Workflow
Review Your Pitch Claims with an Aurora Agent
Trace pitch claims to evidence, correct customer and revenue definitions, and keep proposed wording separate from founder approval.
Review pitch claims by tracing each sentence to a defined source, period, unit and owner before asking an agent to improve the wording. Aurora can help prepare that review using available materials; the founder decides which claims belong in an investor-facing deck.
The result of this workflow is a claim-to-evidence register and a corrected draft, not an investment recommendation, investor-acceptance prediction or verification certificate. The example below is fictional and has not been executed as a live Finta product test.
Separate claim review from slide design
A polished slide can still confuse signed customers with evaluations, monthly revenue with annualized revenue, or a selected interview group with a whole market. Check the statement before changing its typography. Use the existing pitch-deck guide for deck structure; this workflow owns the evidence review behind the words.
Choose the intended audience and current deck version. Freeze a review copy so an approved source does not silently change while a writer is editing the slide. Identify which material can be shared with the assistant and which needs a restricted review by the company's own specialist.
Build a claim register before drafting
| Claim ID | Exact statement | Evidence and definition | Decision owner |
|---|---|---|---|
| C01 | [Copy the actual slide sentence] | [Source, date, period, numerator/denominator, exclusions] | [Named person responsible for the fact] |
| C02 | [Copy the actual customer statement] | [Contract/payment/evaluation distinction and permitted disclosure] | [Commercial owner] |
| C03 | [Copy the forecast statement] | [Assumptions and forecast label, separate from actual results] | [Finance/founder reviewer] |
| C04 | [Copy the market statement] | [Primary evidence, geography and population; unresolved estimate limits] | [Research owner] |
Use four editorial decisions: supported as written, supported after narrowing, awaiting evidence, and remove. Do not assign a confidence percentage merely because the output looks quantitative. A missing source is a blocker, not zero revenue or proof that the opposite claim is true.
Work in reviewed passes
- Manual intake: the founder supplies the approved deck and source index, sets disclosure limits, and chooses the review questions.
- User-triggered evidence pass: ask Aurora to extract exact statements and identify support or conflicts in the material actually available. Keep inaccessible sources explicit.
- Manual fact review: source owners resolve definitions, dates and discrepancies.
- User-triggered writing pass: rewrite only accepted statements, preserving the reviewed meaning.
- Manual final comparison: compare the revised deck with the register and approve the actual version intended for external use.
These passes can be separate responsibilities in an enabled workspace or successive reviewed tasks in an ordinary conversation. The division is an operating design, not proof that agents were provisioned or that a checker has independent access to the underlying systems.
Fictional example: correct the evidence before the sentence
Cedar Loop's draft slide says, “Twelve customers and $240,000 annual revenue.” Its supplied source packet contains eight paying organizations, four unpaid evaluations, and $20,000 in revenue in one selected month. The statement combines two different counting and timing errors.
| Draft statement | What the fictional packet supports | Reviewed wording |
|---|---|---|
| Twelve customers | Eight paying organizations; four unpaid evaluations | Eight paying organizations and four unpaid evaluations, as of the source date |
| $240,000 annual revenue | $20,000 recorded for one specified month; no full-year result supplied | $20,000 revenue in the specified month; any annualized calculation must be labeled and justified separately |
| Retention is strong | No cohort definition or retention schedule supplied | Remove until a defined cohort calculation is reviewed |
| The market wants our product | Nine interviews from a selected group | Describe the nine interviews and selection limits, not marketwide demand |
The founder removes the retention sentence, narrows the customer count, and retains the actual monthly revenue. No new number is needed. The correction makes the story easier to trust because the reader can tell what happened and what remains an assumption.
Preserve the revision trail
Record the old sentence, accepted replacement, reason, source version and approving person. If a source changes after approval, reopen the affected deck claim instead of silently carrying its earlier decision forward. Keep this review bounded to the selected deck. Use the fundraising-materials consistency guide for the separate cross-document reconciliation and release-pack job.
For forecasts, distinguish an input assumption from an observed result. For customer quotes, verify permission and exact wording. For a partnership, distinguish a conversation, pilot, signed agreement and active deployment. Do not let an agent translate a promising discussion into an announced relationship.
Use a current playbook without inventing a reviewer
The Deal Room Reviewer skill template is an existing starting point for missing or inconsistent fundraising material. Inspect its actual instructions and confirm before installation. It does not replace the finance owner, counsel, an audit or the founder's final decision.
Finta's available context and tools depend on the account, organization and connections. A role name does not grant new document access. For this recipe, prepare proposed corrections only. Record changes, sharing and sending are separate actions that need the user's actual authority and the applicable product controls.
Finish with a falsifiable review receipt
- Deck version and source packet are identified.
- Each material claim has a reviewed definition or an explicit blocker.
- Numbers reconcile across every accepted output.
- Illustrations, assumptions and forecasts are labeled.
- The final human reviewer and unresolved decisions are recorded honestly.
- No claim of a live product test, investor reaction or successful fundraise is inferred from this worksheet.
Continue with the investor-update workflow when the job is recurring communication, or the accelerator-interview workflow when the job is defending a specific answer. Those are different canonical tasks.
Copyable AI practice prompt
Try this with Aurora AI in Finta, Codex, or Claude using material you are authorized to share. Tool access and available actions differ.
Review [approved deck text] against [permitted source index]. Build a register of exact claims, evidence location, date, period, unit, definition, conflict and human owner. Classify each claim as supported, narrow, awaiting evidence or remove. Do not infer missing metrics, customer status, partnerships, permissions or investor acceptance. Draft replacement wording only for supported facts and mark assumptions separately. Return the slide and claim IDs affected within this deck. Produce no record change or external message.