For PE deal teams and IC members handed a DCF to trust: 11 Claude prompts that rank every input by leverage, name the one assumption carrying the valuation, run the bear case, and hand you the cell to attack before the vote. On your own data, inside Claude.
Free — runs in your own ClaudeMedium setup · 4 steps11 ready-to-run prompts
Three minutes, four steps, nothing to install by hand
Claude sets it up for you. You just paste.
Never used Claude? It is free and takes 30 seconds to open. Copy the instruction below, paste it into Claude, and it reads this page and walks you through everything, one question at a time.
1
Tell Claude how to talk to you
One tap. It changes how much Claude explains, and how slowly it goes. You can change it any time.
2
Copy your setup instruction
A short instruction plus a link to this page lands on your clipboard. First copy asks for your email once. That unlocks every button across the whole library.
3
Open Claude in a new tab
Free account, no card, 30 seconds. This tab stays open so you can come back.
Claude reads this page, asks one question about your work, then guides you step by step until your first output is right. If anything looks wrong, tell Claude what you see, and it fixes it with you.
▸Prefer the full prompt instead of the link? (optional)
I am comfortable copy-pasting and following instructions, but I am not a developer.
There is nothing to install for this one and no commands to type: it all happens inside Claude. If any instruction below implies a Terminal, translate it into the equivalent click path for me instead.
- Plain English. Define jargon the first time it appears.
- One step at a time, then wait for me to confirm before the next one.
- Tell me what success looks like at each step, and diagnose any error before moving on.
Follow the instructions below with those rules applied.
If you can browse the web, open and read this page in full first — it has the complete guide and every prompt you will run (the vault is under the-vault anchor): https://consultance.ai/library/dcf-attack-kit#the-vault . If you cannot open links, tell me and I will paste the page in — do not guess the prompts.
I want to set up the DCF Attack Kit inside my own Claude and run it on a real DCF this week. Walk me through it one step at a time, wait for me to confirm each step before the next, and define any term the first time you use it. Treat me like a PE deal lead, IC member, or CFO who reviews valuations but has never wired up a Claude Project. This is NOT a Terminal install. There is nothing to install from a command line. It is all done by clicking inside Claude and pasting prompts. If any step needs a command window, you have misread this, so do not send me one. Do not tell me this is "not possible" inside Claude, because it is, so if you hit a wall, ask me a question instead.
One honest thing before we start, so you know the ceiling: this vault is about 60 percent of the build. It gets a real DCF onto the table, finds the assumption doing the lifting, runs the bear case, and writes the one page attack memo. The last 40 percent, wiring live data into your model and data room, role based permissions so the right people see the right cells, an audit trail your reviewer will accept, and running this across many deals at once, is implementation work consultance wires into production with you. That is the honest ceiling of any prompt bundle. When you want that part, reply "wire it" for a 30 minute slot.
Ask me this ONE question first, then stop and wait:
**Where does the DCF you want to attack live right now?**
- (A) An Excel or Google Sheets model somebody sent me
- (B) A PDF or a CIM with the DCF and its outputs inside it
- (C) A data room I would have to export the model from
- (D) I only have the headline number and a few assumptions, I will paste them raw
Once I answer, walk me through the setup in this order, one step at a time:
1. **Open Claude and pin the model.** Tell me to go to claude.ai (or the desktop app), start a new chat, and set the model to Claude Opus 5 for the whole session. Define: pinning the model means picking Claude Opus 5 from the model dropdown so every prompt in this session uses the same one and the numbers do not drift.
2. **Create a private Project.** Click "Projects" in the left sidebar, then "New project", name it something like "DCF Attack — [target]". Explain why: a Project keeps my data in my own tenant. Nothing I load is uploaded to consultance.ai, stored by us, or seen by us. That is the whole reason this is safe to run on a live deal.
3. **Load the DCF the way that fits my answer above.** If (A) or (B), tell me exactly how to upload the file into the Project knowledge, or how to use the Claude Excel add-in to read the workbook in place. If (C), tell me to export first and then upload. If (D), tell me to have the headline EV, the WACC, the terminal method and input, and the forecast horizon ready to paste when prompt 01 asks.
4. **Paste prompt 01, the setup router.** It will ask me what type of DCF I am attacking and where my data lives, then confirm the setup back to me. Define: the router is the first prompt, it sets the whole session up so every later prompt works off the same setup. Tell me not to run anything else until 01 confirms.
Then, once setup is done, run the first-session drill on the highest-value prompts, one at a time:
- **Prompt 02 (rebuild the model).** Tell me what good output looks like: a clean driver table and a UFCF bridge that TIES to the source model. If it does not tie, it stops and lists the lines that do not, which is correct, not a failure. Do not trust anything downstream until 02 ties.
- **Prompt 03 (rank inputs by leverage) and 04 (name the load-bearing assumption).** This is the heart of the kit. Good output is a leverage table sorted by how much each input moves the answer, then ONE named assumption the whole valuation rests on. If it hedges into three equal candidates, tell it to pick one and defend the pick.
- **Prompt 07 (the reconciliation gate).** Run this before the memo. It re-derives the value a second, independent way and prints RECONCILED or NOT RECONCILED. If it prints NOT RECONCILED, the model is wrong until proven otherwise, so do not write the memo yet. Define: reconciliation here means getting to the number a different way and checking the two agree.
- **Prompt 09 (the attack memo).** Good output is one page a partner can act on: the number, the assumption doing the lifting, the bear case, the one or two cells to attack, and the reconciliation status. If it buries the point under caveats, tell it to lead with the cell to attack.
Bonus path, never required: the library page and the bonus-extras file (reverse DCF, WACC forensics, the assumption log) are there when I want them, but I do not need any repo, any install, or any third-party tool to run the vault. Everything runs inside my own Claude.
Before I trust any output: check that every figure traces to a source cell or a labeled assumption, that prompt 07 printed RECONCILED, and that a named human (me or my reviewer) signs off the load-bearing assumption before the number goes to the IC.
Step 2 · run it on your data
Step 1 set it up. These 11 prompts do the work.
the vault
The 11 prompts
Grab the whole pack as one file, or tap any prompt below to copy it on its own. Placeholders that look like {{THIS}} get swapped for your own numbers — and if you ran Step 1, Claude fills them in for you.
One .md file · all 11 prompts, numbered, in order · nothing left out.
<role>
You are a valuation desk lead who has killed more bad DCFs than you have built. Convene a standing board of four named advisors and hold it for the whole session: a sell-side M&A associate who knows exactly which cell to nudge to land the number the deal needs, a buy-side investment committee member who kills weak assumptions for a living, a Big 4 valuation partner who signs fair-value opinions under ASC 820 and IFRS 13, and a CFO who has to defend the number to a board. Surface their disagreement, do not smooth it over.
</role>
<task>
Do NOT analyze anything yet. First set up the engagement. Ask me the questions below, wait for my answers, then confirm the setup back to me in a short block before we start.
1. TYPE: (A) a sell-side DCF in a CIM I received as a buyer, (B) an internal DCF my own associate built, (C) a fairness or valuation opinion I have to sign or challenge, (D) a DCF inside a broader IC pack I am reviewing before a vote.
2. DATA SOURCE: (A) upload the model and support into this Project knowledge, (B) I will paste the raw assumptions and outputs, (C) Claude add-in reading an Excel workbook in place, (D) a governed connector to a data room (name it, I will hedge if unverified), (E) a mix.
3. TOKENS to hold: {{TARGET}}, {{HEADLINE_EV}}, {{WACC}}, {{TERMINAL_METHOD}}, {{TERMINAL_INPUT}}, {{FORECAST_YEARS}}, {{SECTOR}}, {{DECISION}}.
4. OUTPUT BAR: every figure traces to a source cell or a labeled assumption, every assumption is labeled, unverifiable inputs are flagged not filled, and you attack the model as a skeptic, you do not defend it.
</task>
<output_format>A short SETUP CONFIRMED block: TYPE, DATA SOURCE, the eight tokens, the output bar in one line. Then stop and wait for me to say go.</output_format>
<role>You are the sell-side M&A associate from prompt 01, rebuilding the model cleanly so nothing hides.</role>
<task>Work from the data source selected in prompt 01. Reconstruct {{TARGET}}'s DCF as a transparent chain: revenue build by driver, operating margin path, the unlevered free cash flow bridge, the WACC build with each input, terminal value by {{TERMINAL_METHOD}} using {{TERMINAL_INPUT}}, discounting, and the walk from EV to equity value. Tie your rebuilt {{HEADLINE_EV}} to the source model. If it does not tie within a small tolerance, STOP and list the lines that do not.</task>
<output_format>A clean driver table, the UFCF bridge per forecast year, the WACC build with labeled inputs, and the terminal value calc. One line: TIES to the source model, or the lines that do not.</output_format>
<constraints>Use the data source from prompt 01. Every number traces to a source cell or a labeled assumption. Do not invent a plug to force a tie.</constraints>
<role>You are the buy-side IC member from prompt 01. You care which input is carrying the answer, not what the model concludes.</role>
<task>Using the rebuilt model from prompt 02, run a one-at-a-time sensitivity on every material input: revenue growth by year, {{TERMINAL_INPUT}}, {{WACC}} and each component, operating margin, tax rate, capex intensity, working capital. Flex each by a defensible range for {{SECTOR}} and record how much {{HEADLINE_EV}} moves in absolute and percentage terms. Rank most to least impactful.</task>
<output_format>A leverage table sorted by impact: input, base value, low and high tested, EV at each, percentage swing. State the top three doing the lifting at the bottom.</output_format>
<constraints>Use the data source and rebuilt model from prompts 01 and 02. Flex one input at a time. Ranges must be defensible for {{SECTOR}}, labeled as assumptions.</constraints>
<role>You are the valuation desk lead. You reduce the argument to the single assumption that, if wrong, breaks the number.</role>
<task>From the leverage table in prompt 03, name the ONE assumption the valuation most rests on. Interrogate it: what is the value and who set it (seller, associate, or convention); what would an operator in {{SECTOR}} say if told it was baked in for {{FORECAST_YEARS}} years; what is the honest defensible range and where does the model sit in it; what does {{HEADLINE_EV}} become at the conservative end.</task>
<output_format>THE LOAD BEARING ASSUMPTION named in one line. Then who set it, the honest range, where the model sits, and the EV at the conservative end. One sentence on why this is the cell to attack first.</output_format>
<constraints>Use the data source and prior prompts. Make the call, do not hedge into three equal candidates. Pick the one and defend the pick.</constraints>
<role>You are the CFO who has watched terminal value quietly become the whole story more than once.</role>
<task>Isolate terminal value in {{TARGET}}'s model. Report it as a percentage of enterprise value. If {{TERMINAL_METHOD}} is perpetuity growth, compare {{TERMINAL_INPUT}} to long run nominal GDP and a mature {{SECTOR}} rate, and flag hard if it exceeds the long run economy. If it is an exit multiple, compare it to where {{SECTOR}} trades and transacts and check the implied perpetuity growth backed out of it. Cross check the two methods against each other.</task>
<output_format>Terminal value as a share of EV, the comparison to the economy or market, a verdict (defensible, stretched, or a lie sitting at the top), and the EV at a defensible terminal level.</output_format>
<constraints>Use the data source and prior prompts. Do not accept a terminal growth rate above the long run economy without flagging it as the single largest error in the model.</constraints>
<role>You are the buy-side IC member. You build the case the seller does not want in the room.</role>
<task>Build three scenarios for {{TARGET}}: the model as given (pitch), a base with every load bearing input at its honest midpoint, and a bear where the top three inputs from prompt 03 sit at the conservative end of their defensible range at once. For each, report EV, equity value, and implied entry multiple. Show the walk from the pitch number to the bear number, attributing the drop to each input.</task>
<output_format>A three column table (pitch, base, bear) with EV, equity value, implied multiple. A bridge from pitch EV to bear EV by input. One line: how much of the headline is assumption, not business.</output_format>
<constraints>Use the data source and prior prompts. Bear inputs are defensible and conservative, not doom, each surviving an operator challenge in {{SECTOR}}.</constraints>
<role>You are an independent reviewer who did not build any of the above and trusts none of it until it ties.</role>
<task>Re-derive {{TARGET}}'s value a SECOND, independent way from the raw inputs, do not reuse the prompt 02 chain: an implied exit multiple cross check, a reverse DCF backing out the growth the price requires, or a comps and precedent triangulation. Then reconcile: does it land within a stated tolerance of the prompt 02 EV; do the top three leverage inputs still explain the pitch-to-bear gap; are there inputs in the memo that never appeared in the source model. If everything ties and no inputs are invented, print RECONCILED, else print NOT RECONCILED with every break. No downstream prompt writes the memo until this prints RECONCILED.</task>
<output_format>The second independent valuation, the tolerance, a reconciliation table, an invented-input check, and a single hard line: RECONCILED or NOT RECONCILED with the breaks.</output_format>
<review_gate>This is a hard stop. A human owns this gate. NOT RECONCILED means the model is wrong until proven otherwise. Nothing leaves the desk on a number that did not reconcile.</review_gate>
<constraints>Use the data source and prior prompts. The second method must be genuinely independent of the prompt 02 build. Never round a break away to force a tie.</constraints>
<role>You are the valuation desk lead preparing the one table the IC always asks to see.</role>
<task>Build a two-way sensitivity grid of {{HEADLINE_EV}} and implied equity value across {{WACC}} on one axis and {{TERMINAL_INPUT}} on the other, stepped in defensible increments for {{SECTOR}}. Mark the model's current cell. Shade the defensible region versus the region requiring heroic assumptions. State the range the IC should plan around, not a single point.</task>
<output_format>A WACC by terminal input grid of EV, the current cell marked, the defensible band, and one line: the honest value range for {{TARGET}}, not a false-precision point.</output_format>
<constraints>Use the data source and prior prompts, and only run after prompt 07 printed RECONCILED. Increments defensible for {{SECTOR}}. Never present a single point EV as if the grid did not exist.</constraints>
<role>You are the buy-side IC member writing the half page that decides the vote.</role>
<task>Write a one page memo on {{TARGET}} an IC can act on in the time it takes to read it: the headline number and {{DECISION}}, the single load bearing assumption from prompt 04 and why it carries the answer, the bear case from prompt 06 and the honest range from prompt 08, the one or two cells to attack in diligence with the question to put to the seller on each, and the reconciliation status from prompt 07. No hedging, no wall of caveats.</task>
<output_format>A one page memo: headline and decision, the assumption doing the lifting, the range, the attack list with a seller question per cell, the reconciliation line. Written for a partner, not an analyst.</output_format>
<constraints>Use the data source and prior prompts. Only write after prompt 07 printed RECONCILED. Every figure traces to a prior prompt. No invented inputs.</constraints>
<role>You are the deal lead who has to get the truth out of the seller without tipping your hand.</role>
<task>For each cell on the attack list from prompt 09, draft the diligence request and the exact question to the seller or management. Anticipate the answer they will give to defend the assumption and the follow up that pins them down. Cover the load bearing assumption, the terminal input, and the top revenue driver at minimum.</task>
<output_format>A diligence table: cell under attack, the request, the question to the seller, their likely defense, your follow up. Ordered by how much the cell moves the answer.</output_format>
<constraints>Use the data source and prior prompts. Questions must be answerable from real diligence. Label anything speculative.</constraints>
<role>You are the harshest reviewer in the firm, brought in to find what the attack itself missed.</role>
<task>Turn on the whole analysis. Did the attack over-index on one input and miss a second assumption quietly carrying the answer? Are any bear inputs too soft, letting the model off easy? Is the honest range wide because the business is genuinely uncertain or because the model is under-specified? What one thing, if the seller produced it in diligence, flips this from a walk to a deal or a deal to a walk? Then state whether the prompt 09 memo is safe to send as written, or the one revision it needs first.</task>
<output_format>A red team list of what the attack missed, a soft-bear check, a read on the range, the single diligence item that flips the decision, and a verdict: send as written, or the one fix required first.</output_format>
<review_gate>A named human owns the send decision. This prompt informs it, it does not make it.</review_gate>
<constraints>Use the data source and prior prompts. Be adversarial toward your own analysis, not just the seller's model.</constraints>
Got the prompts. Want them wired into your actual stack? We map that on a free AI audit.
For PE deal teams and IC members handed a DCF to trust: 11 Claude prompts that rank every input by leverage, name the one assumption carrying the valuation, run the bear case, and hand you the cell to attack before the vote
Path A · free
You just did it
The setup rail and every prompt above are free and stay free. The cost is your time, and the risk of wiring it wrong on live data.
• Run last quarter's numbers first. Live data is not a test bed.
• Nothing here uploads to us. It runs in your own Claude account, on your own machine.
• A named human reviews and signs every output before it reaches a board, lender, or client.
• Mask account numbers and names to the minimum the task needs.
the fine print
Straight answers on ownership
Prompt set authored by consultance.ai. A DCF attacked here produces a first draft valuation read, not a fairness opinion or investment advice. Your data stays in your own Claude tenant; we never see it. Fair value work maps to ASC 820 and IFRS 13, valuation performance to AICPA SSVS No. 1 and IVS 105, and estimate auditing to PCAOB AS 2501; a named human signs off every figure against source before it reaches an investment committee, a board, or a counterparty. Not legal or investment advice.
Want this running in your business, not just your laptop? We build it and hand you the keys.
DCF Assumption Attack Kit is a finance and data build in the consultance.ai AI Build Library. For PE deal teams and IC members handed a DCF to trust: 11 Claude prompts that rank every input by leverage, name the one assumption carrying the valuation, run the bear case, and hand you the cell to attack before the vote. On your own data, inside Claude. It fits PE deal teams and principals, investment committee members, corporate development, family office principals, and CFOs who receive a DCF and have to decide whether to trust the number. Setup difficulty is Medium, with 4 plain-English steps.
What does DCF Assumption Attack Kit do?
For PE deal teams and IC members handed a DCF to trust: 11 Claude prompts that rank every input by leverage, name the one assumption carrying the valuation, run the bear case, and hand you the cell to attack before the vote. On your own data, inside Claude.
Who is DCF Assumption Attack Kit for?
It fits PE deal teams and principals, investment committee members, corporate development, family office principals, and CFOs who receive a DCF and have to decide whether to trust the number.
How hard is DCF Assumption Attack Kit to set up?
Medium to set up — one guided setup instruction covering 4 plain-English steps, plus 11 ready-to-run prompts on the resource page.
How would consultance.ai build this out?
The vault is about 60% of the build. We wire the last 40% into production in your own environment, deal data never leaving the building: live data wiring into your model and data room, role based permissions so the right people see the right cells, a retained audit trail your reviewer will accept, and the reconciliation gate as a standing pre vote control across every deal at scale. Done with you, then handed over so you own it.
What are the licensing terms?
Prompt set authored by consultance.ai. A DCF attacked here produces a first draft valuation read, not a fairness opinion or investment advice. Your data stays in your own Claude tenant; we never see it. Fair value work maps to ASC 820 and IFRS 13, valuation performance to AICPA SSVS No. 1 and IVS 105, and estimate auditing to PCAOB AS 2501; a named human signs off every figure against source before it reaches an investment committee, a board, or a counterparty. Not legal or investment advice.
Want this built into your workflow?
DCF Assumption Attack Kit is the starting point. On a free AI audit we map where it fits your stack and what consultance.ai would build around it.