Tax advisers
Connect every analysis to verifiable sources.
Eight staff search internal precedents and collect documents before preparing a brief. Versions and tax periods sometimes become mixed.
These cases show how we would apply our method to different professional situations. Business profiles, countries, timelines and gains are illustrative assumptions. They do not describe completed client engagements.
Illustrative audit · 4 pages · no registration
- Simulated schedule
- 8 weeks
- Support planned in the scenario
- 10 consulting days
- Simulated net time saving
- 27 h / month
Assumed internal commitment : 4 person-days. Consulting days are spread across the schedule and do not mean full-time attendance. Internal days are additional. Scope and a quote must be agreed for a real engagement.
01 / Proposed audit scope
Tax advisory firm
Factual preparation of a brief within a defined tax scope, without automated regime selection or advice.
What we would examine
20 redacted briefs, sources labelled by country and period, and questions with known answers.
02 / The process, department by department
| Department | Simulated starting point | Proposed process | Human approval |
|---|---|---|---|
| Professional practice | Manually search precedents | Source passages labelled by date and country | Senior tax adviser |
| Administration | Difficult document version tracking | Document index and missing-information list | File manager |
| Finance & accounting | Fragmented preparation of data tables | Extracted data linked to supporting documents | Engagement reviewer |
The approach to compare
Search an authorised, dated corpus with citations and abstention when sources are missing. Explicitly test period and jurisdiction confusion.
What stays with the professional
The adviser checks source currency, the client’s circumstances and every conclusion. No automated filing, final tax calculation or advice.
03 / From audit to an independent team
8 weeks.
10 consulting days.
A simulated engagement timeline, with a decision and deliverable at every stage. Rollout depends on test results.
Week 1 · 2 consulting days
Observe & measure
20 redacted briefs, sources labelled by country and period, and questions with known answers.
Proposed deliverable Process map, baseline measurement and issue register.
Week 2 · 1 consulting days
Scope & decide
Compare simplification, existing features and AI. Define scope, access, full costs and stop criteria before building.
Proposed deliverable Audit report, priority matrix and test protocol.
Weeks 3–5 · 3 consulting days
Prototype
Search an authorised, dated corpus with citations and abstention when sources are missing. Explicitly test period and jurisdiction confusion.
Proposed deliverable Limited prototype, test set and approval workflow.
Weeks 6–7 · 2 consulting days
Test & decide
Every reference must be retrievable for the correct country and period; unsupported answers must be withheld.
Proposed deliverable Comparative assessment: proceed, adjust or stop.
Week 8 · 2 consulting days
Train & hand over
Train users on routine and exceptional cases. Appoint an owner and document the manual fallback. If testing fails, hand over findings and a correction plan.
Proposed deliverable Usage guide, team workshop and 30-day follow-up plan.
04 / What the scenario aims to achieve
Simulated net time saving
27 h / monthFree time to analyse client-specific issues and discuss options.
- Assumed monthly volume
- 72 brief preparations
- Baseline time per unit
- 60 min
- Target time, including human review
- 35 min
- Monthly monitoring and maintenance
- 3 h
Transparent calculation
72 × (60 − 35) ÷ 60 − 3 = 27 h / month
Volume × (baseline time − target time) ÷ 60 − monthly monitoring.
If only half the volume benefits from the process, with the same monthly monitoring: 12 h / month
This calculation is a simulation, not a client measurement. Target time includes corrections and review. Time released is not a demonstrated cash saving; it depends on actual volume and adoption.
Costs to compare with the benefit
Cost advice, data preparation, integration, licences, training and maintenance before committing. Price, return on investment and revenue effects must be established for each engagement.
How we would verify the result
Compare both methods on a set separate from preparation examples. Measure total time, corrections, critical errors and actual use. Keep difficult cases in the assessment; suspend if quality or confidentiality deteriorates.
Professional acceptance criterion : Every reference must be retrievable for the correct country and period; unsupported answers must be withheld.
After the pilot
During the 30 days after an approved pilot: review metrics weekly, examine exceptions and decide whether to maintain, correct or expand the scope. This follow-up period is outside the scenario schedule and consulting days.
Before a pilot, define permitted data, access and any supplier reuse. Country-specific and professional rules must be checked for the real engagement.
Background reference: CNIL, using a generative AI system.Does this sound like your organisation?
Start with your reality: a time-consuming process, your current tools and the people involved. We can assess whether a similar approach makes sense for you.