Before any workflow we build goes into production, we put a simple financial model in front of the client. It answers one question: how long until this pays for itself? If the answer is more than 12 months, we have a conversation about scope. If the answer is under six months, we start planning deployment.
This is not unusual logic. It is the same math a business owner applies to a new hire, a piece of equipment, or a software subscription. What surprises people is how rarely AI vendors apply it before they start building.
The Core Formula
The foundational ROI calculation for AI workflow automation has three inputs:
Annual Value = Hours saved per week × Blended hourly rate × 50 working weeks
Payback period (in months) = Build cost ÷ (Annual value ÷ 12)
That is it. Two lines of arithmetic. Everything else is a refinement of those two lines.
The "hours saved per week" figure comes from process mapping. Before we propose any solution, we walk the workflow with the team, step by step, and time each phase. We are looking for the steps that are repetitive, rules-based, and high-volume, because those are the ones where AI creates genuine time savings. Judgment calls, relationship decisions, and edge cases stay with the humans.
The "blended hourly rate" is not the loaded labor cost from the P&L. For this model we use a conservative figure that reflects what the employee's time is actually worth in opportunity terms. A $60,000 per year employee doing manual document intake is not costing the company $29 per hour in direct payroll. They are costing the company the value of the higher-value work they are not doing.
A Real Example: Invoice Processing
One of our clients ran a four-person accounts payable team. Each person spent roughly three hours per day on invoice receipt, data extraction, matching against purchase orders, and routing for approval. That is 12 hours per day, or roughly 60 hours per week across the team, on a process that was almost entirely rules-based.
After process mapping, we identified that 22 of those 60 hours per week could be handled by an AI workflow. The remaining 38 hours covered exception handling, vendor disputes, and approval decisions that genuinely required human judgment.
At a $40 blended hourly rate, 22 hours per week over 50 working weeks equals $44,000 per year in recovered capacity. The build cost for the workflow was $9,500. Payback period: just over two months.
That math made the decision easy. More importantly, it gave the client something to show their leadership team that was not an abstract argument about digital transformation. It was a spreadsheet with a payback date circled.
What the Formula Does Not Capture (and What to Do About It)
The basic formula measures time. But AI workflows often generate value in ways that do not show up in hours-saved calculations. Here are three categories worth adding to the model when they apply.
Error reduction value. Manual data entry errors have real costs: rework time, customer impact, compliance exposure. If your team is re-keying data from documents into systems, an AI extraction layer can eliminate most of those errors. The value is not always easy to quantify precisely, but even a rough estimate belongs in the analysis.
Throughput capacity. Some workflows are not bottlenecked by cost but by speed. A team that can process 200 applications per week, regardless of headcount, limits business growth. If AI increases throughput by 40%, that capacity has revenue implications that dwarf the labor savings calculation.
Talent retention and morale. This one is harder to put a number on, but worth noting. High-volume, low-judgment work is draining. Teams that spend their days copying data between systems do not stay. The cost of replacing a skilled employee, including recruiting, onboarding, and the productivity gap, often runs six to nine months of salary. Removing the worst parts of a job has measurable retention value.
When the Math Does Not Work
We have walked through this calculation with prospects and concluded that the project should not move forward. That is a feature of the model, not a failure.
If a workflow affects only two hours per week and the build cost is $15,000, the payback period is years, not months. In that case, the right answer is usually to document the process well and revisit when the volume grows.
AI is not always the answer. It is the answer when the math works, the workflow is stable enough to automate reliably, and the team has the governance structure to maintain what gets built. When those three conditions are met, the ROI is usually compelling. When they are not, spending on AI creates technical debt without a return.
Starting the Calculation
If you want to run this analysis on your own workflows before we talk, start with one process. Pick the one your team complains about most. Time each step. Identify which steps require a human decision versus which steps apply a rule to information. Multiply the rule-based hours by your blended rate. That number is your ceiling. The build cost should be well below it for the project to make sense.
We are happy to review that analysis with you and tell you honestly whether the numbers support moving forward. If they do, we know exactly how to build it. If they do not, we will tell you that too.