# How to measure AI ROI: a calculation method > How to calculate the ROI of AI in your company: a measured baseline, gains actually reallocated, full costs, indicators by function and pitfalls to avoid. URL: https://aiorigin.io/en/ressources/insights/roi-agents-ia-pme-methode-calcul-2026 Insights Method # Measuring the ROI of AI in your company: the method The ROI of AI isn’t found in a market study; it is measured in your own company. The method comes down to three elements: a baseline, verified gains and full costs. UPDATED · OCTOBER 16, 2026 READ · 5 MIN Contents - How do you measure the ROI of AI in a company? - Why time saved often stays invisible - Which indicators to track, function by function - How to go about it: the six-step method - An example of the reasoning: supplier invoices - Mistakes that skew the calculation - How AI Origin supports you IN SHORT To measure the ROI of AI, compare an indicator measured before and after, use case by use case: turnaround time, volume processed, rework rate. The gains you count are the ones actually reallocated; costs include tools, integration, training and review time. ROI is then calculated as net gains divided by costs. ## How do you measure the ROI of AI in a company? You measure the ROI of AI by use case, not for “AI” in general. For each case, you need three elements: - A measured baseline: how long the task takes today, how many cases are processed, how many are reworked or corrected. - The gains observed after go-live, on the same indicators, and how the time freed up is used. - Full costs: tools, integration, training, support, and time spent reviewing what AI prepares. ROI is then calculated the classic way: net gains, meaning gains minus costs, divided by costs. The formula is nothing new. What makes the difference is the quality of the measurement, especially the baseline. Be wary of average gain figures taken from general studies. They describe other companies, other tasks and other tools. The only figure that matters for your decision is the one you measure in your own company. ## Why time saved often stays invisible The first gain from AI is usually time: a summary prepared, a draft reply, a faster search. But that time only becomes a result if it is reallocated. According to BCG, 42% of regular frontline users save at least one day a week, and 66% receive little or no guidance on how to reinvest that time. In other words, without a decision on how to use the time freed up, the gain dissipates: it shows up neither in turnaround times, nor in volumes, nor in the accounts. The ROI calculation must therefore distinguish two things: - The time saved on the task, which is easy to observe. - What the team does with it: more cases processed, shorter turnaround times, work that used to be set aside, such as tenders no one had time to study. Only the second translates into value for the company. It is also the one leadership needs to organize. ## Which indicators to track, function by function Choose an indicator the team already tracks, rather than one created for the occasion. A few examples: FunctionBaseline indicatorWhat AI can move Customer serviceTime to first response, reopened requestsPrepared responses, summarized history SalesVisit reports logged in the CRM, follow-up timeReports dictated, then filed FinanceClosing time, invoices with anomaliesMatching prepared, discrepancies flagged PurchasingOffers compared, request-for-quote cycle timeComparison grids prepared Executive managementTime to produce a summaryMonitoring and summaries prepared OperationsIncomplete files, processing timeCompleteness checks Pair each volume or time indicator with a quality indicator: draft rework rate, errors found during review, complaints. A time saving that comes at the expense of quality is not a gain. ## How to go about it: the six-step method - Define the scope: a specific task, a team, a main indicator and a quality indicator. - Measure the baseline over a representative period, before anything goes live. Take volumes, durations and rework from your tools whenever possible, rather than relying on self-reporting. - List full costs: licenses or tool usage, connector development, training time, review time, maintenance. - Go live with a volunteer team, keeping systematic review. Let the learning period pass before measuring. - Measure again on the same indicators over the same duration, and note how the time freed up is used. - Decide: extend, adjust or stop. Document the calculation so you can repeat it when extending to other teams. When possible, compare with a similar team that does not yet use the tool. This neutralizes seasonal or workload effects. Finally, present each result to the leadership team in the same format: scope, baseline, gains, costs, decision. Use cases become comparable, and the choice of the next ones rests on facts rather than on one team’s enthusiasm. ## An example of the reasoning: supplier invoices Take supplier invoice control in a finance department. The reasoning is built in four stages, without any invented assumption. The baseline. Over a representative period, you pull from the ERP the number of invoices received, the time between receipt and approval, and the number of invoices blocked because they don’t match the purchase order. You also ask the team how it checks today: systematically or by sampling. The change. AI matches each invoice with the purchase order and the goods receipt, then flags discrepancies. The accountant only looks at flagged invoices, and makes the call. This type of use is described on our finance page. The gains. They show up in three indicators: approval time, the share of invoices checked, and anomalies detected before payment. The time freed up goes to a task decided in advance, for example analyzing recurring discrepancies with certain suppliers. The costs. The connector between AI and the ERP, configuring the matching rules, training the team, time spent reviewing flagged invoices. ROI is then calculated with your own values. And if it is low, the calculation shows where to act: integration, instructions, or the choice of use case. ## Mistakes that skew the calculation Counting self-reported time. Off-the-cuff estimates overstate or understate the gain. Prefer data from your tools: timestamps, volumes, turnaround times. Forgetting review time. AI prepares, a person decides. Review is part of the cost, and it decreases as instructions are refined: measure it too. Forgetting integration costs. An assistant that forces people to copy and paste between several applications costs time with every use. Connecting AI to your tools often changes the result. Adding up overlapping gains. If two use cases cover the same process, the total gain is not the sum of the two. Measuring people rather than a process. ROI is about a task and an organization. Individual monitoring holds back adoption and says nothing about the value created. Ignoring what can’t be counted in hours: a reduced risk, a customer deadline met, an error avoided on an invoice, know-how that no longer depends on a single person. Note them alongside the calculation, with the same rigor. ## How AI Origin supports you Measurement starts at the operations mapping step of our method: each use case is ranked by gain and difficulty, with the indicator that will be used to track it. We then help your teams measure their baseline and the result, on your own data. Every gain is measured in your company, starting from your baseline. For cases that require connecting AI to your software, the calculation includes the cost of integration, so the decision is fully informed. See also our guide on moving AI pilots into production. YOUR QUESTIONS ## Frequently asked questions. What is the ROI formula for an AI project? The same as for any investment: net gains, meaning gains minus costs, divided by costs. The hard part is measuring gains correctly, starting from a baseline taken before go-live. Which costs should be included when calculating AI ROI? Tools, development of connectors or business tools, training, support, maintenance and time spent reviewing what AI prepares. Leaving out review or integration skews the result. How long should you wait before measuring AI ROI? Until after the learning period, once usage has stabilized. Measure too early and you underestimate the gain; too late, and no one remembers what the baseline was. How do you measure a gain that isn’t time? With a suitable indicator: errors avoided, complaints, deadlines met, files complete the first time. These gains are presented alongside the financial calculation, with their own measurement. Can you trust the productivity figures published in studies? They show a trend, not your result. Your decision should rest on a measurement made in your company, on your tasks and with your tools. SOURCES ## Texts and studies cited in this article. Every figure on this site is sourced. See all our sources. BCG, AI at Work: Strategy Matters More Than Tools, June 3, 2026. bcg.com READ NEXT ## To go further. - Method · Four steps in a loop - GUIDEWhy AI pilots fail to reach production - Use cases · Proven, delivered in a few weeks LET’S TALK ABOUT YOUR PROJECT ## Let’s talk about your operations. Thirty minutes to review current usage, and tell you where we would start. Your contact The AI Origin team that runs the projects You talk directly with the people who will run your project, from scoping to go-live. No specific project yet? Start with a hands-on session: everyone brings a real problem and leaves with the AI agent that solves it. See step 3, Bring your teams on board Thank you, we have received your request. We will get back to you to schedule a time. ### Write to us in two lines. Your company, the teams involved, what you are seeing. We will get back to you to schedule a time. Write to contact@aiorigin.io Or book a time slot Or write to us: contact@aiorigin.io