Evidence in. Decisions out.
A2V is built on a small number of disciplines applied without exception: real instances instead of workshop opinions, evidence graded for confidence, at most three constraints, and no engagement that ends without a decision taken or explicitly declined.
Operating principles
Eight principles govern every engagement.
Outcome-first
Nothing starts until a business outcome is named, owned and sized. AI is the means, never the topic.
Jointly owned
You own outcomes, sponsorship, data, decisions and adoption. We own method, challenge, evidence discipline and synthesis.
Time-boxed by design
Every stage has a hard edge, and we're explicit about what a time-box cannot credibly achieve.
Evidence-led
Findings are cited, triangulated and graded. Weak evidence is declared weak, not hardened for a cleaner story.
Fluency-informed
Capability is role-based and built in the flow of real work against named use cases — never generic training days.
Governed for safe progress
Risk is screened every engagement and calibrated to appetite, not fear — control without paralysis.
Built to scale
Prerequisites are surfaced as funded roadmap items, so what works once can work routinely.
No theatre
Activity that looks like progress but moves nothing is named openly — including when it's ours to name.
The nine flows
Value leaks where flows cross functional boundaries.
Nine flows carry AI ambition to business value. Each is examined by walking one real instance end to end with the people who touch it — not by holding a workshop about it. Flows break at the places no single function owns; that is where we look.
F1Use-case flowidea → measured value or kill
How an idea travels from someone's head to a measured result — or to a deliberate stop. Both endings are healthy; the quiet middle is not.
Ideas have entry criteria and decision dates; some get killed, on purpose, with the learning kept.
Pilots that never end; selection by enthusiasm; no initiative has ever exited.
F2Decision flowleadership intent → frontline behaviour
Whether a leadership decision survives translation down through management to what people actually do on a busy afternoon.
Intent arrives recognisably; managers adapt it knowingly rather than losing it.
Distortion at every level; the frontline is unaware a mandate exists; compliance is performed, not practised.
F3Data flowsource systems → AI-consumed data
How data reaches the AI that uses it: through an approved, curated path with known provenance — or not.
Access is approved by someone identifiable; content is curated; provenance is known.
Tools pointed at uncurated document estates; nobody can say who approves an AI system's access to a dataset.
F4Governance flowpolicy → operating control
Whether written rules become controls that people know and that demonstrably operate.
Frontline staff can state the rules that apply to them; controls have been tested and exceptions have a route.
A policy nobody interviewed can recall; controls asserted but never tested.
F5Capability flowtraining → confident, competent use
Whether training investment becomes visible skill in real work — with managers redesigning how their teams work, not just attending the same webinar.
Role-relevant capability shows up in the work itself; managers set norms for their teams.
High completion rates, low real usage; capability concentrated in a small enthusiast minority.
F6Risk flowunmanaged experimentation → controlled scaling
How the organisation moves from quiet, unmanaged tool use to controlled scaling — which requires people to feel safe disclosing what they actually use.
Disclosure is safe; the inventory of tools in use grows because people volunteer it; risk appetite is applied, not implied.
Widespread unofficial tool use nobody will admit to; an official inventory that is quietly fictional.
F7Benefits flowproductivity claim → bankable value
Whether claimed value survives contact with evidence: a baseline taken before deployment, and a review by someone with no stake in the answer.
Baseline-first is the rule; benefits are independently reviewed; resources get reallocated on the results.
Claims with no baseline; vanity metrics; projected benefits reported as achieved.
F8Adoption flowpilot → embedded default behaviour
Whether a new way of working survives real pressure and becomes the default — or quietly reverts the moment a deadline bites.
The new way holds under load; norms sustain themselves without a champion pushing.
Reversion under deadline; workaround spreadsheets; adoption that lives and dies with one enthusiast.
F9Trust flowuncertainty → sustained calibrated use
How people calibrate their trust in AI output: checking proportionately, and disclosing errors safely — neither blind trust nor total re-work.
Checking effort matches the stakes; wrong outputs get surfaced and change something.
Re-checking so heavy it cancels the gains — or concealed reliance nobody verifies at all.
Evidence discipline
Three sources that can't all lie the same way.
Every finding is triangulated across what people say (interviews), what systems record (telemetry, licence logs, registers), and what we see (observing real work). Where the three diverge, the divergence is usually the most valuable finding in the engagement.
The rule of the floor
Where frontline evidence and leadership perception disagree, we score toward the floor — and report the perception gap as a finding in its own right.
Confidence, graded
Every score and every figure carries a confidence grade — A triangulated, B partial, C thin. C-grade findings are presented as hypotheses, with what would confirm or kill them.
Ranges, never points
Value is expressed in ranges with the arithmetic shown. False precision is a form of dishonesty this method refuses.
Absence is evidence
The register that doesn't exist, the baseline nobody took, the kill decision never made — what's missing is reported as plainly as what's found.
The So-What Chain
Every finding must earn its place.
A finding only reaches your executive session if it completes a five-step chain. Findings that can't complete it go to the appendix — however interesting they are.
Finding
Something evidenced, cited and triangulated — not an opinion in a suit.
Flow blocked
Which of the nine flows it breaks, and where.
Value capped
What the blockage plausibly costs, as a range with graded confidence.
Constraint
Whether this is one of the few binding constraints — at most three per engagement.
Decision required
The specific choice, and whose it is to make.
The No-Theatre screen
Eight ways activity impersonates progress.
AI theatre is activity that looks like progress while moving nothing that matters. A2V screens for eight forms of it, openly — it is a diagnostic instrument, not a slogan. Naming the pattern kindly, with evidence, is often the moment an engagement earns its fee.
Maturity theatre
Scores presented, nothing owned — measurement standing in for movement.
Pilot theatre
Pilots that never exit; demo-live confused with operations-live.
Innovation theatre
Showcases, labs and press releases while the P&L stays untouched.
Governance theatre
Policies and committees with no recall on the ground — or no appetite to act at all.
Demo theatre
Curated-data wins presented as proof of production value.
Training theatre
Completion percentages presented as capability.
Strategy theatre
Ambition slides with no numbers and no owners.
Benefits theatre
Projected benefits tracked as if already achieved.
The exit artefact
The Commitment Log, not a report.
An A2V engagement ends in a Commitment Log populated live by your executive team: each entry names the outcome committed, its baseline today, the target and date, the owner, and the evidence that will count as done. Declines are recorded as decisions — proudly. An engagement that ends with neither a commitment nor an explicit decline has failed its own standard.
The client owns
Outcomes · sponsorship · access · data · decisions · adoption · benefits evidence
We own
Method · challenge · evidence discipline · synthesis · calibration · risk visibility · quality gates · roadmap structure
This ownership strip is printed on every artefact an engagement produces. When a client asks us to take one of their items — "you decide which use case" — we hand it back: that's a decision only you can own; our job is to make it an easy one.
The difference
Why this is not a maturity assessment.
A maturity assessment describes your state and leaves. A2V walks your real flows, prices your constraints in your money with the arithmetic shown, carries a stop-clause on its own fee as a design principle, and doesn't finish until your executives have committed — with baselines, owners and dates — or explicitly declined. Internal calibration instruments exist behind the method, but you are never sold a score: scores describe, decisions move.