73% of AI projects have no formalized ROI measurement — here's how to change that
Measuring AI ROI is the main blind spot of SMBs and mid-market companies. It is estimated that only about a quarter of companies have set up a structured measurement framework for their AI investments. Yet, based on our observations, AI projects that are properly measured are significantly more likely to be scaled and refunded. The good news: with 4 to 6 well-chosen KPIs and a simple calculation method, you can prove the value of your AI project in less than 90 days.
The problem
Most SMBs launch AI projects with an intuitive conviction that "it's going to save us time" or "it's going to reduce errors." But when the time comes to justify the investment to the management committee or to decide whether to continue, scale or stop the project, the absence of hard data becomes a major handicap.
The concrete difficulties companies encounter:
- No baseline: it's impossible to measure an "after" without having measured the "before." Yet 70% of AI projects start without an initial measurement of the targeted processes.
- Poorly chosen KPIs: measuring model accuracy (a technical metric) instead of business gain (hours saved, cost per processing) doesn't speak to decision-makers.
- Diffuse benefits: AI often generates indirect gains (better customer satisfaction, reduced turnover) that are difficult to attribute and quantify.
- Poorly tracked costs: API, infrastructure and maintenance costs are not consolidated, making the net ROI calculation impossible.
The result: "zombie" AI projects that keep running without anyone knowing whether they create or destroy value. Several studies suggest that a significant share of active enterprise AI projects do not generate positive net value when all costs are factored in.
The AI solution
Our approach is based on a three-level measurement framework, adapted to each phase of the project and each audience (management, business, technical). Here are the three pillars of AI ROI measurement.
Operational KPIs (measured from D+30)
These are the most immediate and easiest indicators to measure: time saved per task, number of tasks automated per day, error rate before/after, processing time. They speak to business teams and quickly prove the value of the project. Example: a lead qualification process that drops from 45 min to 12 min per case. Track these KPIs with our AI automation solutions.
Financial KPIs (measured at M+3)
Translation of operational gains into euros: cost per processing before/after, projected annual savings, customer acquisition cost (where applicable), additional revenue generated. This is the language of the management committee. Formula: ROI = (Annual gains – Total project cost) / Total cost × 100. Check out our case studies for quantified examples.
Strategic KPIs (measured at M+6)
Impact on the company's business objectives: customer satisfaction (NPS), retention rate, time-to-market of new offerings, ability to absorb growth without hiring. These KPIs justify extending the AI program to other processes and refunding the budget.
Implementation
Here is the three-step method to set up an AI ROI measurement framework in your SMB or mid-market company, without complex tooling or advanced statistical skills.
Establish the baseline before deployment
Before launching your AI project, measure the key indicators of the targeted process over 2 to 4 weeks: average time per task (time 20 to 30 occurrences), unit cost (time × fully loaded hourly cost), error rate (number of errors / number of processings), volume of tasks per week. Document this data in a simple spreadsheet. This baseline is your absolute reference for calculating gains.
Define target KPIs and success thresholds
For each AI project, define 4 to 6 KPIs with explicit success thresholds. Example for a customer service chatbot: automatic resolution rate > 40% (minimum threshold), average response time < 30 sec, user satisfaction > 4/5, cost per ticket reduced by 50%. Validate these thresholds with the business sponsor and management. They become your Go/No-Go criteria for going into production.
Build a tracking dashboard
Create a simple dashboard (Google Sheets, Notion or Power BI) with three views: operational view (real-time KPIs for the business team), financial view (cumulative monthly ROI for management), technical view (infrastructure costs, latency, error rate for the IT team). Automate data collection as much as possible via the APIs of the tools used. Schedule a monthly review of the dashboard with stakeholders.
Results
Frequently asked questions
How do you calculate the ROI of an AI chatbot for customer service?
Measure the cost per ticket before AI (agent time × hourly cost) and after AI (tickets resolved automatically × API cost + escalated tickets × agent cost). An AI chatbot resolves on average 40 to 60% of level-1 tickets, reducing the cost per ticket from 8-12 euros to 2-4 euros. The ROI is calculated as follows: (annual savings – project cost) / project cost × 100.
Which KPIs should you track for an AI automation project?
The four essential KPIs are: time saved per task (in hours/week), the error rate before/after (as a percentage), the cost per processing (in euros), and the user adoption rate (as a percentage of active employees). Add a user satisfaction KPI (internal NPS) to measure acceptance.
How long does it take before you can measure the ROI of an AI project?
The first indicators are measurable as early as 30 days after deployment (time saved, volume of automated tasks). A reliable financial ROI requires 3 to 6 months of hindsight to factor in ramp-up, process stabilization and indirect effects (customer satisfaction, retention).
How do you measure the intangible benefits of AI?
Intangible benefits (employee satisfaction, brand image, agility) are measured through quantifiable proxies: internal NPS survey before/after, turnover rate of the teams concerned, customer response time, number of innovations or new offerings launched thanks to the time freed up.
For technical profiles
Here is a comparison of AI ROI measurement approaches by the company's level of maturity:
| Criterion | Spreadsheet + manual tracking | BI dashboard (Power BI / Metabase) | MLOps platform (MLflow / Weights & Biases) |
|---|---|---|---|
| Setup cost | 0 euros | 500 – 2,000 euros | 2,000 – 10,000 euros |
| Setup time | 1-2 days | 1-2 weeks | 2-4 weeks |
| Collection automation | Manual | Automatic via connectors | Automatic + alerts |
| Business KPIs | Yes (manual entry) | Yes (real time) | Partial (technical focus) |
| Technical KPIs | No | Basic | Complete (drift, latency, costs) |
| Sharing with management | PDF export | Shared dashboard | Technical interface |
| Suitable for | 1-2 AI projects | 3-5 AI projects | 5+ AI projects in production |
Our recommendation: start with a structured spreadsheet for your first 1-2 projects, then migrate to a BI dashboard once you reach 3 projects or more. The MLOps platform is only justified if you have a dedicated data team and more than 5 models in production.