ROI calculator for AI projects: how to calculate honestly in May 2026

What this is about

An ROI calculator for an AI project is a sober piece of Excel – no projection, no vision. It relates three quantities: hours saved per month, the genuinely recoverable hourly rate, and the project's one-off plus recurring cost. The output is a number in months – when the project turns net positive.

The formula reads: ROI as percent = (hours saved per month × hourly rate × running months − total project cost) / total project cost × 100. The break-even month arrives when the numerator hits zero. That is the honest version. The dishonest version reads "we save 20h per week × CHF 150 × 52 weeks = CHF 156,000 annually" and ignores three realities: not every saved hour is billable, employees stay employed, and setup cost falls before any savings.

In Swiss fiduciary practice 2026 a reliable corridor shows up: AI packages in the CHF 1,700 to 14,500 range turn positive between month 2 and month 8 depending on project type and load. Deloitte 2026 reports for the Swiss SME market an average CHF 3.70 of ROI per franc invested at 2 to 4 months break-even – that matches our own pilot data.

Why an honest calculation matters

Three reasons. First: management financing the project from cash flow needs a reliable date for the return to positive. Overestimated savings cause cash-flow stress in month 4. Second: employees expected to use the system every day accept an honest story better than an overblown salvation narrative. Third: the question "was it worth it?" gets asked in month 6 – and whoever has a clean prediction conducts that conversation with numbers, not with feelings.

The most common trap confuses "freed-up hours" with "billable hours". If a payroll clerk spends 90 minutes/day on client questions and AI shortens that to 30 minutes, the office gains 60 minutes – but only financially if those 60 minutes are billable as hourly rate, or if the additional capacity serves an additional client. Under flat-fee mandates or unsellable residual capacity (no new client in sight) the effect is real (overload drops) but not directly monetisable.

The schnellstart Frauenfeld case study shows it done right: a 12-FTE fiduciary office automated invoice processing and email triage, gained 13 hours per week, generated CHF 35,000 annual savings on a CHF 4,800 setup (KI-Fundament) – ROI in 7 to 8 weeks. The precondition: freed capacity was actively converted into new-client acquisition, not simply absorbed.

Worked examples from practice

Example 1: fiduciary collections. A St. Gallen office of 8 FTEs handles 220 dunning steps 1-3 per month manually. Avg. 4 minutes per dunning – find the receipt, adapt the template, choose the channel. That is 14.6 hours per month. With AI collections the effort drops to 2 minutes per dunning (approval), saving 7.3 hours per month. At an internal cost rate of CHF 120 that is CHF 876 per month. Project: AI backend pilot CHF 7,800 + Managed Basic CHF 600/month. Break-even at (7,800) / (876 − 600) = 28 months – long at first glance. The calculation misses: collections is usually flat-fee or bundled in the client package, not hour-billable. What is genuinely monetisable here is overtime avoidance (CHF 0 to 200/month) and deadline discipline (fewer receivables write-offs – estimated CHF 400/month from 80 clients upward). Real calculation: (400) / (7,800/24 + 600) = ROI in 35 months. This project would simply not pencil under Managed Basic. With KI-Fundament (CHF 4,800) instead of the pilot (self-operated, no managed), break-even is at 4,800 / 400 = about 12 months – longer than under the old CHF 2,800 entry, but still within the 24-month horizon and economically defensible.

Example 2: client questions via chatbot. A 5-FTE office in Zug receives 12 queries/day via email. Current effort per query: 6 min research + 4 min reply = 10 min, of which 6 min cut by RAG. At 22 working days that is 26 hours/month. 50% billable: 13 hours × CHF 140 = CHF 1,820/month. Project: AI backend pilot CHF 7,800 + Managed Pro CHF 1,200/month. ROI = (1,820 − 1,200) / 7,800 × 12 = 95% annual, break-even month 13 including managed cost. Without managed (self-operated): break-even month 5.

Example 3: Frauenfeld setup (schnellstart case). 12 FTEs, KI-Fundament CHF 4,800, savings of 13h/week, 52 weeks × 13h × CHF 100 = CHF 67,600 gross. 50% genuinely billable: CHF 33,800 – reported was CHF 35,000. ROI in 7 to 8 weeks (= 4,800 / (13 × 100 × 0.5) = 7.4 weeks). Conclusion: KI-Fundament projects often break even in single-digit weeks if the freed capacity actually gets billed.

Example 4: AI value-add sale instead of cost saving. A variant rarely calculated but increasingly important in 2026 fiduciary work. An office builds a new tariff "express VAT preparation in 3 business days" on top of its AI pipeline and charges CHF 50/month extra per client. With 40 of 120 clients buying it: CHF 2,000/month additional revenue – with zero extra staff. Here "ROI" is direct revenue, not cost saving. This calculation is clean because no hour conversion is needed.

ROI calculation in 6 steps

1. 01Pick a process: one concrete process (collections, OCR, client queries) – not "AI in general".
2. 02Measure today's time: three employees log real time per case for 5 days each. Use the average, not the gut feeling.
3. 03Estimate the saved share: 30 to 60% is realistic. "We save 90%" deserves suspicion.
4. 04Set the hourly rate: not the wish rate but the genuinely billable one – subtract flat-fee share.
5. 05Check monetisability: does the freed capacity come back as additional revenue? If not, set CHF ROI to zero and argue via staff relief.
6. 06Compute three scenarios: best case (60% saving, 60% billable), base case (45/45), worst case (30/30). Decide at the base case.

When the calculator helps

The ROI calculator is mandatory in every audit-light report. It becomes concretely useful when (a) the office knows its internal hourly rate – the truly recoverable one, not the wish rate, (b) the share of "billable hours" within total savings can be estimated realistically (rule of thumb 30 to 60% in fiduciary work), and (c) the running cost (server + token + managed) is at hand.

Typical situations: management wants to budget a pilot internally – the calculator delivers the answer to the board. A senior partner is sceptical and needs a reality check – the calculator shows where the plan is strong and where it is weak. Employees want to know what the project means for their day – the calculator clarifies that it is not about staff cuts but about whether the freed time becomes otherwise billable.

When the calculator misleads

The calculator misleads when "rate × time" does not fit reality. Three typical traps.

First: flat-fee mandates. If a client has an annual flat-fee contract and the office already runs full capacity on them, AI does not lift revenue – it only reduces effort per hour. Without a new-client pipeline the office is happier but not richer.

Second: non-monetisable hours. If the saved 5 hours per week land as "earlier home time" instead of additional work, the effect is great for staff but invisible to accounting. That can be the intended outcome – but then the project should not be sold as financial ROI.

Third: one-off effects. Some automations save 80 hours in month 1 (e.g. initial indexing of all contracts) but only 4 hours from month 4 onward. Extrapolating the month-1 effect × 12 yields a pretty number and a bad forecast.

Important: the calculator presented here is a rule of thumb, not a binding business case. For investment decisions above CHF 50,000 we recommend involvement of an external auditor or validation through a study like BAK Economics on Swiss digitisation ROI.

Trade-offs

FAQ

Related topics

Sources

1. schnellstart.ai – KMU-Case Frauenfeld (13h/Woche, ROI 7 Wochen) · 2026-03
2. TYTOS Schweiz – KMU-Leitfaden 2026 (Break-Even 2 bis 4 Monate) · 2026-03
3. AI Automation Hub – Kosten und ROI-Berechnung Schweiz · 2026-02
4. BAK Economics – Digitalisierungs-ROI Studie KMU Schweiz · 2025-11
5. fairlane.systems – Pricing Audit Light / KI-Fundament / Pilot · 2026-05