Forrester's ROI Model Won't Fit Your Team

Forrester's SMB Total Economic Impact study for Microsoft 365 Copilot projects attrition reductions of up to 20% and operating expenditure reductions of up to 0.85% over three years. Those numbers are real, they are sourced, and they are useless to a founder presenting to an investor next month.

The model was never designed for your situation

The TEI model is a three-year projection built on a composite firm that does not exist. Forrester constructs it from aggregated customer interviews, applies risk adjustments, and produces an NPV figure. That method is sound when you are a 500-person company running a multi-year AI program with a dedicated IT function. When you are running a 12-person team and need to show a return before your next funding conversation, the model answers a question you are not asking.

The instinct many founders have is to pull the headline numbers from the TEI study and present them as benchmarks. That instinct is wrong, and an investor who tracks unit economics will spot it immediately. Citing a composite-firm attrition figure as if it reflects your own payroll is not benchmarking. It is borrowing credibility from a methodology without applying the methodology.

What 90 days actually measures

The Microsoft Research randomized experiment across more than 6,000 workers at 56 firms found workers with Copilot access spent roughly 30 minutes less on email per week and completed documents 12% faster. Those measurements came from a controlled design, not self-report. The UK Department for Work and Pensions ran a separate three-month trial with over 3,500 staff and found daily time savings that it then used explicitly as an input to investment decisions.

Both of those studies produced results within a 90-day window. Neither required a three-year projection. The savings were visible in task-level behaviour, not in lagging indicators like revenue growth or headcount reduction.

The problem with a 90-day window is not that the savings are invisible. It is that only 40% of workers with Copilot access used it regularly over six months, according to the same Microsoft Research report. At day 90, you are measuring adoption variance, not steady-state productivity. Your ROI figure will be noisier than the TEI composite, and it will likely understate long-run value. You need to say that to your investor before they ask.

Building the calculation

Start with what you can measure on your own team. Take the number of employees actively using Copilot. Multiply their average hourly cost by the hours saved per week. The UK Government Digital Service trial across 20,000 civil servants found self-reported savings averaging 26 minutes per day among active users. The Microsoft Research figure of 30 minutes per week on email alone gives you a more conservative floor from a randomized design. Use the Microsoft Research figure if you want a number an investor is less likely to challenge.

Over 90 days, the labour cost recovery from time savings is your primary input. It is the only input you can verify on your own payroll within a single quarter.

Attrition cost avoidance is a secondary input, and it requires a caveat. Forrester's SMB study puts attrition reduction in the range of 11–20%. That range comes from a vendor-commissioned model, which Forrester's own "Beyond the Hype" post warns is structurally prone to overstating benefits when the commissioning party selects which customer data enters the model. You should not present 11–20% as your expected outcome. You should present it as the external benchmark range and then show your own voluntary turnover rate from the prior 12 months alongside whatever early retention signals you are tracking. The range gives the calculation a credible ceiling. Your own data gives it a floor.

Supply chain savings are real in the Forrester SMB model, which notes operating expenditure reductions of up to 0.85% with explicit references to supply chain efficiencies. Those savings do not appear in 90 days. They require renegotiation cycles, vendor data, and procurement decisions that take longer than a quarter to close. Include them in a 12-month projection if you have supplier activity that warrants it. Leave them out of the 90-day figure entirely.

The failure rate is the number you need to show

Across 200 real B2B AI deployments tracked between 2022 and 2025, the median breakeven was eight months, not 90 days. SMEs outperformed mid-market firms on ROI over 24 months. But 27% of deployments failed to reach breakeven within that window at all. The failure cases clustered around deployments where adoption was not tracked.

An investor looking at your 90-day calculation is not just evaluating the upside. They are evaluating whether you understand the failure mode. The 27% failure rate is not an argument against the 90-day model. It is an argument for showing the investor your usage data alongside your ROI figure. If you cannot show which employees are using Copilot, how often, and on which task types, your ROI number is a projection wearing the clothes of a measurement.

The TEI model Forrester builds for enterprises has the same problem at a different scale. It is also a projection. The difference is that a three-year enterprise model is expected to be a projection, and nobody at the investor meeting pretends otherwise. A 90-day calculation implies you measured something. Make sure you did.

What to put in front of an investor

Track hours saved per employee per week using task-level data, not surveys. Apply the Microsoft Research figure of 30 minutes per week on email as your conservative floor. Benchmark the attrition implication against Forrester's 11–20% range, but show your own retention data next to it. Leave supply chain savings out of the 90-day figure unless you have a procurement event inside the window. Show the usage rate alongside the ROI number.

The Forrester three-year model is not wrong. It is built for a different conversation. A founder who extracts its benchmarks without applying its method to their own team is not simplifying the model. They are removing the part that makes it credible.