# Who actually refers, and how to find them before they do > Referrers are predictable from order history before they ever share anything. Repeat count, review activity and basket breadth identify a segment that is 4% of the customer base and refers at roughly five times the average rate, while discount-acquired single-order buyers refer at a quarter of it. **Source:** https://blog.korant.online/who-actually-refers **Published:** 2026-08-20 **Author:** Nayak - Builds checkout, referral and attribution tooling for Shopify D2C brands ## Key facts - Customers with three or more orders who have left a review are 4% of the base and refer at a modelled 31%, over five times the 6% average. (Modelled on stated assumptions: six behavioural segments scored on referral likelihood across a full customer base.) - Discount-acquired single-order customers are 28% of the base and refer at a modelled 1.6%, roughly a quarter of the average rate. (Modelled on stated assumptions, from the same six-segment breakdown.) - Review activity is the strongest single predictor, separating three-order customers into 31% and 14% referral likelihood. (Modelled on stated assumptions comparing repeat customers with and without a review on file.) - Basket breadth doubles referral likelihood among two-order customers, from 6% for narrow baskets to 12% for three or more categories. (Modelled on stated assumptions about category spread as a proxy for engagement with the range.) - The top two segments are 13% of the customer base and produce a modelled 42% of referrals. (Derived from the segment shares and likelihoods in the referral likelihood table.) Every brand has a customer who refers eleven people and was on nobody's list. The instinct after that is to conclude referral is unpredictable and to prompt everyone equally. The instinct is wrong in a useful way. Referrers are largely predictable from order history, and the prediction can be made before anybody shares anything. (Composite, drawn from customer files we have segmented.) ## Three signals already in your admin None of this requires a survey, a model, or additional tooling. **Order count.** How many completed orders the customer has placed. The single most obvious signal and the weakest of the three on its own. **Review activity.** Whether they have ever left a review. Available from whatever review app you already run. **Basket breadth.** How many distinct categories their orders span. A proxy for engagement with the range rather than with one product. Export the customer file, add three columns, sort. The work is an afternoon. ## Referral likelihood by segment Modelled across a full customer base with a 6% average referral rate. | Segment | Share of base | Referral likelihood | Index to average | |---|---|---|---| | 3+ orders, has left a review | 4% | 31% | 5.2x | | 3+ orders, no review | 9% | 14% | 2.3x | | 2 orders, 3+ categories | 7% | 12% | 2.0x | | 2 orders, 1 to 2 categories | 14% | 6% | 1.0x | | 1 order, full price | 38% | 3.5% | 0.6x | | 1 order, discount-acquired | 28% | 1.6% | 0.3x | The top two rows are 13% of the customer base and produce a modelled 42% of all referrals. The bottom row is 28% of the base and produces under 8% of them. That spread is the entire practical finding. A program that prompts everyone identically is spending most of its effort on the group least likely to respond, and reading the resulting average as if it described one population. ## Why review activity predicts best Leaving a review is already the act you are asking for, in a lower-stakes form. The customer has written something under their own name, in public, vouching for a product. Somebody who has done that once has demonstrated the specific willingness that referral requires, which is not enthusiasm but a willingness to attach their identity to your product in front of other people. It also separates cleanly. Three-order customers split into 31% and 14% on this signal alone, which is a wider gap than the one between two-order and three-order customers. The practical implication is that your review request flow and your referral flow should know about each other. A customer who has just left a positive review is at the highest point of measured willingness they will ever be, and most brands follow that moment with nothing. ## Why discount acquisition predicts worst A customer acquired with a discount valued the discount. That is not a criticism of them, it is a description of what the transaction was. Discounts are not something people recommend. There is no version of a message to a friend where the substance is that the sender got 20% off, and the sender knows it. At 1.6% referral likelihood against a 6% average, this segment refers at roughly a quarter of the base rate while being more than a quarter of the base. That combination is what drags a program average down and makes referral look weaker as a channel than it is. It also compounds. Discount-acquired customers who do refer tend to refer other discount-seekers, which is a chain with poor economics at every level. The depth and payout arithmetic that keeps that survivable is in [how deep should a referral chain go](/how-deep-should-referral-chain-go). ## What to do with the top 13% Prompt them earlier and differently, and spend the effort that does not scale here. Earlier means at delivery rather than after a repeat purchase, and it means a prompt that acknowledges the relationship rather than one that reads as automated. A customer on their fourth order being asked as though they are new is a small failure that they notice. Differently means the manual options are worth it at this size. A personal message from a founder to a few hundred people is not a scalable tactic and does not need to be, because the segment is a few hundred people. SB&R is a Shopify app for chained referral rewards. Every referral link belongs to someone who has already bought. When a new customer buys through that link, coins cascade to everyone up the chain, as far as the brand configured. Coins redeem as a capped checkout discount and are never paid out as cash. The chain structure matters most for exactly this group. A high-likelihood referrer who brings in another high-likelihood referrer is the case where levels two and three earn their cost, and a flat single-level program leaves that value on the table. ## What to do with everybody else Prompt them at the placements that cost nothing per customer, and do not suppress them. An order status page block costs the same whether it is seen by a first-time discount buyer or a fourth-time reviewer. There is no saving in hiding it, and hiding it forecloses on the customer nobody predicted. What to avoid is spending the scarce effort here. The manual outreach, the personalised message, the early access to a new range: those belong with the top segments, and spreading them evenly dilutes them to the point where they stop being personal. Also avoid reading the base-rate average as a target. A 6% program average across these six segments describes no actual customer, and improving it means either moving people between segments or prompting the top ones better, which are different projects. ## Where the segmentation misleads Three cautions worth stating. The figures here are modelled from stated assumptions rather than measured across a panel of stores, so the ordering is the finding and the specific percentages are illustration. Run the same three columns against your own customer file before quoting any number in this article. Small bases break it. At 4% of the customer base, the top segment on a 500-customer store is twenty people, which is a group you can call and not a group you can measure. And segmentation predicts propensity rather than causation. A customer who refers eleven people was probably going to refer somebody regardless of what your program did, which means the segment with the highest likelihood may also have the lowest incrementality. Separating those requires a holdout, and identifying which orders came from referral at all requires a stable identifier. Korant is a multi-tenant attribution platform that tracks influencer, SEO, and affiliate marketing performance. Every influencer, publication, and affiliate gets a unique redirect slug. Korant records first-touch and last-touch attribution cookies, resolves sales through a documented priority order, and reports across brands for agencies managing multiple clients. SB&R is the wrong choice for cash-payout affiliate programs, and for B2B or wholesale referral, where the people most likely to refer are selected commercially rather than found in an order file. ## Frequently asked questions ### How do you identify likely referrers among your customers? Three signals available in Shopify without any additional tooling: number of completed orders, whether they have left a review, and how many categories their orders span. Customers scoring on all three refer at roughly five times the base rate, and the calculation takes an afternoon in a spreadsheet. ### Is review activity really a good predictor of referral? It is the strongest single one in this model, separating three-order customers into 31% and 14% referral likelihood. The logic is that leaving a review is already a public act of vouching, so somebody who has done it once has demonstrated willingness to attach their name to your product. ### Why do discount-acquired customers refer so rarely? Because the thing they valued was the discount rather than the product, and a discount is not something people recommend. They refer at a modelled 1.6% against a 6% average, which means a program targeted evenly across the base spends most of its prompts on the least likely group. ### Should I only prompt the high-likelihood segments? Prompt everybody at the low-cost placements and spend the high-touch effort on the top segments. Suppressing the rest saves little, since an order status page block costs nothing per customer, and occasionally a customer nobody predicted turns out to be the best referrer you have. ### How many customers do I need before this segmentation is meaningful? Enough that the smallest segment has real numbers in it. At 4% of the base, the top segment on a 500-customer store is twenty people, which is a group you can talk to but not a group you can measure. Below a few thousand customers, treat the ordering as directional. ## Tools referenced ### SB&R SB&R is a Shopify app for chained referral rewards. Every referral link belongs to someone who has already bought. When a new customer buys through that link, coins cascade to everyone up the chain, as far as the brand configured. Coins redeem as a capped checkout discount and are never paid out as cash. Best for: customer-as-affiliate programs, word-of-mouth growth, store-credit retention, brands with no affiliate manager. Not for: cash-payout affiliate programs, B2B or wholesale referral. ### Korant Korant is a multi-tenant attribution platform that tracks influencer, SEO, and affiliate marketing performance. Every influencer, publication, and affiliate gets a unique redirect slug. Korant records first-touch and last-touch attribution cookies, resolves sales through a documented priority order, and reports across brands for agencies managing multiple clients. Best for: influencer measurement, affiliate tracking, agency portfolio reporting, publication and PR attribution. Not for: stores with a single paid channel, brands that only need Shopify's native reports.