The scenario
The following is a model calculation based on our experience and typical benchmarks from the accounting industry. It serves as a realistic orientation – not an actual case study.
Company: Accounting firm in the Zug region Size: 12 employees Specialization: Bookkeeping, tax returns, payroll for SME clients Accounting software: Bexio
The starting point
3 clerks spend approximately 80% of their working time on receipt processing:
- Daily volume: 50–80 receipts (invoices, receipts, bank statements) arriving by email, mail, and phone photos from clients
- Process: Open receipt → identify supplier → verify amount → determine booking category → enter in Bexio → follow up with client if unclear
- Time per receipt: approximately 5–8 minutes
- Error rate: approximately 3–5% (wrong assignment, typos, forgotten entries)
The core problem: The clerks are consumed by routine work and rarely get to more demanding tasks like client advisory or tax consulting.
What the AI system handles
Step 1: Automatically capture receipts
All incoming receipts – whether by email, scan, or photo – are automatically captured by the AI system. It recognizes that it's a receipt and starts processing.
Step 2: Extract data
The system identifies:
- Supplier (name, address, UID number)
- Amount (gross, net, VAT)
- Date (invoice date, due date)
- Booking category (based on supplier and historical bookings)
Step 3: Prepare the booking
The AI agent assigns the receipt to the correct client and booking category and prepares the entry in Bexio. When multiple assignments are possible, the system suggests the most likely one.
Step 4: Quality control
- Clear cases (approx. 85%): Presented directly for approval – one click is enough.
- Uncertain cases (approx. 15%): Forwarded to the responsible clerk with context and a suggestion.
Step 5: Book in Bexio
After approval, the entry is automatically created in Bexio. No manual input, no typing.
The numbers
Time savings
| Metric | Before | After |
|---|---|---|
| Time per receipt | 5–8 minutes | under 1 minute (approval click) |
| Time per week | 35–40 hours | 12–15 hours |
| Time saved | – | 20–25 hours/week |
Error reduction
| Metric | Before | After |
|---|---|---|
| Error rate | 3–5% | under 0.5% |
| Correction effort | 3–4 hours/week | under 30 minutes/week |
Capacity gain
The 20+ hours saved per week enable:
- More time for client advisory and consulting
- Capacity for 15–20% more mandates without additional staff
- Higher employee satisfaction (less monotony)
Investment vs. savings
AI system costs
| Phase | Cost |
|---|---|
| Analysis (1 week) | CHF 4,500 |
| Prototype + Implementation (6 weeks) | CHF 28,000 |
| Ongoing costs | CHF 800/month |
| Total in year 1 | CHF 42,100 |
| Total from year 2 | CHF 9,600/year |
Savings
| Item | Calculation | Per year |
|---|---|---|
| Saved working time (20h × CHF 55) | CHF 1,100/week | CHF 57,200 |
| Less error correction | CHF 600/month | CHF 7,200 |
| Total savings | CHF 64,400/year |
Payback
| Calculation | |
|---|---|
| Investment year 1 | CHF 42,100 |
| Savings year 1 | CHF 64,400 |
| Net gain in year 1 | CHF 22,300 |
| Payback period | approx. 8 months |
| ROI over 3 years | approx. 850% |
All figures are model calculations. Actual values depend on receipt volume, booking complexity, and existing data quality.
What the employees do now
This is the crucial point: it's not about cutting positions. It's about freeing talented people from monotonous work.
After automation:
- Clerk A takes on additional mandates in tax consulting
- Clerk B handles the personal client advisory that was previously neglected
- Clerk C becomes the point person for quality checks and onboarding new clients
The team stays the same size – but becomes significantly more productive and satisfied.
The project timeline
| Week | Phase |
|---|---|
| Week 1 | Process analysis: What types of receipts? What booking rules? What special cases? |
| Weeks 2–4 | Prototype with real client receipts. Live demo for the team. |
| Weeks 5–7 | Integration into Bexio. Fine-tuning recognition rules. Training. |
| From week 8 | Ongoing operations. Monitoring and optimization. |
Conclusion
For an accounting firm with regular receipt processing, AI automation isn't a distant future – it's an investment that typically pays for itself within 8–12 months.
Getting started is simple: an analysis for CHF 4,500 shows you exactly how much potential lies in your processes.
Does this sound like your office? Let's discuss in 30 minutes what's possible for your accounting firm.
