How enterprise leaders use AI for business growth

About 95% of enterprises see zero return on their generative AI investment, according to MIT Project NANDA's State of AI in Business 2025. Only around 5% of integrated pilots reach a measurable impact on the P&L. If you've signed off an AI budget to drive growth, that finding probably stopped you cold.

I understand why that gap stings. You funded the pilots, watched the demos, and read the case studies. The spend is real, but the returns, so far, have been dismal.

This is a guide to AI for business growth as enterprise leaders actually practise it — not the small-business tool roundup the term usually returns. I’ll show you where AI is already producing growth outcomes across the major business functions, why most deployments underdeliver once they go live, and what separates the few that move the business from the many that don't.

AI for business growth at a glance

• Enterprise AI adoption is near-universal: 88% of organizations now use AI in at least one function, according to Stanford HAI.
• Few enterprises have seen a return on their AI investment: about 95% of enterprises see no measurable return on generative AI, per MIT Project NANDA.
• Some of the strongest use cases for enterprise AI sit in strategy, marketing, sales, supply chain, and analytics — the functions that own growth decisions.
• Most enterprise AI use cases underdeliver for one structural reason: separate function-level models produce answers that contradict each other.
• Enterprise Decision AI is the layer that connects them into one causal model, so every function reads from the same truth.

How enterprises are using AI to drive business growth

The clearest growth outcomes show up in the functions that own growth decisions: strategy, marketing, sales, supply chain, and analytics. Across all five, the pattern is the same. AI earns its return when it measures what actually drives the result, and underdelivers when it optimizes activity in isolation. Here's where each function is using it, and what separates the deployments that move the business from the ones that don't.

Strategy and corporate planning

In strategy and corporate planning, enterprises use AI to decompose growth — to see which drivers actually moved the business, not just which activities correlated with a good quarter. The work is causal: separating the effect of price, distribution, media, and macro conditions from one another.

Most enterprises never get there. Only 39% of organizations attribute any EBIT impact to AI, McKinsey found, and most of those put it under 5% — because their models describe what happened without explaining what caused it.

A causal model closes that gap. It replaces the annual planning ritual of negotiated guesses with a clear read on what each lever returns, and that read is what produces results. Picture a Fortune 50 automotive manufacturer running a single causal model across 50 states and 15+ vehicle lines, reaching 90%+ planning accuracy not by forecasting harder, but by understanding which drivers actually move demand.

Causal Clarity is where that starts. It is the difference between knowing what happened and knowing what caused it.

Marketing and advertising

Enterprises use AI to measure incrementality in marketing and advertising. That is, the real, causal effect of spend, net of everything else happening at the same time. This use case exposes the limits of attribution, which splits credit across the touchpoints present at a sale — as if being there were the same as causing it.

That distinction decides where your budget goes. When you know what each channel actually returns, rather than what it claims credit for, spend moves toward the drivers that grow the business, and the halo effects between products become visible rather than assumed.

Most marketing AI never reach that view. They optimize clicks and cost-per-acquisition inside a single channel, then hand the CFO a number the finance model cannot reconcile with anything else in the business. That is how a function that measures everything still cannot say what its spend returned.

Sales and revenue operations

In sales and revenue operations, enterprises use AI to forecast pipeline and quantify what drives conversion, so commercial plans rest on causal signal rather than rep optimism. The use case connects demand-side activity to revenue outcomes the finance team will recognize.

The value shows up in a forecast you can defend. A model that explains why pipeline converts, not just that it might, lets revenue and finance agree on a number before the quarter starts.

Most sales AI aren’t capable of seeing that far. They score and rank accounts inside their own pipeline, with no read on what marketing spend or a supply constraint is doing to the very same accounts they just prioritized.

Supply chain and operations

Enterprises use AI to forecast demand and plan inventory against causal drivers — promotions, pricing, seasonality, and the marketing activity that creates the demand in the first place. This enterprise AI use case ties operational planning to commercial reality.

The outcomes here can be solid. Picture a major omnichannel retailer with 2,000+ stores cutting excess inventory by 18% while growing same-store sales. The gain comes from planning supply against one demand model, rather than a forecast that never saw marketing's plans.

Most demand forecasting still runs on the operations side alone. When there’s no unified plan across the business, supply and demand teams optimize against different numbers and reconcile after the fact.

Data analytics and engineering

Traditionally, data analytics and engineering describe what happened. Enterprise AI goes beyond this to modeling why, and to test a decision before they commit to it. It’s scenario simulation grounded in cause and effect.

This is where most analytics functions hit their limit. They can show you any metric, sliced any way, but they cannot answer the counterfactual — what would have happened without this investment — and that is the question every budget decision turns on.

The difference is causal. Correlation-based machine learning is "flawed, at best" for anticipating the impact of choices, MIT Sloan Management Review argues, because it can tell you what moves together but not what moves what. Causal modeling is the lever for high-stakes calls.

Why most enterprise AI deployments underdeliver

When enterprise AI underperforms, most people think the problem is data quality, or the model, or maturity. But the real problem is often architecture. Each function runs its own AI on its own data, and the answers it produces disagree with the answers next door.

This is fragmentation, and automating it makes it worse. When marketing's model says one thing about a campaign and finance's model says another, you’ve only scaled the disagreement rather than added intelligence.

The mechanism is specific. Traditional marketing-mix models are built on ordinary least squares regression, which estimates one variable's marginal effect at a time. Add a second function's drivers and the model's assumptions break. 

So each team builds its own, and you end up with as many truths as you have models. This is why only 5% of companies are generating substantial value from AI while 60% are laggards, on BCG's reading.

How Enterprise Decision AI connects every use case

Enterprise Decision AI is the layer that connects every use case to one causal model, so strategy, marketing, sales, and supply chain all read from the same truth. It is the category DATA POEM's POEM365 was built to deliver, and it answers the question the function-by-function approach cannot: what is actually driving your business, across all of it?

The engine underneath it all is FOUNT, a Large Causal Architecture that maps cause and effect across every driver at once rather than estimating them one at a time. It is pre-trained on 250+ billion consumer transactions and $5 trillion in spend data, then fine-tuned to each brand.

The proof is in the deployment. POEM365 reaches 90%+ forecast accuracy at go-live and is ready to deploy in 6 weeks, which is what makes connecting your use cases a decision you can act on this quarter rather than next year.

Where AI for business growth proves itself in your industry

Connect every use case to one causal model, and the question stops being which AI to buy and becomes which drivers actually move your business. The clearest outcomes I see come from data-rich, decision-dense industries where those drivers are tangled and the cost of a wrong call is high.

Take consumer packaged goods. Unilever connected live weather data to the demand forecasting behind its ice cream business, and added image-capture AI to 100,000 of its freezer cabinets so it could read stock in real time. The company reports that data from those freezers lifted retail orders and sales by up to 30% in its strongest markets, with forecast accuracy in Sweden improving by around 10%.

Now, AI isn't magic; that's not the point. The point is that the right driver — weather — was wired to a specific commercial decision, and the model could measure precisely what that driver did to demand. 

That is the simple case: one signal, one decision. Most enterprises face the harder version, where dozens of drivers move at once and the job is reading all of them together. That is exactly where one causal model earns its place.

Which drivers dominate business growth in each sector

In CPG broadly, the central problem is portfolio effects: a promotion on one SKU pulls demand from another, and a media push lifts products it never named. Modeling those effects causally is what separates real growth from reshuffled revenue.

In retail and ecommerce, the live battle is the walled garden — Amazon Ads, Walmart Connect, and Instacart each control their own data and attribution. A single causal model across them is the only way to compare like with like and allocate across every channel.

In automotive, scale is the challenge: demand for fifteen vehicle lines across fifty states moves on different drivers, and a forecast built region by region never reconciles into one number the board can plan against.

In QSR, where pricing, weather, and local promotions interact daily, the same causal-planning approach turns volatile demand into a number the operations team can trust.

The pattern holds wherever the drivers are tangled: name them, model them together, and the decision gets clearer.

Frequently asked questions

How do you measure ROI on AI for business growth?

You measure the ROI of enterprise AI by incrementality: the causal effect of the deployment on revenue, cost, or margin, net of what would have happened anyway. Efficiency gains alone are not ROI. The defensible figure compares the decision the model enabled against the counterfactual, not against a previous tool.

What's the difference between an AI pilot and production AI?

An AI pilot tests whether something works in a contained setting; production AI is a live, repeatable application changing a real decision at scale. The gap between them is where enterprises lose the most value: only about 25% of organizations have moved 40% or more of their AI pilots into production, according to Deloitte. Most experimentation simply never becomes a scaled, live deployment.

How can AI automation scale with business growth?

AI automation scales with business growth when it runs on one shared causal model rather than separate per-function tools. Automating a fragmented process scales the fragmentation: every team's automation reads from a different truth, and the conflicts compound as the business grows. A single causal model gives every function the same read, so adding scale adds clarity rather than contradiction.

What role does AI play in pricing strategy?

AI supports pricing strategy by quantifying how price interacts with the other drivers that move demand — promotions, seasonality, competitor activity, and marketing. Correlation-based models treat price in isolation; a causal model measures the actual effect of a price change net of everything else, which is what a defensible pricing decision requires.

How long does enterprise AI take to deploy?

Deployment time depends on the architecture. A pre-trained causal foundation model such as POEM365 is ready to deploy in 6 weeks, with individual use-case extensions added in roughly six weeks each thereafter. Building an equivalent capability in-house typically runs to many months before it supports a single decision.

Do you need to build your own AI agents?

No — you do not need to build your own agents from scratch. A pre-trained platform supplies the causal engine and the agents that run on it, so the work is fine-tuning to your data rather than constructing models. Building bespoke agents per function tends to recreate the fragmentation that caps value.

What do enterprises need before they can trust AI with big decisions?

Enterprises need explainable models and unified data before executives will trust AI with consequential decisions. Governance fails when each function holds its own data and its own black-box model. A single causal model with traceable cause-and-effect logic gives executives a decision they can audit, which opaque correlation-based systems cannot.

Start driving growth for your business with Enterprise Decision AI

So where do you begin? Start by counting your truths. Pull the current revenue forecast from finance, the demand plan from supply chain, and the marketing plan, and lay them side by side. If they disagree, no amount of new use cases will reconcile them — the disagreement is built into the architecture.

The move that changes your return is connecting what you already have to one causal model — what DATA POEM is built around. Pick the decision where the conflict costs you most and model it causally across functions rather than within one. That is where the value the pilots promised finally shows up, and it is the work Enterprise Decision AI was built to do.