the model’s context-window size, but the relevance and accuracy of stating what happened and thus what needs to be done. Even a fairly detailed prompt cannot substitute for well-defined business relationships, objectives and constraints.

People use the IW in plain language, through whichever large language model they prefer. The three layers are laid out like a map, so you can think of it as Google Maps for the organisation. When the IW answers a question, its agents follow a structured, traceable and repeatable path through the graph. It also keeps learning from feedback. Together, these features turn AI from a "black box" into a "glass box", where business users can see how each answer was built. A CFO reviewing a recommendation to cut a brand's media budget can see which definition of margin was used, which past decisions were considered and how the conclusion was reached.
That is what makes an AI answer something a leadership team can act on.

Built to evolve and to act

The IW adapts as the organisation changes. New nodes can be added for a new geography or a category launch. It also learns from new decisions, including ones made in response to events the business has never faced before (for example, clearance markdowns started two weeks early during an unusual heatwave), and remembers them for next time. Because it reasons over a structured graph, it uses tokens mainly at the plain-language input and output stage. That keeps compute costs in check.

IW can also support a ‘harnessed’ cross-functional agentic layer. Agents powered via IW take role-specific actions based on the IW's "so what" guidance. We call this the temporal trinity. It connects:

• the recent past: why something happened
• the near future: what the organisation should do next
• an agentic present: the actions that need to be taken now

This matters because many organisations are struggling to get value from AI agents. Gartner expects more than 40% of agentic AI projects to be cancelled by the end of 2027, often because their business value is unclear.

IW also gives an ability to build a ‘harnessed’ cross functional agentic layer that can act on persona based actions basis the ‘so what’ response from the IW platform. This brings out the temporal trinity of connecting an immediate past (why what happened) with near term future (so what an organization needs to do) with an agentic present (actions that need to be taken now to deliver the so-what direction).

The AI substrate for the enterprise

IW is a powerful answer to make AI deployments at enterprise scale on back of rich and relevant context creation, seamlessly bridge functional silos, build transparency via NLP interfaces and imbue ability to manage complexity both in agility and effectiveness of decision making by considering end to end data points. This is what business gut backed with data rationality can closely look like