Every large company carries an asset its accountants cannot see. Not the plants, the fleet, the inventory, or the brand equity that gets a line in the annual report, but the accumulated judgment of the organization: how it prices, how it handles a difficult customer, why a supplier contract was structured one way rather than another. That knowledge exists. It simply exists badly, spread across customer relationship management records, PDFs, Slack threads, and the memory of people who will eventually retire or resign. Aparna Natarajan, Senior Client Executive at Microsoft, argues that the gap between companies that consolidate this material and companies that leave it scattered is about to show up in valuations. Her case is uncomfortable for most boards, because the cost of doing nothing has been paid and few have bothered to measure it.
The Model Is Not The Advantage
The prevailing assumption inside enterprise AI budgets is that capability comes from the tool. It does not. The same frontier models are available to every competitor in every sector, on roughly the same commercial terms, with roughly the same performance. A bank and its nearest rival can sign near-identical agreements and deploy near-identical systems. Whatever differentiation existed in the procurement decision evaporated the moment the second company signed.
What remains is inputs. “Everyone has access to the same AI. Your competitors bought the same thing you did. The difference is what you feed it,” Natarajan says. The consequence is a reframing of where AI spend should go. If the model is commodity infrastructure, then the budget line that matters is not licensing but the plumbing that connects proprietary history, customer records, and institutional reasoning to the system. “A company that connects its own history, its own customers and its own hard-won judgment gets answers nobody else can get. A company that doesn’t gets the same generic output as everyone else. That gap is the asset.” Read that carefully and it becomes an indictment of a great many AI pilots. A deployment that runs on public data and generic prompts is not a competitive program. It is an expensive subscription to the same answers the competition receives.
The Cost You Are Already Paying
There is a version of this argument that sounds like a growth story, and there is a sharper version that is a leakage story. Natarajan makes the second one. Scattered knowledge is not a missed opportunity sitting in the future. It is a recurring operating expense being incurred right now, invisibly, in every part of the business, and it does not appear on any report a chief financial officer reviews.
She lists where it drains out. “Every time someone rebuilds an analysis that already exists, you pay for it. Every time a good decision from two years ago gets forgotten, you pay for it. Every time an experienced person leaves, you pay for it.” Then the observation that should worry any executive team: “Most leaders have never seen that number. It’s larger than they expect.” Consider what each of those three items represents in a company of scale. Duplicated analysis is salaried hours spent reproducing work the organization already owns. Forgotten decisions are the same mistakes relitigated across cycles, with no memory of why the original conclusion was reached. Departing expertise is the permanent loss of reasoning that was never written down, only witnessed. None of it triggers a variance report, because there is no budget line for work the company did twice. That invisibility is precisely why the problem persists through cost-cutting programs that scrutinize far smaller sums. Put a number against it and the business case for consolidation stops being speculative.
An Asset That Appreciates
The distinction that gives Natarajan’s argument its financial teeth is the one between capital equipment and compounding infrastructure. Most of what a company buys begins depreciating on delivery. “Buy a machine and it starts wearing out. Build this and it improves with use. Every decision adds to it,” she says. A unified intelligence layer behaves inversely to almost everything else on the asset register. Usage does not degrade it. Usage enriches it, because every decision routed through the system deposits another piece of reasoning into the store that trains subsequent judgment.
That profile changes who should care. “That’s the profile of a genuine asset, and buyers, boards and investors are starting to price it that way,” Natarajan says, which moves the conversation out of the technology function and into the room where enterprise value is negotiated. An acquirer evaluating two comparable businesses is increasingly assessing which one can reach what it knows. One has decades of pricing logic and customer history structured and queryable. The other has the same decades locked in file shares and the recollections of people who may not stay through integration. The operating assets look identical. The knowledge positions are not remotely equivalent, and the second company has no mechanism to demonstrate what it holds.
The diagnostic Natarajan offers executives is deliberately blunt, and it separates two problems that usually get conflated. “What do we know that nobody else does? And can we get to it?” Most leadership teams answer the first half with confidence and stumble badly on the second. Proprietary knowledge that cannot be retrieved is functionally equivalent to not having it. The organizations that close that gap will spend the next several years compounding an advantage that competitors cannot buy, because it was never for sale.
Follow Aparna Natarajan on LinkedIn for more insights on enterprise AI strategy, knowledge infrastructure, and turning AI ambition into measurable business results.
