The conversation is just getting underway when a senior IT leader leans forward and cuts in with a question that resets the discussion: What is the ROI and how many roles will this reduce?

This is a familiar scenario that is played out in boardrooms across industries in recent years, but it almost always sets the wrong frame. The instinct is understandable since AI is a significant investment and leadership wants proof of return. But reducing that proof to headcount narrows the value to a fraction of what is actually possible.

In current implementations, I have seen Now Assist deflect up to 35 to 40% of tier one cases. That value is real, measurable and it matters. It is also only the starting point. Organizations that optimize only for that outcome capture roughly 18% of the total value their AI investment generates and spend the remaining 82% wondering why the numbers never quite live up to the promise.

Why headcount ROI is the floor not the ceiling

The instinct to measure AI by headcount reduction is a pattern I see across enterprises that have evaluated technology this way for decades. Using headcount as the only metric is where the problem usually begins.

When headcount reduction is the primary proof of AI value, three things happen. The business case gets approved on Year 1 savings and never revisited. The deployment gets optimized for deflection targets rather than capability growth. Stakeholders conclude the technology has plateaued when efficiency metrics stabilize, right when the compounding was about to begin. This is the linearity trap and it leads organizations to make a procurement decision when they should be making an investment decision.

In practice, deflection tends to plateau around 85 to 88% as automation begins to hit the complexity wall. The smarter reframe is not headcount saved but cost-per-outcome, a metric that keeps improving long after headcount numbers stabilize as the platform handles more volume with the same infrastructure.

Three layers of AI value

Our ROI Blueprint, shaped through our ServiceNow engagements, maps AI value across three layers.

Layer 1 is Operational ROI, the efficiency and optimization layer where most conversations tend to start, but where roughly 80% of boardroom attention is focused. Key metrics include Cost-to-Serve, Process Cycle Time, Automation Coverage, MTTR, Self Service and First Call Resolution. Organizations that measure only here often conclude their AI investment has peaked right when the value is starting to compound.

Layer 2 is Experience ROI, the engagement and enablement multiplier most organizations do not explicitly budget for. Key metrics include CSAT Uplift, NPS Score, Decision Cycle Time, Carbon Reduction and Compliance Rate. Every point of CSAT uplift has a retention impact behind it and every reduction in employee effort score reduces the probability of losing someone whose replacement costs 1.5 to 2 times their annual salary. Layer 2 also captures ESG value that is increasingly a procurement criterion, from carbon reduction and compliance automation to accessibility improvements that affect regulatory standing and employer brand.

Layer 3 is Transformational ROI, the growth and innovation layer most organizations never model in Year 1. Key metrics include Revenue from AI Products, Innovation Index, Customer Retention Uplift and Time-to-Market. This is where AI enables IT to resolve incidents before users report them and services to predict demand rather than react to it. Capabilities become viable at scale that could not have been staffed at any cost because the bottleneck was never headcount but the speed and volume of decision-making required. By month 36, the accumulated operational intelligence becomes a proprietary competitive asset that widens the gap between those who invested early and those who waited.

McKinsey's 2025 State of AI research found that only 6% of organizations qualify as AI high performers while 88% use AI in at least one function. HCLTech’s The Blueprint to AI-led Operating Model research found that organizations with a product-aligned approach are 4x more likely to maximize ROI from every AI dollar spent. The platform may be the same and the timeline may be the same, but the return is very different.

Autonomy not automation

Automation focuses on which tasks a machine can take over instead of a person and leads naturally to headcount conversations. Autonomy shifts the focus to what decisions and capabilities become possible when human intelligence is no longer bottlenecked by routine work, leading to conversations about what the organization can now do that it could not before.

When that question replaces the headcount question, the conversation changes and so does the adoption rate.

Five things to do differently

1. Define value across all three layers before deployment, not after. In many cases, organizations lock measurement into Layer 1 by building their ROI model from the metrics that justified approval. Naming Layer 2 and Layer 3 values in the first scoping conversation creates accountability for realizing them.
2. Set a 36-month value horizon, not a 12-month one. AI ROI follows a J-curve where the first twelve months look modest and the compounding accelerates through Year 3. A 12-month window often leads to the investment appearing underwhelming right when the value is starting to build.
3. Instrument leading indicators alongside lagging ones. Lagging indicators tell you what happened. Leading indicators such as adoption rate by function, deflection trend week on week and proactive resolution rate tell you whether the ROI curve is building momentum or plateauing and serve as the early warning system for Layer 2 and Layer 3 returns.
4. Reframe freed capacity as new capability. When Now Assist reaches a projected deflection level of 60% of level-1 tickets at scale, that freed capacity can go to proactive operations, new service lines or better customer engagement rather than redundancy. Tracking where it is redirected is how Layer 3 ROI gets built.
5. Build a Value Realization Office or equivalent. The single biggest predictor of whether an organization realizes Layer 2 and Layer 3 ROI is whether someone owns the measurement story post-deployment. A named function that tracks value across all three layers and reports to leadership quarterly is the difference between organizations that see the full return and those that wonder why it never materialized.

The questions worth asking

The next time a boardroom conversation opens with a headcount question, answer it. Then ask a bigger set of questions:

1. What capacity has AI made available and what new work is that capacity now doing?
2. What is our cost-per-outcome today compared to 12 months ago across all three value layers?
3. Where has AI improved the quality of a decision, not just the speed of a task?
4. What services or capabilities exist today that could not have been staffed before AI?
5. How much richer is the operational data estate and what is that institutional intelligence worth as a strategic asset?

The organizations that lead with AI will be the ones that asked bigger questions and built the measurement model to match.