The State of Referral Marketing 2026
Benchmarks from 42 real referral programs, each verified against the brand's own published terms on a recorded date. Among them: 89% run double-sided rewards, every published program attaches a qualifying action, and 19% differ materially from what popular roundups still report.
of verifiable running programs reward both sides (31 of 35)
of programs that publish terms attach a qualifying action; nobody pays on an unqualified signup (28 of 28)
of published programs state a reward cap (25 of 28)
of programs had materially changed or been retired versus what circulating roundups describe (8 of 42)
Where these numbers come from
Every figure on this page is counted from ReferralFlo's library of 42 brand referral program teardowns, covering 8 regions. Each teardown records the program's reward, cap, expiry and qualification terms as read from the brand's own published pages or PDFs, never a coupon site or aggregator, with the verification date rendered on the page. The corpus behind this edition was verified on 2026-08-04.
Where a brand runs a program but does not publish an amount, the teardown says so and no figure is invented; those programs count toward structure statistics but not reward-value ones. No currencies were converted and no values estimated, which is why this report measures program structure (sidedness, reward type, guardrails, symmetry) rather than averaging reward amounts across fourteen currencies into a number that would mean nothing.
The sample skew is disclosed rather than hidden: 30 of 42 programs are financial services (fintech, investing, payments, broking), the sector where public reward terms are richest. Read these figures as a benchmark for published consumer programs, weighted toward finance, not a census of all referral marketing.
Double-sided rewards are the default, not a best practice
Of the 35 running programs whose structure is verifiable, 31 reward both the referrer and the friend. Only 2 are deliberately one-sided, and 2 more publish only the referrer's side. Rewarding both sides is no longer a design choice to debate. It is the market's baseline.
Upstox (referrer only) and GXS Bank, whose current savings terms state the programme pays the referrer only.
Zerodha and 5paisa publish the referrer reward but no friend-side amount.
Why the base is 35, not 42: 38 of the 42 programs are running, and 35 of those state enough of their terms for sidedness to be verified. How double-sided rewards work →
Cash is the dominant reward, but a third of programs pay in something smarter
Half of the 28 published programs pay cash or cashback. The rest pay in points, stock, storage, fee waivers or revenue share: rewards that cost the brand less than face value, reinforce the product, or compound loyalty. The reward type distribution is a menu of alternatives to discounting.
Instacart, Octopus Energy, Klarna, Syfe, GXS Bank
Toss, Traveloka, Kredivo
Zerodha, Kotak Securities, 5paisa
Robinhood, Spaceship, Superhero
Dropbox storage, Peloton membership, StashAway fee waiver
Boost Bank
Casper
Classified by the reward's primary component, referrer side, among running programs with published terms. Reward ideas by type →
Every published program qualifies the reward; most cap it; expiry is the overlooked term
All 28 published programs require a qualifying action (a funded account, a completed order, a first trade) before anything pays out. 25 state a reward cap. Only 18 publish an expiry, which is exactly the term most likely to surprise an advocate later.
If a program in this set pays on bare signups, it does not publish that fact. Design yours the way the whole market does: reward the qualified action, cap the liability, and say when rewards lapse. What belongs in program terms →
When both sides are published, most brands pay them equally
Among the 16 double-sided programs whose two rewards are stated in directly comparable terms, 10 pay referrer and friend exactly the same. The rest split evenly: 3 favour the friend and 3 favour the referrer. The often-repeated advice to always overweight the friend's side is not what published programs actually do.
Casper, Octopus Energy, Robinhood, Sharesies, Spaceship, Superhero, StashAway, Boost Bank, Kredivo, Dropbox
Instacart, Peloton, Syfe
Klarna, Kuda, Touch 'n Go eWallet
Pairs paid in different units (points against discounts, waivers against tiers) are excluded rather than converted; 15 double-sided programs fall outside this comparison for that reason.
Sector signatures: brokers share revenue, wallets run lotteries
Reward design clusters by sector. All three Indian broking programs in the set couple an upfront amount with an ongoing share of the referred client's brokerage. Randomised rewards appear only in investing and payments: Robinhood's odds-based free stock, and the mystery-amount draws MoMo and Maya run in Southeast Asia.
- Zerodha, Kotak Securities and 5paisa all pay a continuing brokerage share (10%, 15% and 30% respectively): referral as an annuity, unique to broking in this set.
- Robinhood, MoMo and Maya use randomised reward amounts with published odds or ranges: a lottery mechanic that keeps average cost low while advertising the jackpot.
- Fintech is the corpus's largest sector (15 programs) and behaves most conventionally: cash or cashback, capped, on a funded-account qualification.
19% of programs are not what the internet says they are
8 of the 42 programs (roughly one in five) had materially changed their terms or been retired entirely, while widely-circulated example roundups still describe the old version. Referral program terms drift constantly; undated referral content quietly becomes wrong.
This is the report's strongest argument for its own method. During verification, one bank had four referral pages live at once quoting four different reward amounts, and several vendor example libraries still present long-retired programs as live. It is also why every figure here carries its verification date — including this page's.
Check any program before citing it: all 42 dated teardowns →
How to use these benchmarks
As a design baseline
Structure transfers between businesses far more reliably than magnitude. Copy the market's structure (double-sided, qualified, capped, with a stated expiry) and size the amounts from your own economics, not from another company's reward. The reward calculator does that math, and the benchmarks guide covers how to read any external figure, including this one.
Citing this report
Cite any figure freely with attribution and a link. Suggested form:
ReferralFlo, "State of Referral Marketing 2026", data verified 2026-08-04. https://referralflo.ai/state-of-referral-marketing
Unlike most statistics in this market, every number above is traceable: the programs behind each figure are named, and their teardowns link the brand's own published terms.
Where does this data come from?
Every figure is counted from ReferralFlo's library of 42 brand referral program teardowns. Each teardown records the program's reward, caps, expiry and qualification terms as read from the brand's own published pages (never an aggregator or coupon site), with the verification date on the page.
How is this different from other referral marketing statistics?
Most published referral statistics are third-party figures recycled without their definition, sample or date. Every number here states exactly what was counted, across which programs, verified when. The programs behind each figure are named, so any number can be checked.
Is the sample representative of all referral programs?
No, and the report says so. 30 of the 42 programs are financial services (fintech, investing, payments, broking), because that is where public reward terms are richest. Read the figures as a benchmark for published consumer programs, weighted toward finance.
Can I cite these statistics?
Yes, freely, with attribution to ReferralFlo and a link to this page. The underlying per-program terms are public on the linked teardown pages, so any cited figure can be traced to its sources.
How often is this report updated?
The teardown corpus is re-verified against the brands' own pages in sweeps, and this report is recomputed when the corpus changes. The verification date of the underlying data is shown at the top of the report.
Underlying data verified 2026-08-04 · report published 2026-08-28.
Run a program built like the ones that work
Double-sided rewards, qualification rules, caps and expiry are program settings in ReferralFlo, not engineering projects.