What Is a Good ROI for a Business in 2027? How to Judge AI Returns

Introduction

When executives ask what a good ROI looks like, they often mean a harder question: before a budget review turns into an argument, how do we know whether an AI investment is working? The difficulty is that AI costs arrive in pieces, including license fees, cloud usage, integration work, and staff time, while the benefits tend to show up as vague productivity gains that were never quantified when the project started.

Why There Is No Single ROI Number

A common rule of thumb is that a project should pay for itself within a year. That is a reasonable starting point, but it breaks down quickly. A customer-facing assistant that shortens support handling time follows a different payback curve than an internal drafting tool, and compliance automation, which mainly prevents losses, follows yet another pattern. A single percentage hides these differences and pushes teams toward metrics that are easy to count but may not reflect real value.

Common Mistakes That Distort the Math

The first mistake is counting only the model price. Teams compare API fees against labor savings, declare victory, and overlook data preparation, security review, change management, and ongoing maintenance. The second mistake is treating saved hours as saved cash without checking whether that time was redirected to valuable work or simply absorbed by other tasks. The third mistake is measuring too early. When only a small share of the target group has adopted the tool, the numbers look weak, but the cause has nothing to do with the technology itself.

For a more complete list of cost categories, see our enterprise AI adoption cost guide, which breaks down the items that are often left out of the first budget.

How to Set a Defensible Threshold

Start with a baseline. Whether the unit is a support ticket, a contract, or a report, what does it currently cost to process one? Then define the expected change and the total cost of ownership over the next 12 to 24 months. A practical test is the payback period. If a project cannot reasonably recover its full cost, including internal staff time, within a window the finance team accepts, it needs a narrower scope or a different use case.

Treat the pilot as a hypothesis with a spending cap, not as a finished product. When reporting results, present a range rather than a single figure, and agree on the assumptions before the work begins. Teams that do this can explain their results even when the outcome disappoints. Teams that skip it usually end up arguing.

Where to Look When Returns Stall

When results fall behind, the cause is rarely the model itself. More often, data access is poor, ownership of the workflow is unclear, or the team never actually changed how it works. The patterns that stall AI return on investment for companies tend to recur across industries, so checking for them early can save months.

An Action Plan for This Quarter

1. Choose one process where you can measure the unit cost.

2. Record a baseline for at least four weeks.

3. Write down the full cost: licenses, integration, training, review time, and maintenance.

4. Have the finance lead confirm the payback target before the pilot begins.

5. Review at fixed checkpoints, and decide whether to stop or expand based on the numbers you agreed on in advance.

For most businesses, a good ROI means being able to prove the return within a single process and recover the full cost inside a payback period agreed on in advance. Broader claims are still forecasts, not results.