What Is a Realistic ROI for an AI Project?

The honest answer is that it depends almost entirely on scope, and most published figures are not helpful because they pool together very different types of AI investment. Enterprise AI programmes spanning multiple years and hundreds of thousands of pounds look nothing like a focused commercial workflow transformation. Treating them as the same category produces misleading expectations in both directions.

The Two Very Different Pictures

There are two distinct ROI profiles for AI investment, and confusing them is the most common cause of misaligned expectations.

Broad enterprise AI adoption covers large-scale, multi-function AI programmes across complex organisations. Deloitte's 2025 survey of over 1,800 executives found that most organisations report satisfactory ROI within two to four years, with only 6% seeing payback in under a year. These are programmes with significant upfront investment, long change management cycles, and results that take time to compound.

Focused commercial AI projects are a different category. A tightly scoped engagement targeting one specific workflow or commercial function, with a documented baseline and a serious adoption plan, operates on a considerably shorter timeline. On a recent agentic transformation project with a B2B SaaS scaleup, 1,800 hours were freed per year across Sales, Marketing, and CS, delivering over £100k in annualised operational savings. The first measurable results appeared at week five. Full payback on the implementation investment was reached at month six.

Both figures are accurate. They describe different things. The question is which category your project falls into.

What Drives ROI Up

Scope specificity. Projects targeting one well-defined commercial problem with a clear cost consistently outperform projects with a broad brief. The narrower the scope at the start, the faster the value is realised.

A documented baseline. ROI cannot be measured without a before state. Teams that document hours, costs, and output metrics before the project begins can demonstrate results precisely. Teams that do not are guessing after the fact.

Adoption. The most technically correct AI system delivers zero ROI if the team does not use it. Projects with a serious enablement plan, where people understand how their role changes and are properly trained on the new workflows, consistently outperform those where the technology is handed over and adoption is assumed.

High-volume, repetitive workflows. The best candidates for AI transformation are processes that consume significant time, happen frequently, and follow a pattern with variation. The higher the volume and the higher the hourly cost of the people doing it, the faster the ROI compounds.

What Drives ROI Down

Vague scope. Projects briefed as "use AI to improve efficiency" produce inconsistent results because there is no clear outcome to optimise for. The technology gets built. The commercial value is ambiguous.

No baseline. The most common ROI measurement failure is not establishing the before state. Without it, you cannot prove the after state, even if the results are genuinely strong. This kills future investment even when the first project delivered value.

Measuring too early. ROI measured at week four is rarely representative. The team is still adapting. Workflows have not bedded in. Four weeks of data from a transitional period does not reflect steady-state performance. Measure at three months minimum.

Underestimating the cost of transition. Most ROI models account for the implementation cost but not the productivity dip during transition, the internal time spent on change management, or the ongoing management cost of running AI systems in production. A rigorous ROI model includes all of these.

How to Calculate It Properly

The ROI formula is straightforward:

The challenge is not the formula. It is measuring the inputs accurately.

Value generated should be calculated as time saved multiplied by fully-loaded hourly cost, annualised. If the project also delivers revenue impact or headcount efficiency, those are additional lines, not substitutes for the time saving calculation.

Cost of transformation should include: implementation fees, internal time committed to the project (often ignored), tooling and platform costs on an ongoing basis, training and enablement time, and any productivity dip during transition.

The figure that matters most to a leadership team is the payback period: how many months until the cumulative saving exceeds the total cost. For focused commercial AI projects this is typically six to twelve months. For broader enterprise programmes it is longer.

The Three-Tier Model for Presenting ROI

Before any AI project is approved, the ROI case should be presented in three tiers.

Conservative case. The minimum credible outcome. What you are confident the project will deliver even if adoption is slower than expected and the technology underperforms. This is the number that justifies the investment.

Base case. The expected outcome based on comparable projects and your specific baseline. This is the number that anchors the conversation.

Upside case. What is possible if adoption is strong and the technology performs well. This is the number that makes the case compelling.

If the conservative case alone does not justify the cost of the project, the scope needs revisiting before anything is built. A project that requires the upside case to break even is not a well-scoped project.

What Good ROI Looks Like in Practice

Based on delivered commercial AI transformation projects, a well-scoped engagement should produce:

A payback period of six to twelve months from full deployment. Year one savings of £50,000 to £150,000 for a mid-size commercial team, depending on team size, hourly cost, and the volume of work being automated. Annualised savings that increase in year two as agents are refined, adoption deepens, and additional workflows are brought into scope.

These are not projections. They are the profile of a project that was scoped correctly, baselined properly, and adopted seriously.

The Question Nobody Asks

Most ROI conversations focus on what the project will return. The question that is rarely asked is what the business looks like in twelve months if the project does not happen.

If the Sales team continues spending forty percent of their time on admin, what does that cost over the next year? If the CS team cannot scale without adding headcount, what does the next hire cost, and the one after that?

The cost of inaction is often larger than the cost of the project. Framing the decision that way reframes the question from "should we spend £30,000 on this?" to "can we afford to keep losing £80,000 a year?"

Frequently Asked Questions

What is a realistic ROI for an AI project? For focused commercial AI projects targeting a specific workflow or function, year one ROI of £50,000 to £150,000 in time and capacity savings is achievable for a mid-size team, with payback in six to twelve months. For broad enterprise AI programmes, most organisations report satisfactory ROI within two to four years according to Deloitte's 2025 research. The figures are not comparable because they describe different types of investment.

How long does it take to see ROI from an AI project? For focused projects with a documented baseline and a serious adoption plan, the first measurable results typically appear within the first quarter of deployment. Full payback on implementation cost is usually reached within six to twelve months. Measuring at week four is too early. The workflows have not bedded in and the team is still adapting.

What percentage of AI projects fail to deliver ROI? Multiple studies suggest that between 70 and 85 percent of AI projects do not achieve their intended business outcomes. The primary cause is not technical failure. It is lack of commercial clarity before the project starts, no defined baseline, and underinvestment in adoption.

How do I measure ROI on an AI project? Start with a documented baseline before anything is built: hours spent on the process being automated, fully-loaded hourly cost, and current output metrics. Agree the value metrics before deployment. Measure at three months and twelve months, not at four weeks. Include all costs in the denominator: implementation, tooling, internal time, and transition productivity dip.

Should I include soft benefits in my AI ROI calculation? Soft benefits like improved morale, better data quality, or faster decision-making are real but difficult to prove to a sceptical stakeholder. Include them as supporting context, not as primary ROI drivers. The hard numbers, hours saved, costs reduced, revenue generated, are what hold up under scrutiny.

What is the biggest mistake in measuring AI ROI? Not establishing a baseline before the transformation starts. Without a clear before state, you cannot prove the after state. This is the most common reason a project that delivered genuine value fails to be recognised as having done so, which kills future investment.

Jessica Thomas is a Commercial AI Consultant who builds the commercial thesis before any technology decision is made. If you are planning an AI project and want to get the ROI model right, [book a call](https://calendly.com/jess-jessicathomas/30min).