Let’s cut to the chase. Everyone’s talking about AI, and everyone’s being asked to justify AI investments. But most ROI calculations I see are… optimistic, to put it mildly. They focus on potential, not reality. As someone who’s spent the last decade building and deploying AI systems – not just talking about them – I’m going to tell you how to actually measure the value of your AI projects. This isn’t about theoretical frameworks; it’s about what works, especially in challenging environments like the MENA region.
How rare is a measurable return? The numbers, with their caveats attached
Before measuring your own return, it is worth knowing how unusual a measurable one currently is. Two large executive studies converge. IBM's Institute for Business Value 2025 CEO Study, surveying 2,000 CEOs across 33 countries, reports that only 25% of AI initiatives have delivered the ROI expected of them and only 16% have scaled across the enterprise. Deloitte's global study of 1,854 senior executives across 14 countries finds that most respondents reached satisfactory ROI on a typical use case within two to four years — against the seven-to-twelve-month payback normally expected of a technology investment.
One widely repeated line from that study deserves to be retired rather than repeated, including by me. Deloitte reports that "only around one in five surveyed organizations qualify as true AI ROI Leaders" — but read the methodology and the category is defined into existence: Deloitte scored respondents on financial return, revenue growth, cost savings and speed, then "grouped the top 20 percent of performers as AI ROI Leaders." One in five qualify because one in five is what the label means. It is a definition, not a discovery, and quoting it as evidence of scarcity is a small act of numerical theatre.
The figures from the same study that do carry information are the unglamorous ones: only six per cent of respondents reported payback in under a year, and even among the most successful projects just 13 per cent saw returns within twelve months. Those are measurements, not tautologies, and they say the same thing far more honestly.
Read those with their limits attached: both are self-reported executive surveys run by firms that sell AI consulting, neither is peer-reviewed, and Deloitte's sample covers Europe and the Middle East with no US or APAC respondents. For a non-vendor counterweight, the US Census Bureau's Business Trends and Outlook Survey — government data drawn from a panel of roughly 1.2 million businesses — puts actual AI use among US firms at 17–20% between December 2025 and May 2026, concentrated among large employers: around 37% of businesses with at least 250 employees. Adoption is real, but narrower than vendor surveys imply. One methodological trap worth knowing: Census widened its AI question in November 2025 to cover any business function, so figures before and after that change are not a comparable trend line.
The honest reading is not "AI does not pay." It is that a two-to-four-year horizon is the normal shape of a real return, that the organizations reporting returns skew heavily toward those large enough to measure properly, and that anyone promising clean payback in a single quarter is selling against the available evidence. Which makes how you measure the entire game.
Beyond Cost Reduction: The Opportunity Cost of Not Doing AI
The first mistake is framing AI ROI solely as cost reduction or efficiency gains. Yes, automating tasks with AI can save money. We saw this firsthand with a large logistics client in Saudi Arabia. They were drowning in manual invoice processing. We built an AI-powered system that reduced processing time by 70% and errors by 95%. That’s a clear win. But the real ROI wasn’t just the saved labor costs. It was the opportunity cost of what those employees could now focus on – strategic sourcing, building client relationships, and expanding into new markets.
In Lebanon, and across the MENA region, this is even more critical. Talent is scarce, and skilled labor is expensive. Freeing up your existing team to focus on higher-value activities isn’t just about saving money; it’s about unlocking growth potential that you simply couldn’t access before. Think about the impact on innovation, customer service, and market responsiveness. These are harder to quantify, but they are often far more significant than any direct cost savings.
The “Baseline Drift” Problem & Why A/B Testing is Your Friend
Here’s a common scenario: you implement an AI-powered recommendation engine on your e-commerce site. Sales go up. Great! But how much of that increase is actually due to the AI, and how much is due to seasonal trends, marketing campaigns, or just general market growth? This is what I call “baseline drift.” It’s the biggest challenge in measuring AI ROI.
The solution? Rigorous A/B testing. Don’t roll out the AI to everyone at once. Divide your users into control and test groups. The control group sees the existing experience, while the test group sees the AI-powered version. Measure key metrics – conversion rates, average order value, customer lifetime value – over a statistically significant period. This isolates the impact of the AI. We use this religiously at Webspot when deploying new AI features for our clients. It’s the only way to get a truly accurate read.
I often find companies skip this step because it requires discipline and a willingness to potentially show that an AI project isn’t performing as expected. But that’s valuable information! It allows you to iterate, refine, and ultimately deliver a solution that does drive value.
Quantifying Intangible Benefits: The “Proxy Metric” Approach
Some AI benefits are inherently difficult to quantify. Improved customer satisfaction, enhanced brand reputation, reduced risk – these are all valuable, but they don’t easily translate into numbers on a spreadsheet. This is where “proxy metrics” come in.
For example, let’s say you’ve implemented an AI-powered chatbot to handle customer support inquiries. You can’t directly measure “improved customer satisfaction,” but you can measure:
- Resolution Time: How quickly are issues resolved?
- First Contact Resolution Rate: How often are issues resolved on the first interaction?
- Customer Effort Score (CES): How much effort does it take for customers to get their issues resolved?
- Sentiment Analysis of Chat Transcripts: Use AI to analyze the tone of customer interactions.
These metrics act as proxies for customer satisfaction. If they improve after implementing the chatbot, you can reasonably infer that customer satisfaction has also improved. The key is to choose proxies that are directly correlated with the intangible benefit you’re trying to measure.
The Importance of Long-Term Monitoring & Model Decay
AI isn’t a “set it and forget it” technology. Models degrade over time. Data distributions change. The world evolves. What worked brilliantly six months ago might be producing suboptimal results today. This is known as “model decay.”
Your ROI calculation shouldn’t be a one-time exercise. You need to continuously monitor the performance of your AI systems and retrain them as needed. This requires establishing clear monitoring dashboards and setting up automated alerts. We build these into all our deployments. It’s not glamorous work, but it’s essential for maintaining ROI.
In the MENA region, this is particularly important due to the rapid pace of change and the unique cultural nuances that can impact data patterns. A model trained on data from one country might not generalize well to another.
Don't Forget the Cost of Failure: Risk Mitigation & Ethical Considerations
ROI isn’t just about potential gains; it’s also about mitigating potential losses. What’s the cost of an AI system making a wrong decision? What’s the reputational damage if your AI exhibits bias? These are critical considerations, especially in regulated industries like finance and healthcare.
Since this article was first published that cost has acquired a calendar. The EU AI Act's implementation timeline put the bulk of the regulation into application on 2 August 2026, and the European Commission confirms that the obligations for high-risk use cases in the sensitive areas listed in Annex III "have been extended to 2 December 2027" under the AI Omnibus simplification package, with AI embedded in regulated products running to 2 August 2028. For a MENA business, the relevant question is not whether Lebanon regulates AI — it does not — but whether you sell into, or process data for, anyone who is covered. If you do, compliance engineering is a dated line in the ROI model with a deadline attached, not a vague future risk.
Factor in the cost of risk mitigation – things like data quality checks, bias detection algorithms, and human oversight – into your ROI calculation. Ignoring these factors is short-sighted and can lead to significant financial and reputational consequences.
I often tell my clients:
“AI isn’t about replacing humans; it’s about augmenting them. And the most important augmentation is in judgment – ensuring that AI-driven decisions are ethical, fair, and aligned with your values.”
Practical Takeaways: Start Measuring Today
- Define Clear Objectives: What specific business problem are you trying to solve with AI?
- Establish a Baseline: Measure your key metrics before implementing AI.
- A/B Test Everything: Isolate the impact of AI with rigorous testing.
- Use Proxy Metrics: Quantify intangible benefits with measurable indicators.
- Monitor Continuously: Track performance and retrain models regularly.
- Factor in Risk: Account for the cost of failure and ethical considerations.
Measuring AI ROI is challenging, but it’s not impossible. It requires a disciplined, data-driven approach and a willingness to look beyond the hype. If you’re serious about AI transformation, you need to start measuring the real value of your investments today. If you're looking for a deeper dive, check out my book, Applied AI for Future Ready Organizations. And if you’re facing these challenges in the MENA region, feel free to reach out – I’m always happy to share my experience. You can find more about my work at jonahtebaa.com.
