How to Create Financial Projections (3-Statement Model)

Financial projections are usually a deck artifact — built once for an investor meeting, never opened again. That is the wrong use. A good model is an operational tool: updated monthly, compared against actuals, revised when assumptions fail. The deck version is a byproduct.

This guide walks through building the model for operational use, with the financial-statement discipline that survives investor scrutiny as a side benefit.

1. Three statements, linked

The three financial statements, briefly:

• Income statement: Revenue, costs, and resulting profit over a period. Accrual basis.
• Balance sheet: Assets, liabilities, and equity at a point in time.
• Cash flow statement: Cash movement over the period, categorised as operating, investing, financing^[2].

They link: net income from the income statement flows to retained earnings on the balance sheet. Changes in balance-sheet items (AR, AP, inventory, capex) drive the cash flow reconciliation. Ending cash on the cash flow statement ties to cash on the balance sheet.

Build all three. A single-statement projection (usually just revenue and expenses) hides the working-capital and cash-timing effects that cause most small-business distress. The income statement can show "profit" while the cash flow statement shows you running out of money in month 8 because receivables extended and inventory expanded.

2. Drive from operational inputs, not top-down

A top-down model says: "We'll grow 10% per month." A bottom-up model says: "We'll close 12 new deals at $15k ACV, renewing 85% of our $300k base, resulting in $XX MRR."

Both produce numbers. Only the second produces numbers you can stress-test.

For B2B SaaS, the driver stack:

• Pipeline coverage and historical close rates by stage → new bookings.
• Cohort retention curves → revenue from existing customers.
• Sales ramp time × rep count × quota attainment → new-bookings capacity.
• Headcount plan + loaded cost per role → operating expenses.
• Cash timing: billings vs. revenue, AR collection lag, AP payment timing → cash from operations.

For consumer/transactional businesses, the driver stack centers on acquisition (channel CAC × spend = new users), activation (new users × activation rate = active users), and monetisation (active users × ARPU). Each is its own sub-model.

The useful test: can you explain how a 10% miss in any single driver propagates through to revenue, expense, and cash? If no, the model is not driver-based.

3. Sanity-check against public benchmarks

Projections often drift into impossible ranges because they extrapolate from a small sample of good months. Sanity-check against public data:

• Revenue growth rate. In the Fed's 2024 Small Business Credit Survey, 43% of employer firms grew revenue over the prior year, 18% held flat and 39% shrank^[1]. For B2B SaaS, SaaS Capital's 2025 survey put median growth at 24%^[3]. A projection far above those needs a driver that explains it.
• Gross margin. Varies by category: Damodaran's January 2026 data puts the average at 71.7% for listed system and application software companies, 33.4% for business and consumer services and 33.2% for general retail^[4]. Project a margin far from your category's, and be prepared to explain why.
• Operating margin at scale. Listed companies are mature; an early-stage company often runs at a loss while it grows, so judge the margin the model reaches at scale, and show the year it turns positive.
• CAC payback. David Skok's SaaS Metrics 2.0 suggests recovering CAC within 12 months^[5]. A short payback that the model holds while you scale to thousands of customers deserves scrutiny, because channels usually get more expensive as you exhaust the cheapest audience.

The model should justify any number materially outside these ranges. "Because our team is great" is not a justification; "because we have a structural cost advantage from X" is.

4. Build P50, P75, and downside scenarios

Three scenarios, each serves a different purpose:

• P50 — realistic. Your genuine best estimate of what will happen. This is the headline number for investors and is what you plan sales and marketing capacity around.
• P75 — conservative operational plan. Assumes close rates 20% lower, sales cycles 20% longer, churn 1 percentage point higher. This is what you plan hiring and vendor commitments against. If you commit to headcount based on P50, you will have to cut when it misses.
• Downside — what if the environment breaks. Revenue declines 20%, a major customer churns, you can't raise the next round on schedule. What do you do, and when?

The downside case is not pessimism; it is preparation. A company with a clear downside plan acts faster when conditions deteriorate, which is often the difference between distress and survival.

5. Projections as operational artifacts

The model is useful only if it is updated and compared to actuals regularly. Monthly cadence:

• Pull actuals from accounting and CRM systems into the model.
• Compare actual vs. P50 projection. Document the variance by driver (not just aggregate).
• Identify systematic biases. If new bookings missed by 15% three months running, the driver assumption is wrong, not the arithmetic.
• Revise forward projections where evidence has changed. Do not revise to hide past misses; revise to improve future decisions.

Quarterly, re-run the P75 and downside scenarios with updated inputs. Monthly fluctuations should not move those; structural shifts (new product line, major customer loss, competitive entry) should.

A projection model that stays in sync with reality is worth more than one built to a higher standard and then abandoned. In the typical case, the habit of monthly revision is what distinguishes operationally useful projections from pitch-deck artifacts.

6. Common traps in financial modeling

Four patterns that produce unreliable projections:

• Hockey-stick revenue with no underlying driver. If the model shows exponential growth without explaining where each new customer comes from, the model is aspiration. Investors and experienced operators can spot this instantly, and it damages credibility.
• Static cost assumptions with dynamic revenue. Revenue grows 200% but COGS, support cost, and infrastructure stay flat. Fundamentally implausible: most scaling costs grow with revenue, so model each one against its driver.
• Double-counted revenue. Booking revenue from the same customer in both new-bookings and renewals buckets. Easy to do unintentionally when the model separates segments by channel; discipline the definitions.
• Ignoring working capital in cash flow. A model showing positive net income while the balance sheet accumulates $2M of receivables is showing non-cash profit. Cash flow from operations corrects this if the model is properly linked.

7. Tools and who should build the model

For most small businesses, Excel or Google Sheets is the right tool for financial projections. Purpose-built planning software is useful once complexity grows beyond what a single-sheet model can handle — typically when you have 10+ cost centres or multiple product lines.

Ownership of the model matters:

• Founder-built (pre-seed to Series A). Founders should understand the model deeply enough to explain every assumption. Outsourcing this early is usually a mistake.
• Founder-built, finance-maintained (Series A onward). Part-time fractional CFO or full-time finance hire takes ownership of maintenance and calibration, with founder reviewing quarterly.
• CFO-owned (Series B onward). Full-time CFO or VP Finance owns the model, with monthly board-ready outputs. At this stage, the model is part of standard operating cadence, not a one-off artifact.

The model is a living document. A 12-month-old model without monthly updates is worse than useless — it produces numbers that look plausible but are wrong, which drives worse decisions than having no model at all. Build the habit of monthly update into the calendar, not into someone's good intentions.

8. Numeric worked example — top-down vs bottom-up reconciliation

A pre-Series-A SaaS has $2.1M ARR, 18 reps, and a founder projecting "50% growth to $3.15M in year 1." Rebuild the number bottom-up and see where the assumptions break.

Top-down claim
  Year 1 ARR target        $3.15M   (+$1.05M net-new ARR)
  Implied NRR (if 100% new  150%    (SaaS Capital's 2025 median is 101%)
     from existing)

Bottom-up rebuild
  Existing base ARR        $2.10M
  Expansion (8%)           +$168k
  Churn on existing (5%)   −$105k
  Existing contribution     $2.16M  (NRR 103%)

  Net-new ARR needed       $0.99M to hit $3.15M target
  18 reps × 70% attainment × avg quota $85k = $1.07M capacity
  New-logo timing drag (ramp + sales cycle): 75% realised in-yr
  Realistic new-logo ARR   ~$0.80M

  Realistic year-1 ARR     ~$2.97M (41% growth, not 50%)

The top-down 50% is only plausible with rep capacity and attainment both hitting best-in-class simultaneously. The bottom-up number (~41%) is still strong: SaaS Capital's 2025 survey put median growth for private B2B SaaS companies at 24%^[3]. The gap between $3.15M and $2.97M is where hiring commitments, cash planning, and investor expectations diverge from reality.

9. Failure modes worth naming

• Model built once in Sheets, never re-opened. A projection that isn't reconciled to actuals monthly drifts toward fiction inside a quarter. Put the reconciliation meeting on the calendar; the model earns its keep only if the habit holds.
• Unit economics assumed to hold at 10x scale. CAC at $8k with today's top-of-funnel costs may not hold when you are spending 10x the current paid-acquisition budget. Test the model with a higher CAC, because channels get more expensive as you exhaust the cheapest audience.
• Balance sheet treated as optional. A model that skips AR/AP/inventory line items will show income but miss the working-capital stress that drives most small-business distress. ASC 230 sets out how operating cash flow reconciles from net income^[2]; skipping the reconciliation breaks the cash line.

The models that survive board scrutiny are the bottom-up, driver-linked, monthly-reconciled kind, not the polished single-tab "hockey stick" decks.

Frequently asked questions

What makes a financial projection defensible?

Three linked statements — income, balance sheet, and cash flow — driven by operational inputs (pipeline, retention, headcount), not a top-down growth assumption. Most founder models fail scrutiny because they are driven by 'we'll grow X percent a year' rather than ground-up unit math. Build all three statements: a single revenue-and-expense projection hides the working-capital and cash-timing effects that cause most small-business distress, where the income statement shows profit while the cash-flow statement shows you running out of money in month 8 as receivables extend and inventory expands.

What is the difference between top-down and bottom-up modeling?

A top-down model says 'we'll grow 10 percent a month'; a bottom-up model says 'we'll close 12 deals at $15k ACV, renew 85 percent of our $300k base, resulting in this MRR.' Both produce numbers, but only the bottom-up version can be stress-tested. The useful test is whether you can explain how a 10 percent miss in any single driver propagates through to revenue, expense, and cash — if not, the model is not driver-based. In the worked example, a top-down 50 percent growth claim rebuilt bottom-up came to a credible 41 percent once rep capacity, attainment, ramp, and churn were modeled honestly.

Why build multiple scenarios instead of one forecast?

Because each serves a different purpose. P50 is your genuine best estimate — the headline number for investors and the basis for sales and marketing capacity. P75 (close rates 20 percent lower, cycles 20 percent longer, churn 1 point higher) is what you plan hiring and vendor commitments against, so you don't have to cut when P50 misses. The downside case (revenue down 20 percent, a major customer churns, the next round slips) is preparation, not pessimism — a company with a clear downside plan acts faster when conditions deteriorate. In the Fed's 2024 Small Business Credit Survey, 39 percent of employer firms reported lower revenue than a year earlier, so a model without a downside case is incomplete.

How do I sanity-check my projections?

Compare against public data. In the Fed's 2024 Small Business Credit Survey, 43 percent of employer firms grew revenue over the prior year and 39 percent shrank, so steady growth is not the default. SaaS Capital's 2025 survey put median growth for private B2B SaaS companies at 24 percent. Damodaran's January 2026 data puts average gross margin at 71.7 percent for listed system and application software companies, 33.4 percent for business and consumer services and 33.2 percent for general retail. Any number materially outside these ranges needs a structural justification, not 'because our team is great.'