Most AI ROI promises consist of a single, impressive number with no visible derivation. The framework below makes the derivation visible: grounded in established sources like Forrester and McKinsey, complemented by academic research on the realistic time horizon – and worked through on two real Beyond Prompt projects, not just claimed.
Why most AI ROI promises are too smooth
Talk to AI vendors or consultancies and you'll run into the same number almost every time: a single, impressive percentage with no visible derivation. How the number came about, which costs were included, and over what period it holds rarely gets explained. That's not an accident – a range with a derivation sells worse than a round headline. For an investment decision, it's not much use either.
Four sources, one shared logic
Forrester's Total Economic Impact (TEI) methodology has been the industry standard for technology investment decisions for over twenty years. It consistently breaks an investment down into four dimensions – cost, benefit, risk and flexibility – instead of looking at cost and benefit in isolation.
McKinsey adds an important warning: pure efficiency gains are often competed away and tend to benefit customers more than the investing company. By McKinsey's own estimate, 30 to 40 percent of the potential is lost when processes and incentives aren't redesigned alongside the technology. Real, lasting value only emerges when an initiative stays traceable all the way from capability to financial impact, instead of jumping straight to the desired number.
Academic research adds a third, often overlooked insight: economists around Erik Brynjolfsson (NBER Working Paper 25148, “The Productivity J-Curve”) show that new general-purpose technologies like AI first require complementary, often intangible investments – redesigned workflows, retraining. Short term, that means flat or even declining measured impact before the investment pays off. Important: this applies to introducing a genuinely new capability – not automatically to automating an already well-understood process.
The fourth building block is an established standard in IT business cases: separating hard benefits (directly expressible in euros, like time saved or error costs) from soft benefits (real, but only translatable into money with a traceable rationale, like data quality or decision speed). Counting only hard benefits systematically makes worthwhile initiatives look worse than they are.
The central finding: not every project behaves the same
The biggest mistake in many AI ROI calculations is treating every initiative the same way. The two case studies below show: when an already well-understood process is automated, or an existing system is replaced with a fitting solution, the investment pays back fast and reliably – without the ramp-up time that's realistic for genuinely new AI capabilities. That's exactly why this framework consistently distinguishes between the two cases, instead of applying one blanket AI time horizon.
Theory is one thing – here's the test against two completed Beyond Prompt projects, with real numbers instead of a sample calculation.
Is the ROI framework an offer or a binding commitment?+
No. The framework provides a traceable derivation and a reference range, not a quote. The actual scope for your initiative is set in a free first conversation based on your specific numbers.
Why do the two examples pay back so fast when AI projects are supposed to take years?+
Because both examples automate an already well-understood process or replace an existing system – that doesn't require the new organizational capability the academic research above describes for genuinely new AI capabilities. For an initiative that introduces a fundamentally new capability, the framework would apply a correspondingly longer, more realistic time horizon.
Do you just fold soft benefits like data quality into the number?+
No. Soft benefits are named, but only translated into euros when a traceable rationale can be established for them. Without that rationale, they stay explicitly named as a factor instead of inventing a number that can't be cleanly justified.
Can I run the numbers for my own initiative?+
Right now this framework is described as a methodology, not yet an interactive calculator. A tool for that is in preparation. Until then, the fastest way to get an assessment for your specific initiative is the AI Opportunity Scan or a free first conversation.
