AI ROI Calculator
Hard benefits in currency, soft benefits as measured deltas — the structure a finance team will expect. Five inputs, one initiative.
Pick one — the form below changes to match it.
Soft benefits
Tracked as a delta against baseline, not converted to currency and not included in the total above.
Where this number will be challenged
Where should we send it? You'll get a one-page business case you can put in front of finance.
Already submitted?
Used only to send your summary. See our privacy policy.
AI ROI is net annual benefit divided by total one-time cost, expressed as a percentage. Net annual benefit means total annual value minus annual running costs — the figure a finance business partner will actually accept. This calculator estimates it across six AI value drivers, keeping hard benefits (expressed in currency) separate from soft benefits (tracked as a measured delta), the structure a finance team expects a business case to use.
What are the six AI value drivers?
Three of these produce a figure in currency. Two are recorded as a measured delta against a baseline. One is measured in hours, and only becomes money once those hours change a cost line. Keeping them apart is what makes a business case survive review.
1Productivity & Automation Improvement
Employee time released by the initiative, in annual hours and full-time equivalents.
You need: annual hours on the process today, and a deflection rate you can defend.
2Cost Reduction
Annual bottom-line saving from the labour cost of the released time.
You need: fully-loaded employee cost — salary plus employer pension, NI and benefits, not salary alone.
3Revenue Growth
Incremental annual top-line revenue attributable to the initiative.
Watch the units. A conversion rate moving 3% to 4.5% is 1.5 percentage points, so the delta divides by 100. A metric already denominated in currency, such as average deal size, carries the uplift in the delta itself — multiplying it by a per-unit figure double counts.
4Compliance & Risk Mitigation
Annual value at risk reduced — expected-value arithmetic, the same method as any risk register.
You need: the cost of a single event, and a defensible view of probability before and after.
5Customer Experience Improvement
The change in NPS, CSAT or CES against a stated baseline.
Not converted to currency. Turning an NPS point into pounds requires assumptions about customer economics that are specific to one organisation and rarely survive a finance review. Recording it as a delta keeps it measurable without inflating the financial case.
6Employee Experience & Retention
The change in eNPS, voluntary turnover, or time spent on low-value work.
You need: a baseline you actually measured. If you have never run an eNPS survey, say so rather than estimating one.
How do you calculate AI ROI?
ROI is net annual benefit divided by total one-time cost, expressed as a percentage.
How do you calculate payback period for an AI project?
Payback period is total one-time cost divided by net annual benefit, multiplied by twelve, to give the answer in months.
The common error is paying back the one-time cost against gross annual value rather than net, or folding running costs into the denominator alongside the capital cost. Both flatter payback, and both are the first thing a finance business partner will unpick.
Why the forecast doesn't survive contact with reality
Most AI business cases get approved. Process owners and IT sign off, not finance, and the finance director prospects we talk to say plainly that this isn't theirs to gate. The exposure sits later. A case gets approved, the initiative ships, and nobody goes back to check the forecast against what it actually delivered.
The forecast counts hours saved as money saved, and no one checks afterwards whether a cost line actually moved. Headcount, overtime and agency spend can sit exactly where they were and the case still reads as delivered. As CFO.com puts it: “if a CFO cannot name what gets smaller when AI is added, they do not yet have ROI — they have a cost with a narrative.”
An estimate from one meeting becomes the baseline in the approved case, and the real starting point never gets measured. Once that happens there is nothing left to check the forecast against. Two weeks of real data would have settled it either way.
The deflection rate gets approved as a target and stays a target. Nobody goes back and compares it to what the system actually handled once it was live, which makes it the most sensitive number in the model and the least likely to be revisited.
The approved figure covers the licence. Integration, change management, training and the internal time to run the project turn up later, once the initiative is scaling, and by then the case on file no longer matches what it costs. Scaling a pilot to production typically runs three to eight times the pilot budget (industry guidance, 2026).
A monetised customer or employee experience delta gets folded into the reported result to make the delivered number look closer to the forecast. The two were never the same kind of number. Folding them together hides whether the financial case actually landed.
Every model has a point where it stops working, and without a stated one nobody is ever required to notice it was reached. The initiative just keeps running, unexamined, indefinitely.
What is a deflection rate?
Deflection rate is the share of process volume an AI or automation initiative is expected to handle or eliminate without a person. A deflection rate of 80% means one item in five still needs manual involvement. It should be stated by the person building the business case as an explicit target, not inferred by a scoring model, because they will have to defend it.
Typical ranges, as a starting point rather than a target — SilkFlo working assumption, not an external benchmark: retail and consumer processes 40–60%, financial services 25–40%, public sector 20–35%.
Glossary
- Deflection rate
- The share of process volume an initiative is expected to handle or eliminate. Stated by the business, not inferred by a scoring model.
- As-is hours
- Total annual hours a process consumes today, including time spent correcting and checking the work.
- Fully-loaded cost
- Salary plus employer pension, National Insurance and benefits. Using salary alone understates the saving — a common rule of thumb puts it at roughly 20–30%, though the real figure is organisation-specific.
- Value at risk reduced
- Cost of a single risk event multiplied by the reduction in its annual probability. Standard expected-value arithmetic.
- Hard benefit
- A benefit expressed in currency and included in the financial total.
- Soft benefit
- A benefit tracked as a measured delta against a baseline, reported alongside the financial total rather than inside it.
Frequently asked questions
An AI Value Realization Platform is the independent financial governance layer that tracks an AI or automation initiative from business case through deployment to post-implementation audit against what it actually delivered, closing the gap between the number a project promised and the number it produced. SilkFlo is built as this layer; this calculator uses the same six-driver KPI structure SilkFlo runs across a full portfolio, not just one initiative.
Productivity & Automation Improvement, Cost Reduction, Revenue Growth, Compliance & Risk Mitigation, Customer Experience Improvement, and Employee Experience & Retention. Cost Reduction, Revenue Growth and Compliance & Risk produce figures in currency and are treated as hard benefits. Customer Experience and Employee Experience are soft benefits: tracked as a measured delta against a baseline and not converted to currency. Productivity is measured in hours.
ROI is net annual benefit divided by total one-time cost, expressed as a percentage. Net annual benefit is total annual value minus annual running costs. Payback period in months is total one-time cost divided by net annual benefit, multiplied by twelve. A common error is paying back the one-time cost against gross annual value rather than net, which flatters payback.
Converting an NPS point into currency requires assumptions specific to one organisation's customer economics, and those assumptions rarely survive a finance review. Recording them as deltas against a baseline, for example plus 15 NPS, keeps them measurable without inflating the financial case. Once a reviewer rejects one assumption inside a total, they discount the whole total.
The proportion of process volume an initiative is expected to handle or eliminate. A deflection rate of 80 percent means one item in five still requires manual involvement. It is stated by the person building the business case as an explicit target, rather than inferred by a scoring model, because they will have to defend it.
No. Released hours become a cost saving only when they change a cost line, through headcount avoidance, reduced overtime or agency spend, or a measurable increase in output from the same team. If freed time is absorbed back into existing workload it is a real productivity gain but not a bankable saving. This is the first question a finance team will ask.
Hard benefits are expressed in currency and included in the financial total: cost reduction, revenue growth, and compliance and risk mitigation. Soft benefits are tracked as a measured delta against a baseline and reported alongside the total, not inside it: customer experience and employee experience. The distinction exists because a soft benefit converted into currency introduces assumptions a finance reviewer is unlikely to accept, and rejecting one assumption inside a total causes them to discount the whole total.
No, not in the financial ROI figure itself. Soft benefits such as customer experience and employee experience should be reported as a measured delta against a stated baseline, for example plus 15 NPS, alongside the ROI calculation rather than folded into it. This keeps the financial total defensible and the improvement measurable in its own right.
Multiply the annual hours a process consumes by the deflection rate to get annual hours saved, then divide by 2,080 (260 working days multiplied by 8 hours) to get full-time equivalents released. Released hours only become a financial saving once they change a cost line, through headcount avoided, reduced overtime or agency spend, or measurably more output from the same team.
There is no universal answer, but a payback under three months usually invites more scrutiny than praise, because reviewers assume an input is wrong before they assume the result is real. What can be shown is the method: payback in months is total one-time cost divided by net annual benefit, multiplied by twelve. In SilkFlo's own customer portfolio, a five-person Centre of Excellence team modelling initiatives this way achieved under two months' payback.
The break-even deflection rate is the minimum share of process volume an initiative must handle for its financial benefit to cover one-time and annual running costs within the first year. Below that rate, the initiative does not pay back inside year one at the volumes and costs entered. It is calculated from the same inputs as the rest of the case, so it moves as those inputs move.
Method: the same six-driver KPI structure SilkFlo uses inside its AI Value Realization Platform. Built by SilkFlo, London.
The calculator itself runs entirely in your browser — your inputs never leave your device and nothing is stored. If you request the one-page summary, the email address you provide, together with the figures you entered, is sent to SilkFlo to deliver it and to understand what you're evaluating; nothing else on this page is. Figures are estimates based on the inputs you provide and are not financial advice. Currency selection changes the symbol only — no exchange rate is applied.