Reading the Financials: The Primer for Everyone Who Never Had To Care

Contents11 sections

The first board presentation of a platform rebuild opened with architecture: diagrams, service boundaries, the whole apparatus. Twelve minutes in, a director asked what it did to gross margin. Nobody in the room knew. There was a number for cost: what was missing was the line it would land on.

That gap takes about an afternoon to close.

Three statements, three questions

Every company produces three financial statements. They answer different questions and they disagree with each other on purpose.

The income statement is an opinion shaped by accounting rules; the cash flow statement is closer to a fact. When those two tell different stories about the same quarter, the difference is usually capitalization or timing. That gap is where most technology budget arguments live.

The shape of a software P&L

You only need the top of it: most technology decisions land in one of four places.

One boundary catches technology leaders out. It runs between COGS and R&D, and it decides whether a given dollar of hosting lands above the gross margin line where investors read it or below in operating expense where it quietly joins the cost of running the company. Production hosting for the product customers use is usually COGS. The same cloud account running your CI, staging and internal analytics is usually operating expense. One hits gross margin; the other leaves it alone. If your AWS bill arrives as a single undifferentiated invoice with no tagging, your finance team is guessing at that split; their guess will not flatter you, and tagging is the whole fix.

One company I worked with tagged nothing for two years. When they finally split production from non-production, gross margin moved four points in the right direction (from 71 to 75, against a SaaS benchmark running 70 to 85). Nothing changed technically. They had been reporting internal dev spend as cost of goods sold and wondering why their margin looked thin next to competitors.

This gets sharper with AI. Inference you run on behalf of a paying customer is delivery cost. It lands in COGS and it scales with usage. A feature that delights users and consumes tokens turns gross-margin dilutive at exactly the moment it succeeds, which is the part that surprises people. Part 8 covers that properly.

EBITDA, and why everyone is obsessed with it

EBITDA is earnings before interest, taxes, depreciation and amortization. Start at operating income and add back D&A.

It exists for one reason: you can compare two businesses without their financing and accounting choices getting in the way. One company borrowed to buy servers. Another leases them. A third capitalizes half its engineering team, which flatters its operating expense, thickens its balance sheet, and leaves a five-year amortization tail that the next CFO inherits along with the code. Strip out interest and D&A and a rough view of the operating business underneath comes into focus; rough is the word to hold onto.

Three things follow from that, and together they explain a great deal of corporate behavior.

• Private equity buys on a multiple of EBITDA. At 12x, one dollar of recurring EBITDA carries twelve dollars of enterprise value, which is why cost decisions get emotional.
• Lenders write covenants against it, usually debt to EBITDA and interest coverage. Breach one and you are in a very different conversation with your bank.
• Anything that moves spend across the D&A line moves EBITDA while leaving cash untouched.

That last point is the whole game. Buying hardware instead of renting it shifts where the spend appears on the page without changing by a single cent what actually left the bank; so does extending the useful life you assume for what you already own. Both moves are legal and routine.

The correction belongs right beside it. EBITDA is an earnings measure rather than a cash measure, and it deliberately ignores the capital you spent. A company that buys $40M of servers every year and adds the depreciation back reports earnings that pretend the servers arrived free (Charlie Munger put it more bluntly than I will). Serious buyers adjust for it; internal debates frequently skip the adjustment, which is where a CTO who understands it picks up an advantage.

Metrics you will hear in the room

Free cash flow is the one number classification cannot move. When someone in the room keeps steering back to it, they are refusing to be fooled.

How this changes by company stage

The same proposal earns three different answers depending on who owns the company. Moving between company types is where that lands hardest.

The practical read: at a PE-owned company, framing a migration as a one-time project rather than an ongoing cost increase is accurate reporting — and it is the difference between approval and rejection. At a venture-backed company the same framing is beside the point, and months of runway consumed is the number that decides it.

Book versus tax, two sets of books and both legal

This confuses people who assume there is one truth; there are two, and they serve different purposes.

Book accounting, meaning GAAP, exists to tell investors what happened; tax accounting exists to calculate what you owe. They apply different rules to the same transaction, on purpose, and reconciling them is a normal part of closing the year.

So a server can be depreciated over five years for book purposes and fully expensed in year one for tax purposes under bonus depreciation, and both are correct. Software development can be expensed for book and deducted immediately for tax under Section 174A, or capitalized for book and still deducted for tax.

When someone says "we can write that off," ask which book they mean. Half the time the answer changes the decision.

What I actually do before a budget meeting

1. Know which line my spend lands on: COGS or opex. When the answer is unclear, I go find out before the meeting rather than during it.
2. Split one-time from run-rate explicitly. Two numbers on the same slide, always.
3. State the EBITDA impact as a named line, and state it most carefully when it is unflattering.
4. Bring the cash number next to it. When the two disagree, I explain why before somebody else does.
5. Ask the CFO which metric the board is actually being judged on this year. It changes every year, and nobody volunteers it until asked.
6. Learn the covenant thresholds if there is debt. They are the invisible wall a perfectly good proposal hits.

This is vocabulary rather than finance expertise. Twelve minutes into that board presentation the question was never really about architecture: it was about which line the money would land on, and the line is the thing a technically correct proposal has to name before it meets the people holding the budget.

Glossary

References

• FASB Accounting Standards Codification, Topics 350 and 985, for framework context
• IRC Section 168(k) bonus depreciation and Section 179 expensing, as amended by the One Big Beautiful Bill Act, July 2025
• Flexera 2026 State of the Cloud Report, on cloud cost visibility and FinOps maturity
• Standard SaaS benchmarking conventions for gross margin and Rule of 40, as used by Bessemer, ICONIQ and SaaS Capital periodic reports

Best next read

Part 3, Capitalize or Expense: Your Jira Board Is Now Audit Evidence. You now know why the classification matters. Part 3 is the rulebook that decides it, and it was rewritten in September 2025.

The full series

This series is an introduction for technology leaders. It is accounting and tax context rather than accounting or tax advice. Model it with your CFO and tax advisor before you commit.