KPIs and financials

Fundra extracts what it can't calculate, calculates everything else, and reconciles the messy reality of startup reporting into one comparable series.

Startup reporting is messy. The same company reports in different currencies, in thousands one quarter and units the next, restates last year's figures without saying so, sends a board deck with three KPIs and a spreadsheet with forty, and skips a month entirely. Half the point of Fundra is absorbing that mess and producing one series you can trust and benchmark.

The rule is the same as in the ledger: extract only what can't be calculated, and calculate the rest. Reported ARR is extracted. Growth, burn multiple and runway are not – they're derived, so they're computed the same way for every company and are actually comparable.

Extraction, with a reference

Fundra reads KPIs out of whatever arrives – investor updates, board decks, financial statements, cap tables, spreadsheets, meeting notes. There's no template to impose and no format to insist on: a narrative email, a management-accounts export, a slide with three numbers on it and a full P&L all work, in whatever language they're written in.

Extraction goes through the whole document rather than the headline figures. A board pack that contains a cohort table, a cash bridge and a hiring plan yields all of it, not just the ARR on slide two.

Every extracted value keeps a pointer back to where it came from:

  • Spreadsheets record the sheet name and the exact cell (Financials!C14). Where a figure was assembled from several cells, the whole set is kept.
  • PDFs and decks record the page number and the surrounding text fragment.
  • Emails record the message the number was read from.

So every number in Fundra can be traced to its origin in one click. This is the difference between a number you can put in an LP report and a number you have to go and check.

Extraction also detects the things that silently corrupt a dataset: the reporting currency, the scale (absolute, thousands, millions) and the reporting cut-off date. A company that switches from thousands to units mid-year doesn't produce a 1000x growth rate.

Reconciliation

The same metric for the same period often arrives more than once – a board deck in January, an audited spreadsheet in March, someone's manual correction in between. Fundra reconciles them into a single value rather than making you choose every time.

The rules, in order:

  1. A value you selected wins. If someone has explicitly picked a value for that period, it stays picked.
  2. More recent reporting wins, based on the report date rather than the upload date.
  3. Within the same date, source type decides: spreadsheet → manual → PDF → email. Spreadsheets are the most structured and least ambiguous, so they rank first.
  4. Manual input close to an extraction defers to the extraction. If a hand-entered value is within 10% of what was extracted, the extracted value is kept – it carries a source reference and the manual one doesn't.
  5. Values differing by more than 10% are flagged as conflicting rather than silently averaged, so you see the disagreement and can resolve it.
  6. A resolution can be reopened. If a newer report arrives after you resolved a conflict and differs materially, it's flagged again rather than hidden behind your earlier decision.

Alternatives are never thrown away. Every competing value stays attached to the period with its own source reference, so a restatement is visible as a restatement.

Periods and aggregation

Reported data is normalised onto a period grid – months, quarters, fiscal years and custom periods – and then rolled up. How a metric aggregates depends on what it is:

BehaviourMetrics
SummedFlow metrics over the period: revenue, COGS, gross profit, OPEX, EBITDA, EBIT, net income, net cashflow, GMV, S&M spend, new/expansion/contraction/churned ARR
Last valuePoint-in-time metrics: ARR, MRR, cash balance, headcount, NDR, total assets, total debt, total equity, runway, eNPS
AveragedNet burn rate
CalculatedRatios, recomputed from their components rather than averaged – gross margin is sum(gross profit) / sum(revenue), never the mean of monthly margins

That last row matters more than it looks. Averaging twelve monthly margins gives a different – and wrong – answer to dividing annual gross profit by annual revenue. Fundra always recomputes.

Point-in-time metrics reported only annually or quarterly are also rolled down to the month they apply to, so a company reporting cash balance once a year still sits on the same monthly grid as one reporting every month.

Derived metrics

Once the reported base is reconciled, Fundra computes the rest. These are calculated identically for every company, which is what makes cross-portfolio comparison meaningful.

MetricDefinition
ARRMRR × 12, or the summed monthly ARR for a quarter
MRRARR / 12 when only ARR is reported
Net New ARRARR this period − ARR last period
(+) New ARRNet new ARR, backing out expansion, contraction and churn
(%) Churn Rate(churned ARR + contraction ARR) / previous ARR
Annualized Revenue / GMV / Gross ProfitLast month × 12, falling back to a rolling 3 months × 4, then to the period scaled to 12 months
Gross Profitgross margin × revenue
Gross Margingross profit / revenue
EBITDA MarginTrailing 6-month EBITDA over trailing 6-month revenue
Net CashflowChange in cash balance versus the previous period
Net Burn RateAverage monthly net cash burn over the trailing 3 months. Over a full year, the change in cash net of equity and debt issuance, divided by 12
Runwaycash balance / net burn rate, in months – reported only when burn is positive
ARR / FTEAnnualised revenue or ARR divided by headcount
Magic NumberNet new ARR over the previous period's S&M spend
Burn MultipleTrailing 3-month net burn over trailing 3-month net new ARR
Rule of 40YoY revenue growth + trailing 3-month EBITDA margin
LTV / CACReported LTV over reported CAC
GMV Take RateRevenue over GMV

Most of these fall back through several definitions rather than returning nothing: burn multiple prefers net new ARR but will use net new revenue, revenue run rate prefers annualised revenue and falls back to billed then contracted ARR. A company that reports differently to its peers still gets a comparable number, labelled with what was actually used.

Growth rates – period on period and year on year – are computed automatically for the headline series: GMV, revenue, ARR, MRR, gross profit, gross margin, EBITDA, net income and their annualised variants.

Derived metrics are recalculated whenever their inputs change. Correcting a restated revenue figure from eighteen months ago updates every margin, multiple and growth rate that depended on it – you don't have to hunt them down.

Valuation multiples

Reported performance is only half of a valuation question. The other half is what the market paid for it – so Fundra computes valuation multiples at every valuation event, whether that's a funding round or a mark you set yourself.

At each event, the enterprise value is matched to the company's financial period ending immediately before it – the month end just prior to the valuation date, or the same month end when the valuation falls on one. Enterprise value is converted into the company's reporting currency at that date's rate, so the numerator and denominator are in the same money. Fundra then computes:

MultipleDenominator
EV / ARRAnnualised revenue, falling back to ARR
EV / Gross ProfitAnnualised gross profit
EV / EBITDAAnnualised EBITDA

Each multiple records which period it was matched against, so you can see exactly what the company had reported when the round was priced – not an approximation from the nearest annual figure.

Turning recollection into fact

The useful part is that this exists for every valuation event across the whole portfolio, historically. Questions that are normally answered from memory become queries:

What did our companies look like when they raised a Series A? What was the underlying performance, what round sizes resulted, and at what multiples?

That's a real answer – ARR at the time of each Series A, the round size, the pre- and post-money, and the multiple paid – rather than the two or three examples someone happens to remember. The same applies to any cut: by segment, by vintage, by lead investor, by whether the company was growing 100% or 40% at the time.

Ask it directly through the MCP server or build it into your own analysis through the API.

Reading them back

All of this is available through the API/metrics for the values and /metrics/names for what's available – and through the MCP server if you'd rather ask questions in plain language. Metrics written back through POST /metrics join the same reconciliation flow as everything else.

On this page