Applying sector benchmarks to judge your numbers

Sooner or later someone drops Nubank's disclosed acquisition costacquisition costCustomer Acquisition Cost (CAC) is the total sales and marketing spend divided by the number of new customers gained in a period. It measures how efficiently you grow.View full definition →, single-digit US dollars per customer, onto a slide next to yours, and the room concludes your marketing is twenty times worse than Brazil's. Probably not. Nubank grows largely by word of mouth in markets where it became the default digital account, counts customers on its own definition, and earns card revenue under a regime that pays several times what a large US issuer can. The number is real. The comparison is not, because the two figures were never on the same basis.

That adjustment is the work of this lesson: which published figures are legitimate reference points for you, what you normalise before using them, and when the honest answer is that no external benchmark applies to the thing you just built.

Four questions before you use anyone's number

• What is the denominator? Starling reports accounts, and a sole trader with a personal and a business account is two of them. Ally reports deposit customers. Nubank reports customers and, separately, active customers, with an activity rate that has run in the low eighties percent. Three defensible choices, three incompatible scales. Only one of them matches the funded-account denominator the acquisition-cost lesson insists on.
• What sits in the numerator? Published acquisition costs are almost always blended across organic and paid, and rarely disclose whether referral bonuses paid in account credit, or the sponsor bank's onboarding fee, are inside the figure.
• What period, in what currency? Starling closes its year on 31 March. Nubank reports calendar quarters in dollars while earning mostly in reais, so a currency move alone shifts reported revenue per customer with no change in anybody's behaviour.
• Who published it, and what do they sell? Most onboarding conversion benchmarks come from vendors selling identity verification or account-opening software. Useful, and not neutral: their sample is their own customer base, which skews digital-first and already instrumented.

Three reference classes, and which one is yours

Ally Bank. Branchless US bank, deposits north of $140bn, more than three million retail deposit customers, revenue driven by net interest margin in the low three percents. Acquisition here is expensive per customer and the relationship is long: Ally has reported deposit retention in the mid-90s. If you are deposit-led, this is your class, and card-first benchmarks will make your costs look indefensible.

Nubank. Over 100 million customers across Brazil, Mexico and Colombia, monthly revenue per active customer around $11 by late 2024 against a cost to serve well under a dollar. Credit-card-first, so a large share of that revenue is interest on lending, not fee income your marketing can claim. Read Nubank's efficiency as a ceiling produced by scale and word of mouth in a specific market, not as a target for a paid-acquisition programme in a saturated one.

Starling Bank. Around four million UK accounts, a heavy small-business mix, pre-tax profit near £300m in the year to March 2024. Group revenue also includes Engine, the software platform Starling sells to other banks, so profit per account is not a clean banking number. The £29m FCA fine in October 2024 landed on a later reporting year, which is exactly how a single-year profit-per-customer benchmark misleads.

The adjustments that change the verdict

Card economics are set by regulation, not by your funnelfunnelThe customer journey from awareness to purchase, typically Awareness, Interest, Consideration, Decision, Action, with prospects narrowing at each stage.View full definition →

US issuers above $10bn in assets are capped by the Durbin amendment at roughly 21 cents plus 0.05% per debit transaction. Exempt smaller issuers, including most sponsor banks behind US neobank programmes, earn something closer to 1.2% to 1.5%. EU and UK caps sit at 0.2% on consumer debit and 0.3% on consumer credit. On a $40 purchase that is about 8 cents in Europe against 50 cents or so for a US exempt issuer.

Six times the revenue per swipe, for identical customer behaviour. Import a US card-first LTVLTVLifetime Value: the total revenue (or profit) a customer generates throughout their entire relationship with your business.View full definition →:CAC target into a UK plan and you have set a hurdle the regulation forbids anyone from clearing. Starling's answer was product mix: business banking, lending and subscription tiers rather than swipe volume.

Rate cycle beats vintage comparison

A savings customer acquired in 2021 at near-zero policy rates and one acquired in 2023 produce very different spread income on the same balance. Benchmarks quoting deposit acquisition cost without a date are close to useless, and your own 2021 cohort economics are not a fair baseline for today's spend either. Attach a date to every range you quote.

Funnel definitions carry local law

Starling's onboarding includes a short video selfie; a US application turns on SSN checks. Any pooled account-opening conversion benchmark is averaging different legal obligations, which is why the funnel lesson's stage map, not a headline percentage, is what you compare.

The benchmark ranges (2026, estimates)

Directional ranges from analyst commentary and public disclosures, not precise truths.

• Neobank acquisition cost: roughly $30 to $200 per funded customer in the US (estimate), often €20 to €120 in Europe where referral does more of the work.
• US retail incumbents: roughly $150 to $400+ for a checking relationship (estimate), carrying branch and brand cost.
• LTV:CAC: the 3:1 rule of thumb arrived from SaaS. It travels badly into banking, where tenure runs for years and the discount rate in the model matters more than the ratio. Use it as a smell test, and trust the discounted tenure model the LTV lesson builds over the folklore.
• Digital account opening: application start to completion commonly 40% to 70%, completion to funded 50% to 80% (estimates).
• Twelve-month retention: incumbents and deposit-led players commonly 85% to 95%; card-first apps far more variable, and much lower when measured on the active definitions the engagement lesson settles rather than on registrations.

For how these pieces fit together as unit economics, the a16z classic still holds up: 16 Startup Metrics.

Normalising before you judge: a worked pass

A US card-first programme, "MeridianPay", reports $4.00 monthly margin per funded customer and wants to know how far behind Nubank it is.

Reported: $4.00 per funded customer / month
Active share of funded base: 65%
Per ACTIVE customer = 4.00 / 0.65 = $6.15 / month

Against Nubank's roughly $11, the gap looks like a two-thirds shortfall. Strip out what is not comparable: much of Nubank's figure is interest on credit, MeridianPay has no lending book, and the two operate under different interchange regimes and different average transaction sizes. On the fee-and-interchange slice alone the gap narrows sharply. The real finding is not "we are worse at marketing." It is "we have no credit product," which is a product roadmap decision, not an acquisition one.

🎬 [VIDEO: "LTV to CAC Ratio Explained" - youtube.com - a concise walkthrough of how the ratio is built and where teams misread it]

Reading a funnel against the right peer

Suppose MeridianPay converts 25% of clicks to application starts, 45% of starts to completion, and 55% of completions to funded. Compared to a US card-first peer, the bottom two stages are weak and the fix is onboarding and a reason to fund on day one. Compared to Starling, where identity checks are heavier and a business application asks for company detail, a 45% completion rate is unremarkable. Same three numbers, two different verdicts, decided entirely by which peer you picked before you looked.

Ways benchmarks mislead you

Survivorship. Published neobank ranges come from firms still publishing. Xinja handed back its Australian licence and returned deposits in 2020 having gathered deposits without a lending product to pay for them. Its numbers are in nobody's benchmark set, so the surviving range flatters the model.

The benchmark becomes the target. A team told to hit an 80% completion-to-funded rate can drop the qualifying deposit to $1 and hit it inside a quarter, while the balance per funded account collapses. Any benchmark you pay a bonus against will be met on its own terms.

Registered is not funded. Retention figures counting anyone who has not deleted the app will beat yours every time.

Wrong peer group. A deposit-led balance sheet judged against card-first economics, or a European book judged against a US exempt-issuer one, produces confident nonsense.

Turning benchmarks into a decision

For each metric, record three things before the verdict: the peer class, the reporting basis you normalised to, and the date of the source. Then read it.

• Green: at or above the healthy range for that peer, on a matched basis. Consider scaling.
• Amber: in range but drifting, or matched only after generous adjustment. Fix the named stage.
• Red: below break-even economics, or you cannot state the basis of the number you are comparing against. Stop scaling and repair the measurement first.

Key Takeaways

• No figure is a benchmark until you know its denominator, its date, its currency and who paid to publish it.
• Interchange regulation, not marketing skill, explains much of the gap between US, UK and Brazilian per-customer economics.
• Pick your reference class first: deposit-led (Ally), credit-card-led (Nubank), mixed retail and SME (Starling). Then compare.
• Normalise to a common basis before drawing conclusions, and expect the adjustment to shrink most alarming gaps.
• Published ranges are survivors' ranges, and any benchmark tied to a bonus will be gamed. Your own dated cohort curve stays the most trustworthy reference you own.