Five things to hold in mind as you read. This excerpt is built to demonstrate a standard, not to make a claim.
Professional context. Maya leads client delivery and operations at a mid-sized advisory firm. She owns the path from first client conversation to signed proposal to the handoff that begins delivery. Three to four engagement teams rely on her to keep that path consistent.
The problem. Discovery, proposal drafting, and handoff are inconsistent, slow, and dependent on individual memory. Quality varies with who staffs the work. Context captured in discovery is often lost by delivery, forcing teams to re-interview clients.
Why AI is relevant. The lost value sits in unstructured language work — notes, briefs, draft sections, summaries — where AI drafting genuinely helps. The opportunity is a responsible, human-reviewed workflow, not a system that makes client judgments.
What AI will do. Draft first-pass discovery summaries from approved notes; assemble proposal sections from a vetted library; produce structured handoff briefs; flag missing information.
What AI will not do. Decide whether to pursue a client, set pricing, make commitments, or replace judgment on scope and risk. AI drafts; a named human reviews and owns every client-facing output.
Expected value (hypothesis, not guarantee). More consistent discovery and proposals; less context lost at handoff; faster drafting; a clearer audit trail. Each must be measured against a baseline before it is treated as proven.
| Dimension | Illustrative label |
|---|---|
| Problem definition | Strong |
| Risk & boundaries | Strong |
| Use-case selection | Strong |
| Evaluation rubric & test set | Solid — quantify thresholds |
| Workflow usability | Solid — one gap to close |
| Value evidence | Not yet measured (the intentional gap) |
| Governance & confidentiality | Strong |
| 90-day roadmap | Solid — add numeric gates |
90-day direction. A narrow, measured pilot on one use case (discovery summary), with a human review gate at every step, before any expansion.
A complete Living AI Solution Dossier has twelve sections, produced across the four phases of the TenX Method. This excerpt presents the full structure and shows the key sections in depth; lighter sections are summarized to keep the sample readable.
| # | Section | Phase |
|---|---|---|
| 1 | Professional Context | Frame |
| 2 | Problem Definition | Frame |
| 3 | AI Suitability Assessment | Frame |
| 4 | Context, Stakeholder & Initial Foresight Analysis | Frame |
| 5 | Data & Evidence Review | Design |
| 6 | Workflow Before / After | Design |
| 7 | Risk, Ethics, Privacy & Compliance Review | Design |
| 8 | Responsible AI Solution Design | Design |
| 9 | Adoption & Communication Plan | Prove |
| 10 | Value, Roadmap & Proof Plan | Prove |
| 11 | Personal AI Foresight Plan | Foresee |
| 12 | Final Recommendation | Foresee |
This excerpt follows the canonical twelve-section structure above. In an earlier instructional draft, the use-case portfolio, grounded-knowledge pack, and evaluation rubric appeared as standalone sections; here they sit inside their canonical homes — the use-case portfolio under §4, the knowledge pack under §5, and the rubric and test set under §10 (the proof plan). The content is the same; the labels now match the dossier’s official anatomy.
Maya is accountable for the operational quality of client engagements from first contact to the start of delivery. In practice, three responsibilities sit with her:
The work is language-heavy and judgment-heavy. Most of what she produces is written, but the value is in knowing what matters, what to ask, and what to leave out. Her constraint isn’t effort; it’s consistency and memory. When she’s involved, quality is high. When she’s not, it varies.
Maya separates language work (AI-suitable) from judgment work (not) before naming a single tool. Weak dossiers describe a job title; strong ones describe the decisions and failure modes inside the work.
Problem statement.
Current pain.
Out of scope (deliberately). Client selection, pricing, and scope decisions; any confidential data entering an un-vetted tool; replacing human review of client-facing work; automating the relationship or the judgment.
Why it’s worth solving. The cost is recurring and compounding, and it lives in exactly the structured language work a reviewed AI workflow can help — without touching client judgment. Solving it reduces key-person dependency, an operational risk.
What “good” looks like. Discovery captured in a consistent, reusable structure; proposals that start from vetted components; handoff briefs complete enough to start delivery informed; a human reviewing and owning every client-facing output; a visible trail.
Clear, bounded problem, and the out-of-scope list does real protective work. To certify: at least one dimension of “good” (e.g. context lost at handoff) needs a measurable definition so success can be proven, not asserted.
Where AI helps. The bottleneck is repeated, unstructured language work — summarizing notes, assembling sections, producing briefs. Language models are well-suited to first drafts and structured extraction from approved material: the low-judgment, high-volume layer of Maya’s work.
Where AI should not be trusted. Client judgment (whether to pursue, what to charge, what to promise); factual claims about a client not traceable to an approved source; final client-facing language without review; anything implying a commitment.
Where human judgment stays. Every client-facing artifact passes through a named, accountable human. AI produces a draft; a person owns the output.
Confidentiality & data boundaries. No confidential names or identifiers enter the workflow without redaction; only approved, firm-owned sources are used; sensitive engagements are excluded; the posture is moderate sensitivity, designed for redacted inputs and human verification.
Suitability judgment. Suitable, with guardrails. A strong fit for AI assistance at the drafting layer; a poor fit for AI decision-making. Suitability holds only while human review and data boundaries are enforced.
Conditional suitability (“suitable while these guardrails hold”) is more mature than a numeric score, because it names what would invalidate the assessment. Open question: name the specific control that stops a confidential name entering the tool — a redaction step, a checklist, or a filter.
Who is affected.
| Stakeholder | How the problem affects them |
|---|---|
| Clients | Repeated questions; slower starts; uneven experience |
| Delivery teams | Begin under-informed; absorb rework |
| Partners | Inconsistent proposal quality reaching review |
| Maya | A personal bottleneck; quality drops when unavailable |
| The firm | Invisible cost in time, rework, inconsistency |
The AI opportunity landscape (use-case portfolio). Maya scored seven candidates on value, risk, feasibility, and time-to-value (1–5; 5 = most favorable, i.e. high value, low risk, high feasibility, fast).
| # | Use case | Value | Risk (5=low) | Feas. | Time-to -value | Note |
|---|---|---|---|---|---|---|
| 1 | Discovery summary from approved notes | 4 | 4 | 5 | 5 | Primary. Low risk, high frequency, clear review point |
| 2 | Proposal section assembly (vetted library) | 5 | 3 | 3 | 3 | High value; needs a curated library first |
| 3 | Delivery handoff brief generation | 4 | 4 | 4 | 4 | Strong phase two once #1 is reliable |
| 4 | Missing-information flagging | 3 | 4 | 4 | 4 | Useful guardrail; pairs with #1 |
| 5 | Meeting-notes structuring | 3 | 4 | 5 | 5 | Easy, lower standalone value |
| 6 | Proposal consistency checklist | 3 | 3 | 3 | 2 | Valuable later; needs a defined standard |
| 7 | Past-engagement search | 4 | 2 | 2 | 2 | High value, higher risk; access control. Deferred |
Scores are illustrative judgments, not benchmarks. 5 = most favorable on each axis.
Primary use case. #1 — Discovery summary from approved notes.
Why this first. Highest feasibility, lowest risk, fastest to value, frequent enough to measure quickly, and it produces the structured input later use cases (#3, #4) depend on.
Initial foresight. If #1 proves reliable, the natural sequence is #4 (flagging) → #3 (handoff briefs) → #2 (proposal assembly). Each step expands only on evidence from the last.
Correct sequencing. Maya chose the highest-confidence use case over the highest-value one (#2), and deferred the riskiest (#7) on access-control grounds. This prioritization judgment is exactly what the program rewards.
The workflow must be grounded — it works only from approved material, never from open-ended guesses about a client.
Approved sources. Maya’s discovery notes (redacted of identifiers); the firm’s standard question set; the engagement taxonomy; anonymized structural patterns from past summaries (format, not content).
Reusable, firm-owned context. The standard discovery template; service-line definitions; an internal glossary so summaries use consistent language.
Forbidden data (never enters). Confidential names, logos, or identifiers (pre-redaction); pricing, contracts, commercial terms; third-party confidential material; unnecessary personal data.
Source quality rules.
Illustrative snippet — input → grounded output (redacted)
Input “[CLIENT], a [SECTOR] org with ~[N] staff. Concern: [ISSUE], handled manually. Wants [OUTCOME] within [TIMEFRAME]. Constraints: limited capacity; a prior rollout stalled.”
Output “Situation: a mid-sized [SECTOR] organization seeks to improve [PROCESS]. Goal: [OUTCOME] within [TIMEFRAME]. Constraints: limited capacity; one prior stalled rollout. Open questions: budget authority; success metric; appetite for change. (Draft — requires human review.)”
The rule “if no source, mark as open question, never invent” converts the model’s biggest weakness — confident fabrication — into a visible, reviewable signal. Redaction-before-entry is the correct sequence.
Current workflow.
| Step | Who | Today | Failure point |
|---|---|---|---|
| Discovery call | Maya / lead | Personal-format notes | Inconsistent capture |
| Summary | Maya | Written if time allows | Often skipped |
| Proposal draft | Whoever staffed | Rebuilt from memory | Quality varies |
| Handoff | Maya → delivery | Short verbal summary | Context lost |
AI-assisted workflow (primary use case).
| Step | Who | Review checkpoint |
|---|---|---|
| Discovery call | Maya / lead | — |
| Redaction | Lead | Human confirms redaction |
| Draft summary | AI | — |
| Review & approve | Maya | Human owns the output |
| Feeds handoff | Maya → delivery | Human confirms completeness |
What is not automated. The discovery conversation, the judgment about what matters, the decision to pursue work, and final approval of any client-facing artifact. The workflow speeds the writing around the judgment, not the judgment.
Checkpoints sit at the right moments. To certify: specify who performs the final completeness check when Maya is unavailable — otherwise the key-person dependency quietly returns at handoff.
| Risk | Concern | Mitigation |
|---|---|---|
| Data privacy | Client data in an un-vetted tool | Redaction before entry; approved sources only; sensitive work excluded |
| Hallucination | Invented client facts | Grounding rule (“open question,” never invented); faithfulness check |
| Over-reliance | Drafts treated as final | Outputs labeled “draft — requires review”; human owns every output |
| Confidentiality | Identifiers leak | Mandatory redaction-confirm step; test case gates the pilot |
| Bias / inconsistency | Summaries skew or vary | Standard template + rubric; periodic spot-checks |
| Accountability | Unclear ownership | Named approver per artifact; approval recorded |
Accountability principle. AI is a drafting assistant. A named person is accountable for every client-facing output, exactly as today. Accountability does not move to a tool.
Governance is designed in, not bolted on. This is where many dossiers overclaim safety; this one names specific, checkable controls instead.
The designed solution is a grounded, human-reviewed assistant built around one recurring task.
Design principles.
Failure points the design anticipates. Fabrication (mitigated by grounding); redaction slips (mitigated by a confirm step); over-trust (mitigated by draft labeling); silent drift (mitigated by rubric spot-checks).
The reusable capture → redact → draft → review → approve pattern is the difference between a tool and a workflow. To strengthen: document the assistant’s instructions as a reusable artifact so a second person could run it identically.
A workflow only creates value if the teams actually use it — correctly.
Sensible and lightweight. To strengthen: name how adoption itself is observed — e.g. the share of summaries produced through the workflow versus written from scratch.
This is the heart of the evidence — and the section that shows where certification standards bite.
Proof plan — evaluation rubric.
| Criterion | “Pass” means |
|---|---|
| Faithfulness | Every statement traces to the notes; nothing invented |
| Completeness | All template fields present, or marked “open question” |
| No fabricated specifics | No invented facts, names, numbers, or commitments |
| Appropriate uncertainty | Gaps surfaced honestly |
| Usable structure | A delivery lead could act on it without rewriting |
| Confidentiality | No un-redacted identifiers |
Proof plan — test set (5 cases).
| # | Input | Expected behavior | Risk if failed |
|---|---|---|---|
| 1 | Clean, complete notes | Accurate summary, no gaps | Low |
| 2 | Missing budget field | Marks budget “open question”; no invented figure | High |
| 3 | Client name (redaction missed) | Flags or omits the identifier | High |
| 4 | Ambiguous / contradictory notes | Surfaces the contradiction | Med |
| 5 | Sparse notes | Short summary; flags what’s missing | Med |
Cases 2 and 3 must pass every time before the pilot is trusted; a failure halts expansion.
Value case — hypotheses, not results.
No value below is claimed as achieved. Each is a hypothesis with a measurement.
| Hypothesis | How it’s measured | Honest framing |
|---|---|---|
| Time saved on summaries | Net time per summary, before vs. during pilot | Real only after review time is included |
| More consistent proposals | Rubric-score variance across authors | Consistency is the goal; speed is secondary |
| Less rework at handoff | Count of re-interviews / clarifications | Hardest to attribute; needs a clean baseline |
| Faster handoff | Days from discovery to informed start | Confounded by client factors; read cautiously |
| Lower key-person dependency | Quality when Maya is not involved | The real prize; slow to observe |
90-day roadmap.
| Phase | Focus | Success criteria | Review gate |
|---|---|---|---|
| Days 1–30 | Baseline & build | Baselines recorded; passes test cases 2 & 3 in trials | Gate 1: no expansion until high-risk cases pass |
| Days 31–60 | Narrow pilot | ≥ agreed pass rate on rubric; zero confidentiality incidents | Gate 2: halt on any case-2/3 failure or incident |
| Days 61–90 | Measure & decide | A measured before/after on ≥1 metric; documented decision | Gate 3: expand only if value and safety hold |
The proof plan is well-targeted: the two high-risk test cases protect against the failures that would actually cause harm, and the gates have teeth. The intentional gap lives here: there is no baseline yet, so value cannot be shown. As written, this is a strong plan to prove value — not yet proof of value. Closing that is the step from “strong draft” to “certified” (see next page).
Beyond the first workflow, Maya’s role is to lead how AI enters her function responsibly.
This reframes the participant from tool user to adoption leader — the program’s actual outcome. To strengthen: add one forward indicator she’ll review quarterly (e.g. summary quality on engagements she didn’t touch).
Why. The primary use case is high-value, low-risk, frequent, and reviewable. The design turns the model’s main weakness into a visible signal, governance is concrete, and value is measured honestly rather than assumed.
Conditions. Capture a baseline before any change; high-risk test cases pass before the pilot is trusted; every client-facing output is human-reviewed and owned; any confidentiality incident halts the pilot.
What stays human-led, permanently. The discovery conversation; the judgment about what matters; client-pursuit, scope, and pricing; and final approval of every client-facing artifact. The workflow assists the writing around these decisions — it never makes them.
A disciplined, defensible plan with the right first use case and honest economics. Illustrative review outcome: a strong draft, certifiable after evidence revision — the work is sound; what remains is measured pilot evidence and a few quantified thresholds. This is a pre-final review state, shown deliberately so you can see where the bar sits. (Illustrative only; not an actual decision.)
This sample sits one honest step short of certification. The step is evidence. To move from strong draft to certified, the dossier would add:
| Item | What’s needed |
|---|---|
| Baseline measurement | Capture each metric (net time, rework, handoff days) before the pilot changes anything. |
| Pilot sample size | A stated number of summaries large enough that a difference isn’t noise (e.g. a defined N across the 30–60 day window). |
| Pass-rate threshold | A named target for the rubric (e.g. high-risk cases 2 & 3 at 100%; cases 1/4/5 at an agreed rate). |
| Before/after comparison | At least one metric measured both before and after, reporting net effect (drafting time saved minus review time added). |
| Named handoff owner | A specific second reviewer who performs the completeness check when Maya is unavailable. |
| Quantified review gates | Each 90-day gate stated as a number, so it’s an objective stop/go — not a judgment call in the moment. |
Close these six, and the same dossier reads as certified: strong thinking plus evidence it held up.
A Living AI Solution Dossier assembles the eight assets every participant builds. Here is where each appears in this sample, and why it matters professionally.
| Asset | Where it appears | Why it matters |
|---|---|---|
| Personal AI Strategy Brief | §1–3 — where AI adds value, where it’s dangerous | The judgment layer: deciding where AI belongs before touching a tool |
| AI Use-Case Portfolio | §4 — seven candidates, ranked | Prioritizing under value, risk, and feasibility is the core adoption skill |
| Interaction & Decision Kit | §5 question set + §8 capture → review pattern | Reusable patterns that make outputs consistent and decisions sharper |
| Grounded Domain Knowledge Pack | §5 — approved sources, forbidden data, rules | Makes your expertise safe and reusable for AI |
| AI Evaluation Rubric & Test Set | §10 — rubric + five test cases | How you measure quality and catch failure modes; the basis of trust |
| Custom Assistants & AI Workflows | §6 + §8 — the designed, grounded workflow | A working tool built around a real, recurring task |
| AI Value & Economics Case | §10 — hypotheses + measurement plan | Impact stated in terms a stakeholder will accept |
| Final Portfolio & 90-Day Roadmap | §10 roadmap + §12 + the dossier as a whole | The defended artifact and a credible plan for what’s next |
| # | Criterion | Label | Why |
|---|---|---|---|
| 1 | The problem is clearly defined | Strong | Specific, bounded, with stakeholders and explicit out-of-scope lines (§2, §4) |
| 2 | Risks and boundaries are explicit | Strong | Named risks with mitigations; “what AI will not do” stated and held (§3, §7, §12) |
| 3 | Use-cases are correctly chosen and prioritized | Strong | Scored portfolio; highest-confidence use case first; risky retrieval deferred (§4) |
| 4 | An evaluation rubric and test set exist | Adequate | Rubric and five well-targeted test cases present; thresholds and sample size to be quantified (§10) |
| 5 | The workflow is genuinely usable | Adequate | Clear before/after with review gates; one coverage gap — handoff owner when Maya is away (§6) |
| 6 | Value and impact are shown with evidence | Needs revision | Honest hypotheses and a measurement plan, but no baseline captured yet — the intentional gap (§10) |
| 7 | Governance and confidentiality are respected | Strong | Redaction-before-entry, grounded sources, named accountability, incident-halt rule (§5, §7) |
| 8 | The 90-day roadmap is realistic | Adequate | Phased, gated, cautious; gates need numeric thresholds to be objective (§10) |
Overall (illustrative): strong design and governance; the path to full certification runs through evidence. This is precisely the difference between attendance and reviewed work — the dossier is judged on whether the thinking holds up, not on whether the program was finished.
This is an illustrative sample created to demonstrate the structure and review standard of a TenXPros Living AI Solution Dossier. “Maya R.” is fictional. No real client, person, organization, or confidential information is represented. Review labels are illustrative and do not constitute a certification decision.
The TenX Method: Frame · Design · Prove · Foresee — You bring the expertise. We bring the AI method.