Most organisations do not have an artificial intelligence (AI) strategy. They have an AI shopping list, and they have discovered that the two behave very differently under pressure.
The evidence is not subtle. MIT’s NANDA initiative reviewed more than 300 publicly disclosed enterprise generative AI deployments and found that around 95 percent had delivered no measurable return. The figure is often quoted as proof that AI does not work. That is the wrong reading. It works. What fails is the decision-making that surrounds it, because almost none of these initiatives began with a diagnosed business problem worth solving.
This is the first of six articles that will help you build an AI strategy that survives contact with your board, your budget and your competitors. Each one takes a single force of your business, examines what AI genuinely changes there, and asks whether that change is worth building a strategy around. By the end of the sixth, you will have a defensible answer rather than a list of tools.
We start with Demand, because every other force in the business is downstream of it.
The Six Forces of Business
Across strategic business and technology planning engagements for dozens of organisations globally, Unisphere has consistently found that every business, regardless of sector or size, runs on six dominant forces. They are Demand, Marketing, Sales, Value Delivery, Back Office Operations and Finance. Each is employed or leveraged according to how well it helps that organisation win in the markets it has chosen, with the products or services it takes to those markets.
The framework matters here for one reason. AI does not apply evenly across a business. It changes the economics of some forces profoundly and others barely at all, and the difference is specific to your organisation. Working through the six forces in order is how you find where your advantage actually lives, rather than where the vendor demonstration suggested it might.
What Demand actually means
Demand is not your sales pipeline and it is not a market sizing exercise. It is the deep, evidenced understanding of what your market wants. That means the specific features and functions customers value and will pay for, the price elasticity in your category, the channels customers prefer to buy through, the levels of service expected before and after the sale, and where each competitor is positioned in the customer’s mind rather than in your own competitive matrix.
Almost every executive team believes it understands demand. Very few can evidence it. What usually exists is a composite of long-tenured opinion, the loudest customer’s most recent complaint, a win and loss process that was designed but never operated, and a product roadmap shaped by whoever argued most persuasively in the last planning session.
That gap is the single most common root cause of failed technology investment we encounter. An organisation invests heavily in delivering something the market never asked for, then concludes that execution was the problem.
What AI genuinely changes here, and what it does not
The honest answer is that AI does not help you understand demand. It collapses the cost of synthesising demand signal you already own.
Consider what most organisations are already sitting on. Support tickets going back years. Lost deal notes. Recorded discovery calls. Onsite service reports. Churn and cancellation reasons. Quote to close data showing exactly which configurations win and at what discount. Product review text. Website search queries recording the words customers use, which are almost never the words in your marketing. Historically, turning that unstructured material into decision grade insight required a research project, a budget and a quarter. That cost has fallen close to zero.
This is a genuine change in the physics of the Demand force, and it is available now. Language models are effective at exactly this task: reading enormous volumes of messy human text and surfacing patterns, themes and frequency. Applied to eighteen months of lost opportunity notes, this routinely produces conclusions that contradict the executive team’s stated understanding of why deals are lost. That contradiction is the valuable output.
Be equally clear about the limits. AI has no access to demand signal you never captured. If your sales team records a lost deal as “price” in a dropdown and moves on, no model will recover what actually happened in that room. If you have never varied price, no analysis will reveal your elasticity. If you do not speak to customers who did not buy, your data describes only the market you already won. Instrumentation comes first. This is unglamorous, it is usually a process and accountability problem rather than a technology one, and it is where the work genuinely starts.
There is also a category error worth naming. Using a general purpose model to tell you what your market wants, based on its training data, is not demand analysis. It is a competent summary of published industry commentary. It is available to every competitor you have, at the same moment, for the same price. It cannot be the basis of an advantage.
The bench strength question
Before building a strategy on this force, ask a harder question. Is your organisation successful because of superior demand understanding?
For some businesses the answer is yes, and it shows. They ship products the market was waiting for, they price with confidence, and they enter adjacent categories successfully because they read the signal early. For most, the answer is no, and the success came from somewhere else entirely: a hero product, an efficient cost base, a distribution position, genuine agility, or a delivery capability competitors cannot match.
Both answers are useful. If demand understanding is your bench strength, AI applied here compounds an existing advantage, which is the highest return move available to you. If it is not, you have found a weakness, and the strategic question becomes whether closing it changes your competitive position or simply makes you marginally better at something that was never the constraint.
Answering that honestly is more valuable than any tool selection.
From insight to strategy
Here is the discipline that separates the two.
“We will use AI to better understand our customers” is not a strategy. It is a sentiment. There is no obstacle diagnosed, no trade-off committed to, nobody accountable, and no way to know whether it worked.
A strategy is a cohesive set of actions that overcomes a significant obstacle or captures a significant opportunity. It carries guiding policies stating what will and will not be done, so that trade-off decisions can be made quickly and the plan stays on track. It names an accountable owner. Its activity and its objective are both measurable. And critically, it is achievable with the resources you actually have.
Applied to Demand, a real strategic objective sounds more like this. We lose a material share of qualified opportunities to one competitor, and our own notes cannot tell us why. Within two quarters we will know, and our roadmap will be governed by that evidence rather than internal advocacy. The guiding policy might be that we will commission no new market research until we have exhausted the signal we already hold, and that no roadmap item proceeds without demand evidence attached. The accountability sits with a named executive. The measure is competitive win rate against that competitor.
Then apply the defensibility test. Richard Rumelt’s standard is that a strategy holds when replicating it would be too expensive, too slow or too complex for a competitor, or where you hold a genuine protection such as a patent, sole distribution rights, or control of a scarce resource.
Your accumulated customer interaction data passes that test. It is scarce, it is proprietary, it compounds with time, and a competitor cannot buy it at any price. Your access to a frontier model does not pass it, because your competitor signed up for the same one this morning.
That distinction is the foundation of everything that follows in this series.
Where this leaves you
Before reading Part 2, do one thing. Write down, in a single sentence, the evidence base for your organisation’s current understanding of what its market wants. If that sentence describes a system, you have a bench strength worth compounding. If it describes a person, you have a diagnosis.
Part 2 examines Marketing, the act of relaying that understanding compellingly and setting the correct expectations. It also introduces the equation that governs the next three forces: customer experience equals expectation minus reality. AI has made it remarkably cheap to inflate the first term, which is a strategic risk that very few organisations have priced.
Unisphere works at the convergence of technology, strategy and AI with the business knowledge to connect them. If you would like a structured diagnosis of where your AI strategy should concentrate, rather than a list of things you could buy, we should talk.