The red pen is the new production line
Session a3 built the rails; this session builds the craft that runs between them. The old economics hid a truth: review was never really optional, it was just cheap to skip, because expensive drafts arrived pre-filtered by the effort that made them. That filter is gone. What arrives now is fluent, plausible, fast, and unfiltered - and the partner who reads it is the only quality gate left standing. The good news: review is a teachable craft with a structure, not a heroic act of attention. That structure is the pyramid, and you will run it on a real page before you leave.
The bottleneck moved 7 min live
The BCG field study's production numbers - work completed 25.1% faster with GPT-4 - have a second-order consequence nobody prices in on day one: if drafting accelerates and review does not, review capacity, not drafting capacity, now gates the firm's throughput. The queue did not disappear. It moved to the partner's desk.
LiveFluent volume: the new failure mode4 min▶
The old failure mode announced itself: a weak draft read weak, and a rushed analysis looked rushed. The new one does not. Fluent volume is more pages, all plausible, arriving faster than scrutiny scales - every paragraph polished, every error wearing the same confident prose as every fact. The danger is not that the work got worse; it is that bad work stopped looking different from good work at skim speed. Skimming was a viable review strategy precisely when effort correlated with quality. That correlation is gone.
LiveThe second warning: homogenization3 min▶
The study adds a finding that turns review from a defensive act into an offensive one: AI-assisted answers converge on similar ideas. Un-reviewed work is not just risky - it sounds like every rival's un-reviewed work, because two firms pasting the same prompt get cousins of the same deck.
Review is therefore doing double duty now: catching what is wrong, and adding what only your firm would say. The first duty protects the engagement; the second protects the franchise. Part 3 turns that second duty into a bar the whole practice can hold.
Self-studySizing your own review gap2 min▶
A quick self-audit for your practice: how many deliverable pages reached clients last month, and how many received a genuinely slow read - facts traced, logic pressed - before shipping? The ratio is your exposure. If drafting volume grows at AI speed and slow-read capacity stays flat, the ratio degrades every month without anyone deciding anything. The fixes are structural, not motivational: review time on the engagement plan as a line item, passes with named owners (Part 2), and rituals that scale scrutiny (the self-study card after the pyramid).
The review pyramid 8 min live
Four review passes, run cheapest first, each catching a failure the others miss. This is field practice, labeled honestly - the structure working reviewers converge on, not a certified standard. The order matters: a fact error found in pass 1 saves the partner-hour that pass 3 costs, and a page that fails pass 2 is not worth pass 3's attention yet.
LiveFour passes, four failures, four owners5 min▶
| Pass | Question | What it catches | Who runs it |
|---|---|---|---|
| 1 · FACTS | Does every number, citation, and quote trace to its source? | The Deloitte-shaped failure: fabricated references, misattributed quotes, invented statistics - each fact gets a named owner | The drafter, then a second pair of eyes |
| 2 · SPECIFICITY | Is this OUR client, or any client? | Slop - fluent text with zero client anchors; the consultant track's scorer prices exactly this: its AI-written slop draft scores 8 of 100 because anchors are zero | Engagement manager |
| 3 · JUDGMENT | Would we sign this recommendation with our name? Does the logic survive a hostile read? | Plausible reasoning that collapses under one sharp client question | Partner |
| 4 · VOICE | Does it sound like the firm, not like a model's average? | Homogenization - the deck any rival could have shipped | Partner, reading aloud |
Passes 1 and 2 are delegable and cheap - juniors can run them with a checklist, and the consultant track trains them to. Passes 3 and 4 are why partners exist. A practice that lets partners burn their hours on pass-1 work has its pyramid upside down.
Self-studyReview rituals that scale3 min▶
Three rituals turn the pyramid from a heroic habit into practice infrastructure. The red-team hour: before any major readout, one hour where a colleague not on the engagement attacks the deck as the client's most hostile reader - pass 3, institutionalized. The second-partner rule: high-stakes decks get a second partner's pass 3 and 4, on the calendar, not as a favor. The read-aloud test: the fastest pass-4 instrument ever devised - sentences a partner would never say out loud have no business shipping under the partner's name. None of these require new tools; all of them require the engagement plan to budget review as work, which is the leader's move.
The anti-slop bar as brand 5 min live
Your clients now read AI-fluent text all day - in their inboxes, their vendors' proposals, their own teams' memos. Fluency has stopped signaling effort, and generic competence has stopped signaling anything. What the field study says survives is exactly what homogenized output cannot fake: work that is unmistakably specific and judgment-dense. That is not a style preference. It is the moat.
LiveSetting the bar, out loud4 min▶
The bar, stated as policy: nothing generic ships under the firm's name. Every page carries at least one thing only your firm, on this engagement, for this client, could have written - their number, their constraint, their politics, your judgment call. The study's homogenization finding makes the pure-AI baseline everyone's baseline; a deliverable at that baseline is priced accordingly, which is session a2's lesson arriving from the other direction.
And the bar is set in review, not in a memo. The leader models it by catching slop out loud, kindly - "this paragraph could be about any company; give me the sentence only we could write" - in front of the team. Said kindly, it teaches the whole room in ten seconds what a quality bar sounds like. Said never, the bar does not exist, whatever the memo claims.
Self-studyWhere the bar meets the invoice2 min▶
Session a2 asked what happens to pricing when clients know AI did the drafting; this session supplies the answer's other half. What a client cannot get from a model at any price is exactly what the pyramid's upper passes produce: facts a named human stands behind, sentences only your firm could write, and a recommendation a partner will sign. The review discipline is not overhead on the fee - increasingly, it IS the fee. A practice that can show its review structure in a procurement conversation is answering "you used AI, why am I paying?" before it gets asked.
Run the pyramid on one page 🖊 ★ 20 min · everyone works
No lab widget on the leader track - a real page and a red pen are the exercise. Take one page of a recent deliverable from your practice; if nothing is at hand, use the course's Cadence engagement facts as a stand-in exec summary and audit that.
Pick one real page. A recent exec summary, a key slide's narrative, one section of a report. One page only - the point is depth per page, and discovering what a slow read actually costs and catches.
Pass 1 - facts. Mark every number, citation, and quote: source seen, or source absent? No judgment yet, just marks. Count the absences - each one is a Deloitte-shaped exposure with your name on it.
Pass 2 - specificity. Circle every sentence that could apply to any company in the industry. Be ruthless: if swapping the client's name for a rival's changes nothing, circle it.
Pass 3 - judgment. Write, in the margin, the one hostile question this page cannot answer. Every page has one. If you cannot find it, you are reading as the author, not the client.
Pass 4 - voice. Read the page aloud. Mark the sentence you would never say to this client across a table. That sentence is a model's average wearing your letterhead.
Tally which pass found the most. That is your practice's weak gate - the pass to staff, ritualize, or train against first. Bring the tally to session a5; the operating model builds on it.
Homework 📝 ◐ 25 minutes
- Run the pyramid on a second page from a different engagement or author, and compare weak gates. One page is an anecdote; two pages is the beginning of a pattern worth acting on.
- Put review on one live engagement plan as a named line item - which passes, whose hours, before which milestone. If it is not on the plan, it is the thing that gets skipped at AI speed.
- Catch one piece of slop out loud, kindly, this week - in a review, name the generic paragraph and ask for the sentence only your firm could write. The bar exists the first time the team hears it.
Three questions before you go 🎯 ◐ 90 seconds
1 · With AI in the workflow, the firm's production bottleneck moved...
The study's speed gain means pages arrive faster than scrutiny scales. The queue did not disappear; it moved to the partner's desk - and the new failure mode is fluent volume: more pages, all plausible, none pre-filtered by effort.
2 · The four review passes, in order, are...
Cheapest first, each catching a different failure: facts (the Deloitte-shaped check), specificity (our client or any client?), judgment (would we sign it?), voice (the firm, not a model's average). A fact error caught in pass 1 saves the partner-hour pass 3 costs.
3 · Specificity is a quality gate, not a style preference, because...
Homogenized output converges on what any firm could say, and it reads fluently while doing so. Client anchors are the measurable difference between a deliverable and a confident blog post - which is why the scorer prices them, and why pass 2 circles every sentence that could belong to any company.
What this session covered from the official universe
Certificates, videos, and graded assessments stay with the official providers - this course teaches the working core.