How to Calculate the ROI of an AI Consulting Engagement

Vague promises about AI productivity gains will not get budget approved. Here is a concrete framework for calculating and justifying AI consulting ROI before you start.

The hardest part of getting an AI project approved is not convincing executives that AI works — it is quantifying what it will actually deliver for your specific situation. Generic claims about productivity gains do not hold up in a budget committee. You need numbers that are tied to your operations.

Here is the framework I walk clients through before we scope any engagement.

The cost side: what you are actually paying for

Most ROI calculations undercount costs. A thorough accounting includes: consulting fees (the largest line item), infrastructure costs (model API calls, vector databases, compute), internal staff time during the project (your team's hours are not free), change management (training, process redesign, communication), and ongoing maintenance after launch.

For a mid-complexity enterprise AI project, the total first-year cost — consulting plus infrastructure plus internal time — typically runs between $150,000 and $500,000. Ongoing costs in year two are 20-30% of that, mostly infrastructure and support.

The value side: three categories of return

I organize the value from AI projects into three categories, because they have different calculation methods and different credibility levels with finance teams.

1. 1Labor cost reduction. The most concrete category. Calculate the hours currently spent on the target process, multiply by the fully-loaded cost per hour, apply the percentage reduction the AI system achieves, and you have a hard dollar number. For document processing automation, a 70% time reduction on a 10-person team spending 40% of their time on the process is a large, defensible number.
2. 2Cost avoidance. Harder to count but often larger. Fraud that is not committed, compliance fines that are not incurred, contracts that do not have unfavorable terms because someone missed a clause. These require risk-adjusted estimates but are legitimate value.
3. 3Revenue impact. The most speculative category. Faster customer onboarding, better support experiences, more accurate recommendations — all can drive revenue, but the causal chain is longer and harder to defend in a spreadsheet. I recommend modeling this conservatively and leading with labor cost reduction instead.

A worked example

Say you want to automate invoice processing. Currently, 8 people spend 60% of their time processing 5,000 invoices per month. Their average fully-loaded cost is $80,000 per year. The AI system will handle 80% of invoices automatically, reducing team time on this process by 80%.

Annual labor saving: 8 people x $80,000 x 60% effort x 80% automation = $307,200 per year. Project cost: $250,000. Year one ROI: 23%. Year two ROI (with $75,000 operating cost): 310%. Payback period: 9.8 months.

How to build the internal business case

Three things make a business case credible: the numbers come from your own data (not vendor claims), you include a realistic downside scenario, and the business owner of the affected process has co-signed the assumptions.

• →Pull actual time data from your team — time studies or manager estimates both work
• →Model three scenarios: 50%, 70%, and 85% automation rates
• →Include one-time costs and ongoing costs separately
• →Get the process owner to validate the assumptions in writing

The business case also needs to answer: what is the risk if this does not work? For most projects, the answer is that you spent money and time but the current process still runs. That is a manageable downside. Make sure finance understands the downside is bounded.