THE 160% AI ROI GAP: WHY ONLY THE TOP TIER WINS AND WHAT THEY DO DIFFERENTLY

Here is a number that should make every business leader sit up and take notice. The ServiceNow Enterprise AI Maturity Index for 2026 reports that top-tier companies achieve 160% ROI on their AI investments. Meanwhile, Gartner's latest data tells us that only 28% of AI projects deliver any meaningful return at all, 20% fail outright, and the remaining 52% produce results so mediocre they barely register on the balance sheet.

That gap — between the 160% winners and the 28% who break even — is not about technology. It is about how you think about AI in the first place. And the uncomfortable truth is that most organisations are making the same three mistakes, regardless of industry, size, or budget.

THE FIRST MISTAKE: SPEEDING UP BROKEN PROCESSES

The single biggest predictor of low AI ROI is treating artificial intelligence as a productivity overlay on existing workflows rather than a catalyst to redesign them from the ground up. McKinsey's six-year longitudinal study found that fewer than 100 companies globally have captured more than two-thirds of all enterprise AI value. What do they share? They do not ask "how can AI make this 20% faster?" They ask "if we rebuilt this process with AI from scratch, what would it look like?"

A 20% speed increase on a broken workflow produces a broken workflow that runs 20% faster. It does not produce a competitive advantage. It does not transform your cost structure. It does not change your market position. It just means you fail faster.

THE SECOND MISTAKE: EXPECTING YEAR-ONE RETURNS

IBM CEO Arvind Krishna laid out the real timeline in a recent interview. Year one is net negative. You are building infrastructure, training teams, cleaning data, and establishing governance. That is not failure — that is investment. Year two is where the returns start appearing, often at approximately 10x on what was deployed. By year four, mature enterprises track billions in cumulative savings. IBM itself is on track for over $5 billion in AI-driven savings by the end of this cycle.

But here is the problem: most organisations do not have a four-year attention span. They launch a pilot in quarter one, expect measurable EBIT impact by quarter two, and declare AI a disappointment by quarter three when the numbers do not materialise. The Morgan Stanley research from 2025 — still widely cited in 2026 — projected that agentic AI and automation could reduce annual S&P 500 operating expenses by approximately $920 billion, roughly 28% of expected pretax earnings. That is a multi-year transformation, not a quarterly initiative.

THE THIRD MISTAKE: NO GOVERNANCE, NO SCALE

The difference between a pilot that dies in a spreadsheet and a programme that scales across the enterprise is governance. The top-tier companies in the ServiceNow index share a common architecture: an orchestration layer that manages agent deployment, tracks ROI by use case, enforces security boundaries, and creates audit trails. They treat AI as an operating system for the business, not a collection of tools.

This is where the concept of the control tower becomes critical. Organisations that embed AI into a central orchestration platform — whether ServiceNow, their own stack, or a hybrid approach — report significantly higher satisfaction and measurably better returns. Those that let individual teams buy their own AI tools, connect them to nothing, and report results in isolation? They populate the 52% mediocre middle.

A FRAMEWORK FOR BRIDGING THE ROI GAP

If you are reading this and wondering where your organisation sits on that curve, here is a practical framework to assess yourself. It has four questions, and honest answers will tell you whether you are heading toward 160% or the 28% average.

1. Are you redesigning or accelerating? If your deployment plan starts with "we will plug AI into our existing CRM/case management/supply chain system and see what happens," you are accelerating. You will get incremental gains and limited ROI. If you start with "what would this process look like if we rebuilt it around autonomous agents with human oversight," you are redesigning. That is where the 160% lives.

2. Do you have a multi-year budget or a quarterly experiment? The winning organisations allocate AI spend as a three-to-five-year capital programme, not an operational expense line item reviewed every quarter. They accept year-one negative returns because they have modelled years two through five and know the curve bends upward.

3. Who owns governance? If every department manages its own AI tools independently, you do not have governance — you have a sprawl that will eventually produce a security incident, a compliance failure, or both. Centralised orchestration is not bureaucracy; it is the prerequisite for scale.

4. Can you name your top three metrics? The highest-ROI organisations define exactly three operational metrics before deployment — cycle time reduction, cost per transaction, error rate — and measure them continuously. If you cannot name your metrics without referencing a dashboard that nobody has opened in six weeks, you are not ready to scale.

THE BOTTOM LINE

The gap between AI hype and AI ROI is not narrowing. If anything, mid-2026 data shows it is widening, because the easy implementations have been done and the hard work — process redesign, governance, multi-year commitment — is what separates the top tier from the rest. The same McKinsey study that found only 100 companies capturing most AI value also found that organisations scaling AI across multiple functions achieve roughly double the profit margins of those stuck in isolated experiments.

The question is not whether AI can deliver returns. It clearly can — the 160% figure is proof. The question is whether your organisation is built to execute differently, or whether you are hoping that layering a chatbot on a broken process will somehow produce a different result from the 52% who are still waiting for their ROI to show up.

AI is not a magic wand. It is an operating model transformation. If you treat it like one, the numbers follow. If you treat it like a tool purchase, they do not.

— Priya Sharma