Most people know what ROI means, but it’s harder to calculate for AI projects. The numbers are less predictable than with traditional platforms because many AI projects never reach stable production. IDC says only about 44% of custom AI apps and 53% of third-party AI apps make it from proof of concept to production. That’s why it’s important to look at ROI through a risk lens, not just cost versus benefit.
One useful approach is to use a risk-adjusted formula:
AI ROI = (AI Business Value Income / (Initial Investment + Annual Costs)) × Success Probability
where,
>AI Business Value Income (over N years)
Consider a 2 to 3 year period and include both direct and indirect value:
Direct: time saved, fewer tickets, higher conversion, lower fraud.
Indirect: improved customer or employee experience and quicker decisions. For these, use measurable stand-ins like CSAT, churn, time to resolution, or hours saved, and estimate conservatively.
>Initial Investment This covers more than just building the app. For AI, you often pay upfront for data discovery, security and privacy work, evaluation design, integration, and change management or training. Many proofs of concept fail later because the initial investment was too low.
>Annual Costs AI projects are ongoing. Each year, you’ll have costs for cloud use (like tokens, inference, or embeddings), search and storage, monitoring, human review, updating content or models, and governance or audits. Costs may change as usage increases, models improve, and compliance rules tighten.
>Success Probability (0 to 1) Most spreadsheets skip this part. In typical IT projects like ERP upgrades, data platforms, or CRM rollouts, teams often assume a high chance of success, usually between 0.9 and 1.0, because if you deliver the project and users adopt it, the results are usually steady.
AI projects are different. Success probability shows your actual chance of launching, getting people to use it, and keeping quality steady over time. It depends on data readiness, project ownership, ongoing maintenance or retraining, model drift, and teamwork across business, data, and IT. Assuming a perfect score of 1.0 is unrealistic. In most cases, a range of 0.6 to 0.8 is more realistic.
Example from the Microsoft + Azure ecosystem
Picture Sara, an HR specialist, using an internal Support Copilot for IT and HR. The Copilot is built with Azure OpenAI Service and Azure AI Search, using retrieval over SharePoint policies and FAQs, and is available in Microsoft Teams. Whenever Sara needs to check a policy, update onboarding processes, or resolve common employee questions, she chats with Copilot to get instant, relevant answers within her daily workflow. Around point: 50,000 internal tickets per year, each taking 15 minutes on average, with a total cost of €35 per hour. Each year, that adds up to 50,000 tickets times 15 minutes, which is 750,000 minutes or 12,500 hours, costing €437,500.
If you deflect 20% of tickets, that’s 10,000 tickets avoided, saving 2,500 hours and €87,500.
10% faster means 40,000 tickets save 1.5 minutes each, totaling 1,000 hours and €35,000 saved. Total yearly value is about €182,500, or €547,500 over 3 years. Intial investment, including building, integration, security, and rollout, is €150,000.
Annual Azure and operations costs are €60,000 per year, totaling €180,000 over three years. Total cost is €330,000.
Without factoring in risk, the ROI ratio is 547,500 divided by 330,000, which equals 1.66. If you apply a success probability of 0.7, based on good sponsorship and a well-maintained knowledge base: The risk-normalised AI ROI is 1.66 times X 0.7, which equals 1.16.
What does 1.16 means? For every €1 invested, you can expect about €1.16 of measurable value in that timeframe. If adoption climbs or deflection improves, the return often accelerates. The key is to measure continuously and treat the probability of success as a KPI, not as an afterthought.
In current AI programs, how many time the ROI been calculated?
Comments
Post a Comment