Wall Street Prompt · Marketing Cloud

The common mistake: treating the prompt as the whole job. Type one prompt. Get one clean answer. Ship it. The failure mode: a polished output with a stale or invented number inside it. The fix is not a better prompt. It is a system around it.

PART 1: OFFICIAL SOURCE Anchor the task to the primary document. -> 10-K, 10-Q, 8-K -> The earnings call transcript -> The official press release Not the model's memory. Not a summary of a summary. If there is no source, there is no task.

PART 2: THE PROMPT Give the model a job, not a wish. -> Tell it: extract, do not conclude -> Every figure: source + page + date -> Output: positive / negative / unclear 'Unclear' is a real answer. It goes on your follow-up list, not into your model.

PART 3: THE CHECKLIST The same checks, every single time. -> Source attached to every number -> Period and currency stated -> Units consistent across the table -> Unclear items flagged, not hidden Quality should not depend on your mood that day.

PART 4: HUMAN REVIEW The step people skip. -> Read the output against the source -> Re-check every figure the work turns on -> Decide what it means -> Sign off The model drafts. You are the one who approves it.

WHY IT MATTERS One prompt gives you an answer. The mini-system gives you an answer you can defend in the room. Same inputs. Same checks. Same standard, every time. That is the difference between a trick and a workflow.

RUN THE SYSTEM 1. Official source 2. The prompt (extract + cite) 3. The checklist (same checks) 4. Human review (you sign off) AI builds the draft. You own the judgment. Save this. Run it on your next task. Educational only. Not advice.

A prompt is not a workflow.

A single prompt gives you a single answer. It does not give you a way to trust that answer, repeat it, or defend it. That is why AI finance work breaks: people ship the output of one prompt and find out later the number was stale, the source was the model's memory, and nobody checked.

The fix is a mini-system around the prompt. Four parts:

1. Official source: anchor every task to the primary document. The filing, the transcript, the press release. Not the model's memory. 2. The prompt: extract and cite, do not conclude. Every figure comes back with a source, a page, and a date. 3. The checklist: the same checks every time, so quality does not depend on the day. Source attached, period named, units consistent, unclear items flagged. 4. Human review: you read the output against the source and sign off. The model drafts; it does not approve.

The common mistake is thinking a sharper prompt solves it. The failure mode is a clean answer with a bad number buried inside. The discipline is the system.

AI builds the draft faster. You still own the judgment and the sign-off.

Save this one and run it on your next AI finance task.

Educational content only. Not investment advice, and not a recommendation to buy, sell, or hold any security. Wall Street Prompt. Always verify against the primary source.

The Mini-System: Why a Prompt Is Not a Workflow

Slides

Caption

Design notes

CTA