Five Advisors, One Red Team, Twelve Minutes. The Council I Sell Was Built for Me First.
Listen to this article
Read by Coach Nigel Casey · 8 min read audio
Five Advisors, One Red Team, Twelve Minutes. The Council I Sell Was Built for Me First.
A real advisory board of five fractional executives plus a chairman costs 8 to 15,000 Swiss Francs per engagement and takes two to three weeks to convene. The output is five independent perspectives, an anonymised peer-review round, a chairman synthesis, and a structured deliverable you can act on.
The Strategic Council returns the same shape of output in under twelve minutes, for a few cents in inference. That sentence sounds like AI-hype if you stop reading it there. The point is what comes next.
I built it as my own decision instrument first. Wednesday's post about the Audio Purge cron is the canonical proof of what that means in practice. The cron exists because this Council surfaced an FADP exposure I was not tracking when I ran my own assessment. That is the test of an advisory council. Not whether it gives you a smart deck. Whether its output produces protective infrastructure.
Then I lifted the Council, structurally unchanged, into MyBusinessAccelerator.io as a paid product. Sell the council you use yourself is my motto.
What the Council actually is.
Eight agents, working in two rounds, against a strict JSON schema.
The Council Director orchestrates everything. It receives the question, sets the round structure, runs the schema gate that rejects malformed output, and returns the chairman synthesis at the end. The Director is the only agent that sees the whole flow.
Five independent advisors run in the first round, each in isolation, each with a calibrated system prompt and a small list of banned phrases that signal generic AI consulting language. The five are: the Operator on execution, the Growth Advisor on distribution and reach, the Margin Advisor on unit economics, the Red Team whose schema-level mandate is to disagree with the others, and the Capacity Advisor on what the founder can actually run.
In the second round, Peer Review takes the five outputs, anonymises them as A through E, and routes each advisor four critiques of work they did not write and do not know the source of. This is the part most AI-advisory tools skip. It is also the part that produces the highest-signal disagreements.
The Chairmam then reads the original advice, the peer critiques, and the schema-validated structure, and produces the final dashboard payload. One JSON object, schema-validated, ready to render on the prospect's screen or feed into my own decision dashboard.
Eight minutes from question to dashboard, in production. Twelve minutes is the conservative number with retries.
The constraint that shaped it.
The constraint was honesty. Most AI advisory tools converge on agreement. Five LLM calls with similar system prompts will produce five paraphrases of the same answer, and the user reads it as consensus rather than artefact. You leave with the comforting feeling that the AI agrees with you, and you have learned nothing.
A real advisory board does not converge by default. The growth person and the margin person disagree because the things they care about disagree. The operator disagrees with both because someone has to actually do the work. The capacity person disagrees with all three because the founder cannot do everything. Disagreement is the signal. Convergence is the noise.
So the Council had to be designed against convergence, not for it. That is the whole architecture.
The decision.
Three structural choices that prevent the Council from collapsing into AI-soup agreement.
First, the Red Team is schema-mandated to disagree. Its output schema fails validation if it agrees with all four other advisors. The Director re-runs it until the disagreement is structural, not cosmetic. This is enforced at the schema level, not the prompt level, because prompts get politely ignored and schemas do not.
Second, peer review happens on anonymised work. Each advisor sees four others labelled A, B, C, D and writes a critique without knowing whose work they are reading. This breaks the AI tendency to defer to whichever output looks most authoritative, because authority cues have been stripped. The critique is on substance because nothing else is available.
Third, every output is schema-validated. The Chairman's final payload has to populate a fixed shape: dashboard headline, sector benchmark, SWOT, KPIs, ROI projection, roadmap, list of structural risks. If the Chairman returns malformed JSON, the Director reruns the whole final synthesis. Schema is the only gate that does not care how persuasive the agent's voice is.
The alternative I rejected was a single mega-prompt that asks one LLM call, to produce all five perspectives at once. Faster, cheaper, lower latency. Also a different product. A single LLM running a five-perspective format is not five perspectives. It is one perspective wearing five hats. The Council uses five separate calls because the disagreement has to be real, not stylistic.
What it produces.
The output is a JSON dashboard payload that drops into a templated dashboard for the prospect, and a parallel internal payload that feeds into my own decision dashboard. Same Council. Different rendering.
For the prospect on MyBusinessAccelerator.io, the dashboard renders their personalised analysis with their brand colours pulled from their site, sector benchmarks, SWOT with concrete actions, KPIs with target ranges, an ROI projection range, and a 30/60/90 roadmap. Eight minutes from quiz submission to a fully generated dashboard.
For me, on the founder dashboard, the same Council writes into a different template: structural risks the operation is not currently tracking, dependencies between agents and runtimes, the next three architectural decisions on the critical path, the things the Capacity Advisor says I cannot afford to commit to, this quarter.
The Audio Purge cron came out of one of those internal runs. The Council surfaced the FADP exposure as a structural risk; within days the cron was production code; the discharge of the legal obligation now happens nightly without my attention. That is the canonical example of the test I keep coming back to. The Council surfaces the absence. The code closes the gap.
What it cost.
Six weeks of structural design before the first real output was usable. Most of that time was on the schema, the banned-phrase lists for each advisor, the calibration of how harshly the Red Team should disagree, and the Peer Review round mechanics. The rest was on the Chairman synthesis, which is the hardest agent in the system because it has to honour contradictions without resolving them prematurely.
The cost per run is a few cents in inference. The cost per redesign cycle is much higher than that, because the schema is load-bearing across the whole system. Changing the dashboard payload shape means changing the Chairman, which means revalidating the Peer Review outputs, which means re-tuning the five advisors. The schema is not a JSON detail; it is the contract that everything else hangs from.
The trade-off I underestimated was that the Council changed how I make decisions about EFO. Before the Council, I was making founder-only decisions with founder-only blind spots. After the Council, I run anything material through it before committing. That is not a productivity gain. That is a different quality of decision.
Why this is in the report.
The Strategic Council eight agents are documented in Section 04 of the AI team report at `nigelcasey.com/agent-team-report.html`. They are cross-referenced in Department 2.1 (Advisory Council, productised on MyBusinessAccelerator.io) and Department 3.1 (Strategic Counsel, the private mirror I run on EFO). Same eight agents in three places. Counted once at primary location.
The principle that runs across the three deployments is simple. Sell the council you use yourself. Do not productise something you would not personally route a real decision through first. The MBA Funnel and Dashboard Generator (next post in this series, week 2) is what makes this Council into a paid product for SME founders. The Audio Purge cron from Wednesday is what proves it works on me first.
The test of an AI advisory council is not whether it gives you a smart deck. It is whether its output produces protective infrastructure that your business has to live by, every night, after the deck is forgotten.
A real advisory board of five fractional executives plus a chairman costs between 8 to 15,000 Swiss francs per engagement and takes two to three weeks. The Strategic Council returns the same shape of output in under twelve minutes for a few cents, but only because three structural choices prevent it from collapsing into AI-soup agreement: a Red Team schema-mandated to disagree, peer review on anonymised work, and full schema validation of every output. I built it as my own decision instrument first; the Audio Purge cron from Wednesday is the canonical proof that it produces protective infrastructure rather than smart-sounding decks. Then I lifted it, structurally unchanged, into MyBusinessAccelerator.io. I sell the council I use myself.
If you're running an SME and any of this looks like work you should be doing, that is the side of things I help with.
Language Analysis
Select a category above to highlight those words in the text.
Learning Materials
Key Vocabulary
advisory board
A group of senior advisors who provide structured external perspective to a company, typically meeting periodically and producing a structured deliverable.
“A real advisory board of five fractional executives plus a chairman costs CHF 8,000 to 15,000 per engagement.”
engagement
In professional services, a single defined piece of work delivered for a client; the unit of billing and scope.
“The Strategic Council delivers the same shape of output as a multi-week consulting engagement, in twelve minutes.”
to convene
To bring people together for a meeting; to assemble formally.
“A real advisory board takes two to three weeks to convene.”
chairman
The person who leads a board or council and is responsible for the final synthesis of its discussions.
“The Chairman synthesises the contradictions into a final dashboard payload.”
inference
In machine learning, the process of running a trained model to produce output; the runtime cost of using an LLM.
“The Strategic Council runs in twelve minutes for a few cents in inference.”
to converge
To come together at a single point; in argument, to arrive at agreement, sometimes prematurely.
“Most AI advisory tools converge on agreement; this Council is designed against that.”
consensus
General agreement among a group; sometimes the absence of dissent rather than the presence of conviction.
“The user reads the AI output as consensus rather than artefact.”
artefact
In research and engineering, an outcome produced by the process itself rather than by the underlying reality being measured.
“Five paraphrases of the same answer is an artefact, not a consensus.”
to mandate
To require formally; here, to enforce at the schema level rather than as a prompt suggestion.
“The Red Team is schema-mandated to disagree.”
peer review
A process in which work is critiqued by others of comparable standing, often anonymously, to improve quality through structured disagreement.
“Peer review on anonymised work breaks the AI tendency to defer to authority.”
to anonymise
To remove identifying information from a piece of content so that it can be assessed on its substance alone.
“Each advisor sees four others labelled A through E without knowing whose work they read.”
payload
In computing, the actual data being transmitted in a request or response, as opposed to the metadata around it; here, the structured JSON output of the Council.
“The Chairman produces the final dashboard payload as a single JSON object.”
to validate (against a schema)
To check that data conforms to a defined structural specification; if it does not, to reject it.
“Every output is schema-validated against a fixed shape.”
to honour (a contradiction)
To acknowledge and preserve a difference rather than smoothing it over; in synthesis, to keep contradictions visible to the reader.
“The Chairman has to honour contradictions without resolving them prematurely.”
Grammar Notes
Two-clause closing aphorism with chained nouns (X is the signal. Y is the noise.)
Two short sentences, identical grammatical shape, opposite content. The first asserts what carries information; the second asserts what does not. The pair compresses an entire argument about AI-advisory design into ten words. It is a Nigel-voice signature for closing a section: state two equivalences and let the reader complete the inference themselves.
“Disagreement is the signal. Convergence is the noise.”
Common mistake: Compressing into one sentence (Disagreement is the signal, while convergence is just noise) is grammatically valid and rhetorically weaker. The two-step construction makes the contrast carry weight.
Conditional fragment for rhetorical setup (X if you stop reading there)
A subordinate conditional clause appended to a complete main clause, where the conditional pivots the reader from a likely first reaction to the actual point. The grammar is a standard if-clause; the rhetorical move is to acknowledge the obvious objection inside the same sentence and immediately defuse it.
“That sentence sounds like AI-hype if you stop reading it there.”
Common mistake: Splitting into two sentences (That sentence sounds like AI-hype. But the point is what comes next.) is grammatically fine and rhetorically less efficient. The conditional fragment compresses the acknowledgment and the pivot into one move.
Compound modifier with hyphen (schema-mandated, schema-validated, schema-level)
A noun joined to a past participle (or to another noun used adjectivally) by a hyphen functions as a single compound adjective. Schema-mandated, schema-validated, and schema-level all signal that the discipline is enforced by the structural contract, not by instructions. The hyphenation matters because it tells the reader to read the two words as a single concept.
“the Red Team is schema-mandated to disagree”
Common mistake: Dropping the hyphen (schema mandated, schema validated) is increasingly common in tech writing and increasingly ambiguous. The hyphen is the signal that this is one concept, not two.
Imperative aphorism for closing positioning (Sell the X you Y yourself.)
A short imperative sentence with an embedded reflexive clause. The reflexive yourself is doing the load-bearing work: it asserts not just that the seller has used the product, but that they continue to use it on their own decisions. As a positioning principle for consulting and productised expertise, it compresses an ethical commitment into one sentence.
“Sell the council you use yourself.”
Common mistake: Adding hedges (You should ideally try to sell the council you use yourself when possible) destroys the conviction. The imperative is the device; softening it makes it advice rather than a principle.
Comprehension Questions
- 1.Compare a traditional advisory board engagement (cost, duration, output) with the Strategic Council described in the post. What does the Council preserve from the traditional shape, and what does it change?
- 2.What are the three structural choices that prevent the Council from collapsing into AI-soup agreement? Why is each one enforced at the schema level rather than the prompt level?
- 3.Why does the post argue that the Council had to be built as the founder own decision instrument before it was productised on MyBusinessAccelerator.io? What does the Audio Purge cron from Wednesday post demonstrate about that sequencing?
- 4.What does it mean that the Council was lifted, structurally unchanged, into the productised version? Why is structural sameness the load-bearing claim, rather than visual or feature sameness?
- 5.Apply: in your own organisation or industry, where do important decisions currently get made by a single person with single-person blind spots? What kind of structured disagreement (a Red Team mandated to dissent, an anonymised peer review, a schema-validated output) would change the quality of those decisions, and what would it cost to put that structure in place?
Run your own diagnostic
Use the same Strategic Council I run my own decisions through. The assessment preview is free. The specific central human intelligence it is based on is verified in person during the call.
Start the free diagnostic →