What Claude AI can—and cannot—do in a DCF model
Claude AI can help an analyst organise evidence, write formulas, explain assumptions, review a workbook, and run structured scenario analysis. It does not replace the valuation model, primary-source research, or professional judgment. A DCF remains an assumptions-driven framework: the quality of its output depends on revenue growth, margins, reinvestment, working capital, terminal growth, and the discount rate.
The most reliable approach is to use Claude as a research and modelling copilot, while Excel, Google Sheets, Python, or a financial modelling platform remains the calculation layer. This separation makes every number traceable and easier to audit.
For developers building a finance workflow, Claude’s API can support document extraction and review. A useful starting point is this guide to building a personalised AI assistant with the Claude API, particularly when you need repeatable prompts, structured outputs, and access controls.
Core DCF mechanics to preserve
A conventional DCF discounts forecast unlevered free cash flow to its present value and adds the present value of terminal value:
- Revenue and operating assumptions: volume, price, customer additions, churn, utilisation, or capacity.
- EBIT and taxes: operating profit after applying a normalised tax rate.
- Non-cash charges: depreciation and amortisation.
- Reinvestment: capital expenditure and changes in net working capital.
- Free cash flow: typically EBIT × (1 − tax rate) + D&A − capex − change in NWC.
- Discount rate: usually WACC for enterprise value, or a suitable cost of equity for an equity cash-flow model.
- Terminal value: commonly the perpetual-growth method or an exit multiple, with both requiring explicit justification.
Claude can explain these relationships and generate a model skeleton, but it should not invent a WACC, terminal multiple, or forecast merely because a prompt asks for a complete valuation. Every input should have a source, date, unit, and rationale.
A practical Claude AI workflow for DCF models
1. Define the valuation question
Start with the asset, valuation date, currency, forecast period, and purpose. A listed Indian company, an early-stage SaaS business, and an infrastructure project need different operating drivers. State whether the output should be enterprise value, equity value, or a range.
A useful prompt is:
> Build a five-year DCF framework for [company] as of [date]. Separate historical facts, analyst assumptions, and derived calculations. Do not fill missing data with estimates without labelling them. Return assumptions in a table with source, unit, period, and confidence level.
2. Gather and classify source material
Provide Claude with annual reports, investor presentations, exchange filings, earnings transcripts, and management commentary. For Indian companies, distinguish consolidated from standalone figures and check whether amounts are reported in ₹ crore, ₹ million, or ₹ lakh. Do not rely on an unverified web summary when the filing is available.
Ask Claude to extract facts into a structured table rather than directly producing a valuation. Include:
- Fiscal year and reporting period
- Revenue, EBITDA, EBIT, tax, capex, and working capital
- Net debt, leases, minority interest, and investments
- Segment performance and management guidance
- One-off items and accounting changes
- Source page or document reference
This source-first process is more dependable than asking for a finished model in one prompt.
3. Translate business drivers into forecasts
Ask Claude to propose driver-based assumptions, then challenge them. For example, revenue may be built from users × average revenue per user, stores × same-store growth, or capacity × utilisation × realisation. Costs may be tied to revenue, headcount, commodity prices, or operating capacity.
Require three cases:
- Bear case: slower growth, weaker margins, higher reinvestment, or a longer working-capital cycle.
- Base case: the most defensible interpretation of evidence.
- Bull case: stronger execution, but still within an explicitly stated operating range.
Claude is especially useful for identifying where a forecast jumps abruptly from historical performance. It can flag an unexplained margin expansion or a terminal growth rate that exceeds the long-term economic growth of the market.
4. Use Claude to build, not silently alter, the workbook
You can ask Claude to generate Excel formulas, Python code, or a model layout. Keep inputs, calculations, and outputs on separate sheets or modules. Use consistent sign conventions and make units visible in every table.
A robust model should include:
- Historical financials and source notes
- Assumption inputs with low/base/high values
- Forecast income statement and cash-flow bridge
- WACC calculation with component inputs
- Terminal value calculation
- Enterprise-to-equity-value reconciliation
- Sensitivity tables and checks
- Date, currency, version, and reviewer fields
If Claude returns code, run it in a controlled environment and test it against hand-calculated examples. AI-generated formulas can be syntactically correct while economically wrong.
Validation and control checks
Never accept a Claude-generated DCF without independent checks. Ask the model to produce a review checklist, then verify each item yourself or through a separate calculation script.
Minimum controls include:
- Balance sheet and cash-flow consistency where applicable
- Correct discount-period convention
- Correct treatment of debt, leases, cash, and minority interest
- No double counting of depreciation, capex, or working capital
- Terminal value as a sensible share of enterprise value
- WACC greater than terminal growth in the perpetual-growth method
- Sensitivity to revenue growth, margin, WACC, and terminal growth
- Reconciliation from enterprise value to equity value and per-share value
- Clear distinction between reported data and estimates
For a high-stakes investment committee, maintain an audit trail of prompts, source files, model versions, and manual changes. Do not upload confidential deal information to a consumer AI workspace without reviewing the provider’s data-retention and enterprise-security terms.
India-specific considerations
Indian DCF work often requires extra attention to inflation, currency, tax structure, working capital, and conglomerate complexity. A domestic investor may model cash flows in rupees, while a global investor may translate them into dollars. The currency of cash flows and the discount rate must be consistent.
Check GST treatment, capitalised costs, lease accounting, exceptional items, related-party transactions, promoter holdings, contingent liabilities, and subsidiary-level debt. For banks and insurers, a standard FCFF DCF is often unsuitable; dividend discount, excess-return, or other sector-specific approaches may be more appropriate.
Claude can compare annual reports and highlight inconsistencies, but it should not be treated as a regulator, auditor, or source of live market data. Confirm market prices, government bond yields, beta inputs, and corporate actions from current, reputable sources as of the valuation date.
Prompts that produce better results
Use prompts that constrain the task and expose uncertainty:
- “List every assumption that materially affects value and rank it by sensitivity.”
- “Separate facts from estimates; cite the document and page for each fact.”
- “Review this formula for unit, sign, and timing errors.”
- “Create a bear/base/bull table without changing historical figures.”
- “Explain why this terminal value may be economically unreasonable.”
- “Return JSON with field names, units, source, confidence, and missing-data flags.”
Avoid prompts such as “give me the correct share price.” Valuation is not a single factual answer; it is a range conditional on assumptions.
Where Claude adds the most value
Claude is most useful for document-heavy, repeatable work: extracting guidance, comparing periods, drafting an investment memo, reviewing formulas, and explaining model movements to non-specialists. It can also help teams standardise analysis across companies and reduce time spent on repetitive checks.
If you are comparing model providers or building a production workflow, review Claude vs Gemini API for developers in India. For local or sensitive deployments, how to deploy large language models locally offers relevant architectural considerations, although local deployment may involve trade-offs in capability, cost, and maintenance.
Final takeaway
Claude AI for DCF models is valuable when it improves traceability and challenges weak assumptions—not when it produces an impressive-looking valuation with no evidence. Keep calculations deterministic, label uncertainty, validate every output, and retain human ownership of the investment conclusion. Used this way, Claude can make valuation research faster and more consistent without weakening financial discipline.
Indian founders building finance, accounting, or research products can explore support through AI Grants India, including opportunities relevant to responsible AI applications.