company-financials

Company Financials

Pull a single company's profit-and-loss figures from Economic Mind and present them as a clean, sourced table. When the user asks about margins or trends, add a derived-metrics block and short commentary.

Inputs

• Company (required): a name (preferred) or a companyId if the user gave one.
• Years (optional): default = the tool's default (last 3 calendar years). For a margin/trend request, default to the last 5 years so a trend is visible. Honor any explicit range.
• Currency (optional): e.g. USD, EUR, GBP, NOK, SEK, DKK. Default USD.

Procedure

1. Resolve the company

If the user gave a companyId, use it. Otherwise call search_companies with queries: ["<name>"].

• Top-pick-vs-confirm: if the top hit's score is at least 1.5× the second's (or there is only one hit), use it and announce: Using <companyName> (id <companyId>) — top match.
• Otherwise show the top 3 (Company, ID, HQ, What they do, Relevance) and ask the user to pick before continuing.

2. Pull the financials

Call get_company_financials:

• companyIds: [<companyId>].
• years and currency: pass only if the user specified them (for a margin/trend request, pass the last 5 years). Otherwise let the tool defaults apply.
• Do not pass lineItems — the default returns all 14, which is what we want.

If the company comes back in notFoundCompanyIds (or rows is empty), tell the user there is no P&L data for it and stop.

3. Build the P&L table

Read the company's rows (one per year, each with year, conversionRate, and a sparse reportedLineItems map).

Render a Markdown table with years as columns, newest on the left:

• A first data row: FX rate = conversionRate per year (1.0 when currency is USD; e.g. NOK 10.75 means 1 USD = 10.75 NOK).

• One row per line item, in this order, using each year's value from reportedLineItems (use the keys exactly as they appear in the response):

  Revenue · COGS · Gross profit · Operating expenses · EBITDA · D&A · EBIT · Capex · Free cash flow · Net income · Labor costs · Net interest · Tax · Pre-tax profit

• Sparsity is data, not error: a line item missing for a year stays blank — never write 0. Values are in millions of the chosen currency.

Above the table, state: company name, HQ (from the search result, if any), currency, and the year range.

4. (Optional) Margin & trend analysis

If the user asked for margins, trends, or "analysis", add a Derived metrics block below the P&L (compute only from values already in the table; leave blank where an input is missing):

• Gross margin = Gross profit / Revenue
• EBITDA margin = EBITDA / Revenue
• EBIT margin = EBIT / Revenue
• FCF conversion = Free cash flow / EBITDA
• Capex intensity = Capex / Revenue
• Revenue growth (YoY) = Revenue[year] / Revenue[prior year] − 1 (earliest year blank)

Then 3–6 bullets of Commentary: margin direction and magnitude (at least gross and EBITDA), any anomaly worth flagging, and the revenue-growth pattern. Cite the figures you reference (e.g. "EBITDA margin fell from 28% in 2024 to 21% in 2020"). Do not write commentary about metrics that are blank.

5. Sources

Add a short Sources section. Each reportedLineItems entry carries a source string (the underlying data provider, e.g. "Proff (NO)"). List the distinct sources used, and state that figures are as reported in national filing registries via Economic Mind.

6. (Optional) CSV

Offer: "Want this as a spreadsheet? I can write a CSV." If yes, write <companyName>_PnL_<currency>_<firstYear>-<lastYear>.csv (ASCII, spaces → underscores) with the table, and name the file.

Optional: ownership context

If — and only if — a web-search tool is available in this environment, you may add a one-line ownership note (Public / Private / PE-owned / State-owned / etc., plus the primary owner) with a cited source and access date. If web search is unavailable or nothing reliable turns up within a couple of searches, omit it — do not guess. The core deliverable is the sourced P&L; ownership is a nice-to-have.

Rules

• Show the P&L verbatim from get_company_financials — no rounding beyond what the tool returns, no dropping line items.
• Sparse cells stay blank, not zero. Values are in millions.
• Apply the per-year conversionRate when explaining non-USD figures; small rounding differences in non-USD responses are expected.
• Never invent or recompute reported figures. Derived metrics are computed only from values already shown, and only when the user wants analysis.