Do we still need the 100-hour analyst grind if pitch decks and comps build themselves?

The shift

Producing a first-pass comp set, football field, or pitch deck goes from 20-40 hours of analyst assembly to minutes — pulling filings, building trading comps, formatting slides, and drafting the narrative are now pattern-matching and synthesis, which AI does abundantly. What doesn't move: someone accountable for the number that goes in front of a client, and the judgment to know which comp is wrong for this deal.

The axioms

• Junior hours are the cheap input that buys back senior attention — the grind exists because analyst time was scarce-but-cheaper than banker time, so you spend more of it to save some of the latter.
• The grind is the training path — reps under time pressure are how you install pattern-recognition for when a number is off, before someone touches judgment-bearing work.
• Client-facing output must carry zero errors, because a wrong multiple in the room threatens the deal and the firm's name — accountability sits on a specific person for a specific number.
• The grind is a sorting mechanism — it's a cheap, visible signal of stamina and precision that firms use to decide who gets staffed on judgment work later.
• Pyramid economics — the multi-level fee markup (MD judgment priced on top of analyst hours) only works if the base layer's time is a real, billable cost.

Invalid axioms

1. Junior hours are the cheap input that buys back senior attention. Assembling comps, building the football field, formatting the deck was scarce because it took a trained person's night. That's now abundant — a model pulls the filings, builds the grid, drafts the slide narrative in minutes. Habit-trap: staffing plans and fee models still budget 60-80 hours of analyst time per deliverable as if the assembly itself were the expensive part, when the expensive part left the building.
2. The grind is a sorting mechanism for stamina and precision. Surviving all-nighters was a visible, cheap-to-observe proxy for who could be trusted with more. When the grunt work disappears, so does the proxy — you can no longer tell who's rigorous from who just stayed up late formatting a table. Habit-trap: promotion committees still reward hours logged and decks shipped, which is now a measure of tool operation, not judgment.
3. Pyramid staffing economics. The fee model depends on analyst hours being a real cost that MDs mark up. If a comp set takes 20 minutes instead of 20 hours, the input the pyramid was priced on has collapsed — firms that keep billing (or staffing) as if it hadn't are pricing a scarcity that no longer exists.

Unchanged axioms

1. Client-facing output must carry zero errors. A model can generate a plausible comp set fast; it can't be liable when the wrong precedent transaction understates a company's value by 20% in front of the board. Someone with their name on it still has to verify every number against ground truth before it leaves the building — that verification labor doesn't disappear, it just moves later in the process and gets more concentrated.
2. Judgment on which comps are actually comparable. Screening for "same sector, similar size" is now trivial. Deciding that a comp is structurally wrong — different capital structure, one-off item distorting margins, a business mix shift the multiple doesn't capture — is judgment under ambiguity a model doesn't reliably have. This was always the scarce skill; the grind just used to be the only way to build it.
3. Trust and the standing to advise on a specific deal. Clients pay for a banker who can look them in the eye and own a recommendation, not for a document. That standing is relational and accountable, not something a faster deck-builder confers.

New axioms

1. How do you train judgment if the reps that used to build it are gone? The 100 hours weren't just output, they were the mechanism by which analysts learned what "wrong" looks like in a model — by making the mistakes themselves at 2am. If that mechanism disappears, someone has to design a replacement path to the same pattern-recognition, or the firm is promoting people who've never actually built a model from scratch.
2. Who verifies at the new volume and speed? When a junior can generate five versions of a comp set in the time it used to take to build one, the bottleneck moves entirely to review — and review capacity didn't get five times bigger. Someone has to own catching the confidently-wrong multiple before it's in a client's inbox, at a speed the old process never had to handle.
3. What replaces the grind as a sorting signal? If stamina-under-formatting-pressure no longer distinguishes candidates, firms need a new, legitimate way to identify who has the judgment to advance — and haven't built one yet.

Where it breaks

Firms are cutting analyst headcount and hours on the assumption that deck/comp assembly was the expensive part (INVALID #1) while still expecting the next cohort to arrive at VP with the same error-catching instincts the grind used to produce (NEW #1) — nobody has actually replaced the training mechanism, they just removed it. Second: staffing models are compressing junior hours because output is fast (INVALID #3), but the freed-up time isn't being redirected to review capacity (NEW #2) — the verification bottleneck is quietly absorbing the hours the deck-building used to occupy, unstaffed and unpriced.

ai-axioms.vercel.app/axioms/finance/finance-ib-analyst-grind-ai-axioms