1. Positioning, audience and learning outcomes
This is aprofessional services proposal independent of the retail courseThe course targets knowledge-intensive work such as consulting, accounting and tax, legal, risk, engineering/design. The emphasis is not on AI replacing professional conclusions, but on helping teams more disciplinarily organize sources, process documents, prepare drafts, collaborate on delivery, and retain auditable professional review. Different professions have different practice obligations, client contracts and data restrictions; the general modules teach methods only—statutory or regulated professional judgments must be made by qualified persons.
Recommended participants:Partners/directors, project managers, consultants and analysts, research and knowledge management, business development/client service teams. Recommend 12–18 people per class, grouped by function or profession; no programming experience required, but participants should have document-handling and basic information retrieval skills.
Standard version:Two days of in-person or live interactive training, 6 hours per day, total 12 hours; lunch and breaks excluded. Recommended instruction language: Cantonese with Traditional Chinese materials as primary, keeping English client documents and bilingual delivery exercises. For cross-border teams, case studies and languages can be grouped and adjusted.
Observable capabilities after completion:Participants can select appropriate AI-assisted tasks, distinguish public, internal and client-confidential information; use approved tools to produceevidence-backedresearch and document drafts; identify fabricated citations, erroneous figures and overconfident conclusions; retain versioning and review records; and propose a small pilot that complies with client authorizations, professional responsibilities and organizational policies.
2. Pre-course diagnostics and materials required
Tungverse recommends conducting 30–45 minute needs interviews with the business lead, information security/privacy lead and each professional line manager to confirm the types of data the team handles, available tools, client contract restrictions, deliverable approval workflows, and the most time-consuming but standardizable tasks. Participants complete a ~10-minute pre-course self-assessment and a short “source verification” baseline task before the course for comparison after completion, without assuming any specific improvement magnitude.
Training materials use only public, licensed or synthetic content.Prepare a fictional client engagement brief, three verifiable public-source excerpts, a fictional meeting minute, a fictional tender/proposal requirement, two report samples with different styles, and a 20–30 line simulated analysis table. Even de-identified data must be confirmed by the client for re-identification risk or confidentiality constraints. If an enterprise AI environment is not yet approved, all exercises use synthetic materials and demo accounts.
| Pre-course checks | Why it matters | Who confirms |
|---|---|---|
| Client contracts, confidentiality/privilege and disclosure obligations | Determine which data, tasks or tools are off-limits | Client relationship lead, legal and relevant practice leads |
| Approved AI tools, storage locations, logging and access permissions | Restrict allowable inputs and retention methods | IT / information security / privacy lead |
| Approved public data and fictional cases | Enable source traceability and verifiable delivery practice | Knowledge management and training lead |
| Approval criteria and brand tone for proposals/reports | Ensure drafts meet organizational and client requirements | Practice leads and business development team |
3. Two-day course overview
| Agenda | Module | Duration | Key outputs in session |
|---|---|---|---|
| Day 1 | 1. AI task map for professional services | 90 minutes | Task/risk matrix and candidate scenarios |
| Day 1 | 2. Client confidentiality, privacy and professional responsibility | 90 minutes | Data classification and AI usage red lines |
| Day 1 | 3. Reliable research and source verification | 90 minutes | Research memos with sources and citation checklist |
| Day 1 | 4. Document reading, summarisation and knowledge organisation | 90 minutes | Traceable document summaries and knowledge entries |
| Day Two | 5. Proposal, report and draft deliverables | 90 minutes | Draft deliverables including assumptions, evidence and items for review |
| Day Two | 6. Client meetings and collaborative communication | 90 minutes | Reviewed meeting minutes and follow-up actions |
| Day Two | 7. Practice scenarios for professional line handoffs | 90 minutes | Workflow and responsibility diagrams for each line |
| Day Two | 8. Project implementation, assessment and pilot design | 90 minutes | One‑page pilot plan and before/after measurement records |
Eight modules of 90 minutes each, totaling 720 minutes / 12 hours. The timing below includes instruction, practice and review; it can be adjusted according to the client's professional-line mix, but do not compress verification and risk exercises.
4. Detailed module contents
Module 1 | AI task map for professional services (90 minutes)
Objective:Distinguish between knowledge work that can be assisted and judgments that must be made by professionals.Contents:Review the engagement lifecycle—research, discovery interviews, analysis, drafts, review, client delivery and knowledge retention; select scenarios using four axes: value, data availability, consequences of errors, and verification cost.Exercise:Compare three tasks—public market research summaries, a fictional proposal draft, and legal/tax opinions involving real client confidentiality—and classify them as "suitable for synthetic-data practice immediately", "requires prior approval and controls", or "must not be handled by AI independently".Timing:15 minutes scenario introduction, 25 minutes task breakdown, 35 minutes group screening, 15 minutes review.Output:Each group produces an opportunity/risk matrix together with a list of professional decisions that should not be automated.
Module 2 | Client confidentiality, privacy and professional liability (90 minutes)
Objective:Be able to identify confidentiality, personal data, intellectual property and client authorization risks before inputting, and arrange professional review after outputs.Contents:Classification of public/internal/confidential/restricted data; purpose limitation, minimal data, account and permission controls for client data; confidentiality, legal privilege, conflicts of interest, disclosure and professional duties relevant to different professions; and storage, retention and incident reporting of AI prompts and outputs.Exercise:Twelve data cards (annual reports, internal methodologies, client engagement letters, witness statements, compensation data, anonymised transcripts, etc.) for classification; identify sensitive fields in a client brief proposed for AI input and rewrite them into safe synthetic practice material.Timing:20 minutes principles, 25 minutes classification, 30 minutes red‑line exercise, 15 minutes feedback.Output:Departmental data handling decision tree and a two‑stage "before input—after output" checklist. Hong Kong organisations may refer toOffice of the Privacy Commissioner for Personal Data checklist: Guidance for employees using generative AI; this does not replace legal, professional or client contractual advice.
Module 3 | Reliable research and source verification (90 minutes)
Objective:Create research notes that colleagues can re-check, identifying nonexistent sources, outdated material and conclusions lacking evidence.Contents:First define the research question and source hierarchy; distinguish primary documents, secondary commentary and AI summaries; record URLs/document versions/page numbers/dates; separate “sourced facts,” “inferences” and “to be verified.”Exercise:Using three publicly available sources provided in advance, have the AI propose a comparative framework first, then have participants verify each citation and number; deliberately include one plausible-looking but nonexistent citation and require finding and removing it.Timing:15 minutes demonstration, 20 minutes methodology, 40 minutes group verification, 15 minutes cross-review.Output:One-page research memo, source list and error correction log.
Example prompts:
You are a research assistant. Based only on the public sources I provide, extract into a three-column table: Confirmed Facts / Inferences Requiring Verification / Missing Information. Each fact must cite source name, date and paragraph or page number; if the source does not mention it, write “no evidence found.” Do not fabricate quotations, statutes, cases or figures. Finally, list items requiring professional review. [pastepublic or approvedsource excerpts]
Module 4 | Document reading, summarisation and knowledge organisation (90 minutes)
Objective:Use AI to assist reading long documents while preserving original context, exceptions and version information.Contents:Segmenting text, summarising and comparing, extracting key definitions/dates/responsible parties; version differences; tags, effective dates, ownership and expiry alerts for internal knowledge bases.Exercise:Compare two versions of a fictional scope of engagement or policy document to produce a “change item — source location — potential impact — reviewer” table; convert recurring public issues into sourced knowledge entries, and identify reservation clauses the AI missed.Timing:15 minutes scenario, 20 minutes demonstration, 40 minutes practice, 15 minutes peer review.Output:Includes a paragraph‑referenced differences table, a knowledge entry template and version management rules.
Module 5 | Proposals, reports and draft deliverables (90 minutes)
Objective:Turn verified research and requirements into clear client documents.Draftrather than letting AI determine recommendations.Contents:Break down RFP/engagement requirements, content structure, assumptions and scope, evidence chains, methods and limitations, chart descriptions; how to avoid overpromising, fabricated capabilities and unapproved client case examples.Exercise:Using a fictional RFP and source pack, produce a two‑page proposal: problem understanding, scope of work, deliverables and clarification questions; compare line by line with the original requirements and flag all statements requiring managerial approval.Timing:15 minutes structure, 20 minutes demonstration, 40 minutes drafting and revision, 15 minutes review.Output:A draft proposal/report for managerial review and an “evidence — claim — approver” reconciliation table.Referenceobservations of the Thomson Reuters Professional Services AI report workflow; do not cite its industry statistics as course outcome promises.
Module 6 | Client meetings and collaborative communication (90 minutes)
Objective:Fromapproved meetingmaterials, produce verifiable records and action items, avoiding turning speculation into client commitments.Contents:Advance notice and authorization for recordings/transcripts, confidentiality management; separate decisions, items to confirm, owners and deadlines; bilingual communication and tone; manual review before sending to clients.Exercise:From a fictional meeting transcript, produce minutes, identify AI‑misattributed speakers and one delivery date that was never agreed; draft separate “internal follow‑up” and “client confirmation” versions.Timing:15 minutes authorization and risks, 20 minutes demonstration, 40 minutes practice, 15 minutes verification.Output:Meeting minutes template, action tracker and rules for marking “not effective until confirmed by client.”
Module 7 | Professional line escalation simulations (90 minutes)
Objective:Apply common source handling, confidentiality and review methods across different professional lines.Timing:15 minutes group briefing, 20 minutes design process, 40 minutes simulation execution, 15 minutes peer review. Each team must document data sources, allowed tools, confidentiality boundaries, professional sign-off, record-keeping method, and stop/escalation criteria.
| Team split | Synthetic or public data practical | Professional gatekeeping focus |
|---|---|---|
| Management consulting / market research | Compare public sources and draft a sourced issue tree and presentation outline | Distinguish facts, assumptions and recommendations; do not fabricate client outcomes |
| Accounting / tax / risk | Extract checklists and items for verification from fictional policies or processes | Do not allow AI to independently produce tax, audit or financial conclusions; qualified personnel must review |
| Legal and compliance | Identify issues and compare versions for fictional clauses or public regulatory materials | Do not create case law, legal authority or legal opinions; maintain confidentiality/privilege and require lawyer review |
| Engineering / design and other consultancy | From a fictional requirements document, compile specification questions and delivery risks | Do not permit AI to unilaterally approve safety, technical specifications, or professional sign-offs |
Output:A single workflow diagram: Human original process → AI-assisted steps → Evidence and quality checks → Professional sign-off → Client delivery. These examples are for training and do not indicate that Tungverse provides any form of licensed professional advice.
Module 8 | Project implementation, assessment and pilot design (90 minutes)
Objective:Complete a demonstrable, auditable AI-assisted work package and decide whether it is worth a small-scale adoption.Exercise:Each team chooses one of: research memo, document comparison, draft proposal, or meeting follow-up; first complete a baseline using the original process, then redo using approved AI tools and synthetic materials, recording time, factual errors, missing sources and amount of human edits. Present “AI draft — verification evidence — human corrections — final version” and note which recommendations were rejected.Timing:10 minutes explanation, 20 minutes baseline/plan, 45 minutes completion and presentation, 15 minutes scoring feedback.Output:One-page pilot plan specifying the task, data types, client authorization, responsible person, approved tools, review gates, baseline metrics, exit conditions and review date.
5. Assessment criteria and measurement methods
Recommend using synthetic work packages of similar difficulty for pre-/post-course comparison.Passing recommendation: overall score at least 80/100 and no major confidential data or privacy breaches; this is an internal course threshold and is not an external professional qualification. Regulated deliverables must still be approved following existing practice workflows.
| Scoring dimensions | Points | Verifiable evidence |
|---|---|---|
| Source and factual accuracy | 30 | All key claims are traceable; citations, figures, versions and dates are correct |
| Confidentiality, privacy and client authorization | 25 | Data classification and tool use compliant; no unapproved client data input |
| Analysis and professional appropriateness | 20 | Clearly distinguish facts, assumptions, limitations and matters for professional judgment |
| Human reviewability and auditability | 15 | Maintain records of changes and approvals with clear owners and escalation points |
| Client deliverable availability | 10 | Clear formatting, appropriate terminology, aligned with the engagement scope |
| Total | 100 | Do not trade speed for accuracy or confidentiality |
Pilot should be measurableCompletion time, source error rate, rework proportion, review time, revision rounds of client-facing versions, participant confidence and incident countsCollect a baseline first and let the organisation set targets; any claimed time savings or service quality improvements may only be stated externally after actual measurement. Client satisfaction and internal adoption rates can be supplementary but do not replace professional quality metrics.
6. Post-course pilot, deliverables and boundaries
Recommended four-week pilot (optional):Week 1: professional line and risk owners approve a single low-risk scenario, tools and data; Weeks 2–3: trial within the team with spot checks of outputs, source tracing and manual corrections; Week 4: review quality, review costs, client feedback and any incidents, then choose to scale, modify or stop. Sharing knowledge across clients, using real engagement materials, recording transcripts or deploying automated agents require separate approval and are not default course activities.
Tungverse suggested deliverables:Instructor lesson plans and slides, synthetic engagement and source data package, research/document comparison templates, prompt templates, source verification checklist, data classification and AI usage redlines, pre/post tests, routing exercise scorecards, pilot plan. Clients must provide their own professional approval standards, client authorisations and list of available tools; Tungverse can assist with training and process design but does not replace the client’s legal, accounting, engineering or other licensed professional responsibilities.
Implementation redlines:Unapproved public AI tools must not process real client confidential information, personal data, privileged material or unpublished financial figures; AI must not independently issue legal/tax/audit/engineering opinions; non-existent cases, citations, standards or figures must not appear in deliverables; recordings and verbatim transcripts must be approved under organisational policy and applicable authorisations. If policy is unclear, pause inputs and external deliverables and refer to the responsible owner for a decision.
7. References and customization questions
- Tungverse corporate training overviewandProfessional services application overviewBrand positioning and industry direction.
- Office of the Privacy Commissioner for Personal Data Hong Kong: Checklist for employee use of generative AIEmployee use policy, allowable input data, output checks and training resources.
- Thomson Reuters: 2026 AI in Professional Services ReportIndustry context for professional services workflows, governance and measurement; this course does not rely on the report’s data to make performance promises.
- NIST: AI Risk Management Framework and generative AI supplemental materialsCommon reference for trustworthy AI and risk management.
Please confirm before formal customisation:Primary service lines (consulting, accounting, legal, engineering or others), participant background and numbers, common deliverables, applicable jurisdictions, client data and contract constraints, enterprise AI tools available, instruction language, whether recording/transcripts are permitted, and who will provide final professional sign-off. These choices will change the depth of cases and risk boundaries in Modules 5–7.