Digital products can distribute at near-zero cost and still have poor unit economics. Cloud workloads, AI inference, data suppliers, app-store commissions, payment fees, customer support, implementation, fraud and acquisition spend can grow with revenue. A subscription dashboard may show attractive recurring revenue while cash payback takes years. A marketplace can report gross merchandise value even though its take rate does not cover disputes and incentives.
Unit economics turn these broad results into an operational question:
When one more economically meaningful unit is acquired, used and retained, how much value does the company receive and which costs does it cause?
The answer depends on defining the unit correctly. It could be a paying customer, a company account, an active workspace, an order or transaction, a cohort of sellers or buyers, an API operation, a generated outcome, a licensed device, a project, or a unit of consumption.
Pick the one your costs actually scale with. Economics measured per customer in a product whose cost scales per workspace will look healthy until the average customer opens a second workspace.
A useful model connects four layers:
- revenue quality — what is actually earned and collected;
- service economics — variable cost and contribution;
- acquisition economics — the cost and timing of winning demand;
- retention economics — how contribution persists, expands or disappears.
No single ratio can replace this system.
Start with the decision, not the formula
Unit economics are managerial tools. Define the decision before creating a spreadsheet.
Examples include:
- Can the company afford paid acquisition?
- Which package should receive more marketing?
- Does an AI feature need usage limits?
- Is an enterprise segment attractive after implementation?
- Should annual prepayment receive a discount?
- Does a marketplace commission cover transaction risk?
- Is a lower regional price contribution-positive?
- Should a service remain bundled or be sold separately?
- Can a free plan support its own conversion economics?
The decision determines the unit, time horizon and cost boundary. A transaction-level model helps set a marketplace fee. A customer-cohort model helps set acquisition budgets. An account-level model helps compare enterprise packages.
Do not begin with “What is our LTV?” Begin with “What must we decide, and which cash flows change because of that decision?”
Units and states: precise definitions
A customer can mean a billing entity, user, account, workspace or active subscription. Mixing these definitions corrupts every ratio.
Create a metric dictionary.
| Term | Example definition |
|---|---|
| New customer | First paying billing entity activated in the period |
| Active account | Account with valid paid entitlement at period end |
| New logo | First contract with a legal organization |
| Reactivation | Previously churned account returning to paid status |
| Expansion | Additional recurring value from an existing account |
| Net revenue | Product consideration after refunds, credits and indirect tax |
| Variable service cost | Cost expected to change with delivery or customer activity |
| Acquired cohort | Units whose first paid activation occurred in the same period |
Specify how trials, free accounts, refunds, failed payments, paused subscriptions, internal accounts, partners and merged workspaces are treated.
Choose a time grain. Monthly cohorts fit many self-serve subscriptions. Quarterly cohorts may be more stable for enterprise products with few deals. Transaction products may require daily operational views and monthly strategic views.
The revenue bridge
Revenue is not the checkout sticker price.
For a subscription product:
net product revenue = invoiced product consideration
− discounts and credits
− refunds and chargebacks
− indirect taxes collected for authorities
For usage pricing:
usage revenue = billable units × realized net unit price
For a marketplace:
net marketplace revenue = commission and fees
− seller or buyer credits
− refunds funded by the marketplace
− indirect taxes on marketplace fees
Separate recurring product revenue, variable usage revenue, one-time product revenue, implementation and services, pass-through amounts, advertising or affiliate revenue, and financing or other non-product income.
Pass-through is the line that flatters a business most. Money that arrives and leaves again is not revenue you can spend, and including it makes every ratio below it wrong.
Each stream can have different margins and retention. Combining a high-margin subscription with labor-intensive implementation can conceal both product strength and service weakness.
Track contracted, billed, recognized and collected revenue where timing differs. A signed annual contract does not immediately become cash or fully recognized revenue.
Gross margin versus contribution margin
Definitions vary across accounting standards and companies. Consistency and reconciliation matter more than adopting a fashionable label.
A managerial gross-margin view often begins with net revenue and subtracts cost of revenue:
gross profit = net revenue − cost of revenue
gross margin = gross profit / net revenue
A contribution view can include a broader set of costs caused by serving and monetizing the unit:
unit contribution = net unit revenue
− variable service cost
− payment and channel cost
− variable support and operations
− refunds, credits, fraud and bad debt
contribution margin = unit contribution / net unit revenue
Do not silently change definitions between dashboards. Reconcile managerial metrics to finance records and label non-GAAP measures clearly.
Fixed product development, general management and base platform costs remain real. Unit contribution does not prove the company is profitable; it shows how much is available to recover acquisition and fixed costs.
A complete variable-cost ledger
Digital delivery cost is easy to underestimate because invoices arrive from many systems.
Infrastructure
Include compute, storage, database, bandwidth, observability and backup costs that change with customer activity. Allocate shared services using a defensible driver such as compute time, records, requests or active accounts.
AI and external APIs
Include model tokens, image or audio generation, embeddings, search, moderation, OCR, maps, communications, data enrichment and other supplier charges. Retries and failed workflows still consume resources.
Payments and channels
Include processor percentage and fixed fees, foreign-exchange cost, app-store or marketplace commissions, dispute fees, payout costs and billing-provider charges.
Customer operations
Include variable support, onboarding, verification, moderation, review and customer-success work caused by the unit. Use loaded labor cost.
Risk and recovery
Include expected refunds, credits, fraud, chargebacks, bad debt, service-level credits and warranty-like work.
Delivery partners
Include subcontractors, licensed content, data royalties, shipping or fulfillment where the digital product coordinates a broader service.
Create a cost event ledger tied where possible to account, product, feature and usage event. Aggregate invoices alone cannot explain why a segment became expensive.
Shared costs and false precision
Some costs are neither fully fixed nor directly traceable. A database cluster may step up after aggregate growth. A support team has base capacity and becomes variable over a longer horizon.
Use three categories:
- direct variable — clearly caused by a unit, such as an API call;
- capacity-variable — grows in steps, such as support headcount or database tier;
- fixed within the decision horizon — does not change because of the decision being tested.
For short-term overage pricing, direct incremental cost may be sufficient. For annual planning, capacity-variable cost matters.
Document allocation drivers. Examples: storage by gigabyte-month, compute by job duration, support by ticket minutes, success by account allocation and shared infrastructure by weighted active usage.
Avoid distributing every company expense evenly across customers. That can make a small account look unprofitable because of costs it does not cause. Also avoid declaring all team cost fixed to make contribution appear attractive.
The useful operational level for cost
Average cost per customer can hide extreme variation.
Calculate where relevant:
cost per successful workflow = total workflow variable cost
/ successful customer workflows
cost per accepted AI output = generation, retry and review cost
/ accepted outputs
contribution per transaction = net transaction revenue
− payment, protection, support and fulfillment cost
account contribution = account net revenue
− attributable service and variable operating cost
Use percentiles, not only averages. A small number of workloads may create most infrastructure cost. Support-heavy customers can distort an otherwise healthy package.
Analyze failed units. A failed payment, abandoned AI run or disputed marketplace transaction can create cost without revenue. Success-only denominators may hide this burden.
Fixed-cost leverage
Software is attractive partly because product development can serve additional customers. That does not make fixed costs irrelevant.
operating profit = total contribution
− fixed product, sales, administrative and platform costs
break-even units = fixed costs / contribution per unit
This simple formula assumes stable contribution and capacity. In reality, costs step up. Add thresholds for infrastructure, support, compliance and management.
Create a capacity map:
| Resource | Current capacity | Step-up trigger | Annual step cost |
|---|---|---|---|
| Core infrastructure | 20,000 active accounts | 16,000 | €90,000 |
| Support | 1,500 tickets/month | 1,250 | €55,000 |
| Verification | 8,000 checks/month | 7,000 | €36,000 |
Growth near a threshold can temporarily reduce margin before new capacity is utilized. This is different from structurally negative unit contribution.
An honest acquisition-cost calculation
The simple formula is:
customer acquisition cost = acquisition spend
/ new acquired customers
The difficult part is defining spend and acquired customers.
Acquisition spend includes advertising, media production, agencies and affiliates, sales compensation, solution engineering, events and sponsorships, acquisition tools and data, the labour behind content and SEO, partner incentives, referral rewards, and a fair allocation of management time.
Content and SEO labour is the line most often left out, because it is salaried rather than invoiced. Leaving it out makes organic acquisition look free and paid acquisition look worse than it is.
Use loaded labor cost where teams spend material time on acquisition. Excluding founders can make early CAC look unrealistically low.
Report CAC separately: blended, paid-channel, organic programme, sales-assisted, self-serve, new-logo, reactivation, and the cost of acquiring expansion.
Blended CAC is the number to publish and the worst one to decide with. It moves whenever the mix moves, and the mix moves for reasons unrelated to the efficiency of any channel.
An “organic” customer is not necessarily free. Content, community, product loops and brand require investment. Their cost may be shared over a longer period, so cohort attribution needs a stated policy.
Match acquisition spend to the right cohort
Marketing spend this month may create customers next month. Enterprise sales cost can precede a contract by several quarters.
Use lag analysis. Connect lead creation, qualified opportunity, close and activation dates. For self-serve paid media, attribution windows may be short. For content and sales, use multi-period cohort analysis.
Avoid dividing this month’s spend by this month’s customers when demand has strong seasonality or a long cycle.
One approach is to assign direct campaign and sales costs to opportunities, then allocate programme costs across the cohorts they influence. Another is to use a rolling period long enough to smooth timing.
Always retain a cash view. Even when attribution assigns cost to a future cohort, the company paid the bill earlier.
CAC by segment and channel
Blended CAC can produce dangerous decisions.
Suppose:
| Segment | New customers | Acquisition cost | CAC | Monthly contribution |
|---|---|---|---|---|
| Self-serve teams | 500 | €75,000 | €150 | €32 |
| Sales-assisted mid-market | 40 | €96,000 | €2,400 | €390 |
| Enterprise | 6 | €108,000 | €18,000 | €2,100 |
None can be judged by CAC alone. Payback is approximately 4.7, 6.2 and 8.6 months before considering contribution changes, implementation and collection timing. Retention quality may also differ.
Within each segment, compare channels. A channel can bring low-cost customers who churn quickly or high-cost customers with strong expansion. Optimize acquired contribution, not cheapest signup.
Payback as a cash-risk measure
CAC payback asks how long retained contribution takes to recover acquisition cost.
A rough formula is:
CAC payback months = CAC / monthly contribution per customer
This assumes stable monthly contribution. A cohort-based calculation is better:
cumulative cohort cash contribution at month n =
collected cohort revenue through month n
− service and collection cost through month n
Payback occurs when cumulative contribution equals acquisition cash outflow.
Account for annual prepayment, monthly collection, implementation cost, usage that ramps, early churn, expansion, refunds, sales commissions paid before the money arrives, and delayed invoices.
Commissions paid before collection are what make a growing company short of cash. The sale is booked, the commission is paid, and the invoice is on sixty-day terms.
A subscription can have strong accounting margin and weak cash payback. Annual billing can improve cash timing even when the nominal discount lowers total revenue.
Payback matters because capital is finite. A profitable cohort with 30-month payback can exhaust cash during rapid growth.
Measure retention before estimating lifetime value
Retention determines whether acquisition creates a durable asset.
Logo retention
logo retention = customers retained from opening cohort
/ customers in opening cohort
Gross revenue retention
gross revenue retention =
opening recurring revenue − churn − contraction
/ opening recurring revenue
Expansion is excluded.
Net revenue retention
net revenue retention =
opening recurring revenue − churn − contraction + expansion
/ opening recurring revenue
Define whether usage fluctuation, currency movement, service revenue and reactivations are included. Keep the metric comparable across periods.
Analyze retention by acquisition cohort, segment, package, channel, geography and activation state. A blended curve can hide that one channel attracts poor-fit customers.
Revenue retention does not replace product outcomes. A customer can remain contracted while not adopting, creating renewal risk that has not yet appeared in revenue.
Build retention curves correctly
A cohort table follows the same group from acquisition through subsequent periods.
| Cohort | Month 0 | Month 1 | Month 3 | Month 6 | Month 12 |
|---|---|---|---|---|---|
| January | 100% | 91% | 80% | 71% | 61% |
| February | 100% | 93% | 82% | 73% | — |
| March | 100% | 94% | 84% | — | — |
Use both customer and contribution retention. If the retained customers are the most expensive to serve, logo retention can look healthy while contribution deteriorates.
Watch for censoring. Recent cohorts have not had time to reach month 12. Do not average missing future observations as zero or compare immature cohorts directly with mature ones.
Seasonality and annual contracts require appropriate windows. Contracted retention may only become visible around renewal, while usage gives earlier evidence.
Lifetime value from contribution
Lifetime value should represent expected economic contribution, not revenue.
For a stable subscription with approximately constant churn, a shortcut is:
LTV ≈ average recurring revenue per account
× contribution margin
/ periodic account churn rate
This shortcut is fragile. It assumes a stable churn process, constant contribution and an infinite geometric lifetime. It behaves absurdly when churn is near zero and ignores discounting.
A better model forecasts cohort contribution by period:
lifetime contribution value = Σ from t=1 to T
expected retained contribution at t / (1 + discount rate)^t
A forecast can include churn and contraction, expansion, price changes, changing usage cost, support cost by lifecycle stage, refunds and bad debt, and a conservative terminal assumption.
The terminal assumption does most of the work in the result. Stating it explicitly is what stops a lifetime-value number from being an argument rather than an estimate.
Use an evidence horizon. If only six months of data exist, show observed contribution for six months and scenario ranges beyond it. Do not present a five-year LTV as measured fact.
LTV to CAC: what the ratio hides
LTV:CAC = expected lifetime contribution / acquisition cost
The ratio omits timing. Two cohorts can have equal LTV:CAC but radically different payback.
Use at least four views together:
- CAC;
- payback period;
- observed retained contribution;
- scenario LTV:CAC.
A high ratio may indicate underinvestment, but it may also reflect overstated lifetime, omitted cost or a small high-intent channel that cannot scale.
A low ratio may indicate poor pricing, retention, qualification or delivery economics. Cutting marketing is only one possible response.
Avoid universal targets. Capital cost, risk, stage, segment and growth opportunity affect the acceptable range.
Expansion economics
Expansion can arise from more seats, usage, workspaces, locations, modules or service level. It improves lifetime value only if incremental contribution is positive.
expansion contribution = net expansion revenue
− incremental service, success and sales cost
Track cost to acquire expansion separately. Account management, customer success and sales effort are not free.
Determine whether expansion is: organic product usage, contract true-up, sales-led cross-sell, price increase, currency movement and service attachment.
These have different predictability and cost. Net revenue retention lifted by price increases is not the same operational evidence as wider successful adoption.
Contraction and downgrades
Churn is not the only loss. Customers can reduce seats, usage or package scope.
Build a revenue bridge:
closing recurring revenue = opening recurring revenue
+ new revenue
+ expansion
+ reactivation
− contraction
− churn
+ price and currency effects
Analyze contraction reasons. Over-purchased plans indicate packaging problems. Usage decline may reveal poor outcomes. Seasonal contraction may be normal for consumption products. Currency changes should be separated from customer behavior.
A save offer can preserve revenue while reducing contribution if it creates permanent support or discount obligations. Measure retained economics after intervention.
Refunds, bad debt and credits
Booked revenue that is routinely reversed should not support acquisition decisions.
Track by cohort: trial and first-payment refunds, chargebacks, invoice non-payment, service credits, marketplace claims, promotional credits, and unused prepaid liability where it applies.
Unused prepaid liability is money on your account that belongs to the customer. Treating it as revenue is how a good cash quarter becomes a bad accounting one.
realized revenue rate = collected net revenue
/ initially billed consideration
High refund rates can signal misleading acquisition, poor activation or billing confusion. Bad debt can differ substantially between card, invoice, geography and segment.
Use expected loss in quotes where payment terms or customer risk vary.
Connect pricing to unit economics
Price changes affect more than revenue per customer.
A higher price can change conversion, the mix of customers you attract, payment failures, the sales cycle, support expectations, retention, whether an acquisition channel remains viable, usage behaviour, fraud incentives, and expansion.
Channel viability is the second-order effect that surprises people. A price change that halves conversion can make a paid channel unprofitable even though revenue per customer rose.
A lower price may improve conversion but attract customers whose support cost exceeds contribution. A usage model may align revenue with value but make budgets uncertain. An annual discount reduces nominal revenue while improving cash and sometimes retention.
Evaluate pricing scenarios at the cohort level:
expected contribution per eligible prospect =
paid conversion
× expected retained customer contribution
− variable pre-purchase cost
This is more informative than maximizing conversion or average revenue separately.
Usage-based and AI economics
Average gross margin can hide workloads whose cost scales nonlinearly.
Create a cost model by event:
workflow variable cost = model and API calls
+ compute and storage
+ retries and failures
+ review or moderation
+ support and abuse allowance
Then connect events to account revenue.
usage contribution = net usage revenue
− metered variable workload cost
Track cost-to-revenue distribution by account and plan. Identify heavy users, inefficient features and failure loops.
A credit meter can simplify the customer experience while internal costs remain granular. Review credit conversion when supplier prices, routing or product behavior changes.
“Unlimited” plans need fair-use boundaries, capacity controls or portfolio economics strong enough to absorb extreme use. Do not rely on the average while a few accounts can create unlimited liability.
Marketplace economics on both sides
A marketplace should not use gross merchandise value as revenue.
gross take rate = marketplace fee revenue / defined GMV
transaction contribution = marketplace fee revenue
− processing
− protection, fraud and disputes
− variable support
− fulfillment or verification
− transaction-funded incentives
Acquisition economics exist for buyers and sellers. Incentives may subsidize one side to create liquidity.
Useful cell-level analysis includes:
contribution per eligible visit = match probability
× completion probability
× contribution per completed transaction
− visit-level variable cost
Segment by category, geography and transaction type. A liquid cell can subsidize an immature one deliberately, but the transfer should be visible.
Model enterprise deal economics separately
Large contracts often require sales engineering, security review, legal work, implementation and support.
first-year enterprise contribution = collected first-year revenue
− product delivery cost
− implementation and account delivery
− deal-specific sales, legal and assurance cost
− expected credits and bad debt
Also calculate steady-state recurring contribution. A contract can have weak first-year economics but strong lifetime value if implementation is bounded and renewal is likely.
Do not count unrecovered custom engineering as general product investment unless there is evidence of broad reuse. Attribute the customer-specific portion to the deal.
Compare standard and exception-heavy cohorts. A few prestigious logos can create revenue while weakening overall capacity and roadmap economics.
Free-user economics
Freemium, trials and public APIs create cost before revenue.
free cohort cost = infrastructure and supplier cost
+ support, moderation and abuse
+ onboarding and communication
free-to-paid cohort contribution = paid conversions
× expected paid lifetime contribution
− total free cohort cost
− incremental acquisition cost
A free plan can also generate referrals, content, liquidity or data value. Estimate these carefully rather than treating every free user as a lead.
Segment free users by activation and conversion potential. Cost limits should target low-value abusive behavior without breaking the experience for qualified users.
Sensitivity analysis
Unit economics depend on uncertain inputs. Test how conclusions change.
For a subscription, vary conversion, the realised price, contribution margin, early churn, long-term retention, expansion, CAC, and collection timing.
Early churn and long-term retention deserve separate variables. They have different causes, respond to different work, and averaging them hides which one is actually moving.
For an AI product, also vary supplier cost, retry rate and model mix. For a marketplace, vary match rate, incentives, disputes and take rate.
Create a matrix.
| Scenario | CAC | Monthly contribution | Month-12 retention | Payback | 24-month contribution after CAC |
|---|---|---|---|---|---|
| Downside | €420 | €38 | 48% | 15 months | €90 |
| Base | €360 | €45 | 62% | 9 months | €410 |
| Upside | €320 | €51 | 72% | 7 months | €690 |
If the decision only works under the upside case, it is not yet robust. Identify which assumptions can be tested cheaply.
A worked B2B SaaS example
Illustrative scenario: the figures are assumptions for the calculation, not observed results from a real project.
A workflow product sells two packages.
| Input | Team | Business |
|---|---|---|
| Net monthly revenue/account | €120 | €620 |
| Infrastructure and APIs | €14 | €86 |
| Payment cost | €4 | €8 |
| Variable support and success | €12 | €105 |
| Expected refunds and credits | €2 | €6 |
| Monthly contribution | €88 | €415 |
| New-logo CAC | €520 | €5,200 |
Simple payback is approximately 5.9 months for Team and 12.5 months for Business. However, the Business package also requires a €3,000 implementation project that costs €2,100 to deliver. The customer prepays the annual subscription but pays implementation after acceptance.
Cohort analysis shows:
- Team month-12 logo retention of 58%;
- Business year-one logo retention of 88%;
- Team expansion is limited;
- Business accounts expand contribution by 12% after the first year;
- one sales channel produces Business CAC of €8,400 and lower retention;
- accounts launched within 30 days retain materially better.
The company takes four actions:
- stops the weak acquisition channel rather than judging all Business customers by blended CAC;
- standardizes implementation to reduce time to value and first-year delivery cost;
- preserves annual prepayment despite pressure for monthly invoices;
- changes sales qualification to require an implementation owner.
It does not automatically raise Team prices or eliminate the Business package. The unit model identifies the actual constraints: channel quality and implementation.
Build a trustworthy data model
A unit economics dashboard needs stable identifiers across billing, product analytics, CRM, support and finance.
Maintain mappings from user to account, workspace to billing entity, account to legal customer, subscription to invoice, acquisition source and campaign, product events to usage cost, support and service work, refunds, disputes and credits, and the contract and package version.
These mappings are what make unit economics computable at all. Without them, every number in this article is an estimate assembled by hand each quarter.
Store event timestamps and monetary currency. Preserve original currency and conversion rate used for reporting.
Use immutable source events where possible. A corrected invoice should not erase the history needed to explain a cohort.
Reconcile totals. Sum of account net revenue should match the financial source within documented differences. Sum of attributed cloud cost should reconcile to supplier invoices plus an explicit unallocated amount.
A precise-looking dashboard without reconciliation is dangerous.
A review cadence
Weekly operational review
Monitor payment authorization, refunds, usage cost anomalies, support load and newly acquired cohort quality. Act on failures quickly.
Monthly economics review
Review revenue bridge, contribution by segment, CAC, payback progression, retention, expansion and capacity thresholds. Compare actuals with cohort forecasts.
Quarterly strategic review
Revisit package economics, channel allocation, price, supplier contracts, segment focus and fixed-capacity investments. Update LTV scenarios only when evidence changes.
Assign metric owners. Finance can own reconciliation, product can own usage and activation, growth can own acquisition attribution, and customer teams can own service and retention inputs. Shared definitions prevent each function from publishing its own margin.
Use economics to make decisions
A dashboard should produce action.
If contribution is weak
Investigate supplier routing, infrastructure efficiency, packaging, minimums, overage, support design, fraud and price.
If CAC is high
Investigate segment fit, qualification, channel saturation, sales productivity, creative, conversion and attribution before simply cutting spend.
If payback is slow
Consider annual prepayment, faster activation, lower acquisition cost, higher contribution and better early retention.
If retention is weak
Investigate customer expectation, activation, product outcome, reliability, package fit and acquisition quality. Price discounts rarely cure missing value.
If LTV appears extremely high
Challenge the retention horizon, expansion assumptions, omitted service cost and channel scalability.
If growth creates cash pressure
Model collections, supplier payment timing, commissions, implementation and capacity step-ups. Profitable growth can still require financing.
Common failure modes
Using revenue instead of contribution for LTV
This ignores delivery cost and overstates the amount available to recover acquisition.
One CAC for the whole company
Blended CAC hides channel, segment and sales-motion differences.
Treating founder labor as free
Early economics look attractive because essential selling, support and implementation work is omitted.
Assuming current gross margin is marginal margin
Capacity step-ups, support and supplier tiers can make the next cohort different from the average.
Using one month divided by churn
Short-term churn is noisy, cohort age differs and expansion is ignored. Build retention curves.
Ignoring cash timing
Annual contract value can take months to collect while acquisition and delivery costs are paid now.
Mixing GMV, billings and revenue
Marketplace volume, invoice value, recognized revenue and cash are distinct.
Allocating every fixed cost to each customer
This can obscure whether the incremental unit is valuable. Keep contribution and company profitability as connected but separate views.
Excluding refunds and bad debt
The model optimizes invoices that never become retained cash.
Forecasting beyond the evidence
A five-year LTV from three months of data is a scenario, not an observation.
Optimizing averages
Heavy workloads, support-intensive accounts and weak channels disappear inside blended metrics.
Unit economics checklist
Definitions
- Define the decision and economic unit.
- Define customer, account, activation, churn and expansion states.
- Choose cohort date and time grain.
- Separate contracted, billed, recognized and collected revenue.
- Publish a metric dictionary and owner.
Revenue and cost
- Build a bridge from list price to net revenue.
- Separate product, usage, services and pass-through streams.
- Identify direct, capacity-variable and fixed costs.
- Include infrastructure, APIs, payment, support, refunds and risk.
- Reconcile account-level totals to financial and supplier records.
- Measure failed as well as successful operations.
Acquisition
- Include paid media, sales, tools, programmes and labor.
- Match spend to the correct acquired cohort.
- Separate self-serve, sales-assisted, paid and organic economics.
- Calculate CAC by segment and channel.
- Track cash outflow timing.
Retention and lifetime value
- Build logo, revenue and contribution retention curves.
- Separate churn, contraction, expansion and price effects.
- Use observed cohorts before long-range forecasts.
- Calculate discounted expected contribution.
- Review payback alongside LTV:CAC.
- Publish downside, base and upside scenarios.
Decisions and governance
- Review package and account contribution.
- Connect anomalies to product and billing events.
- Track capacity step-up thresholds.
- Assign weekly, monthly and quarterly review cadences.
- Record decisions and expected metric effects.
- Re-estimate only when evidence or economics change.
Unit economics are not a certificate that a startup is healthy. They are a disciplined way to connect customer behavior, product delivery and cash. A positive contribution does not remove fixed costs. A high LTV estimate does not create immediate liquidity. A cheap acquisition channel does not guarantee retained value.
The useful model is the smallest one that captures the decision’s real cash flows, exposes uncertainty and can be reconciled to reality. When that model is segmented and reviewed consistently, it becomes more than a finance exercise: it guides pricing, packaging, product efficiency, customer success and growth allocation.
