How to get referrals from customers
One good referral conversation is a skill. Thirty of them a quarter is a design problem, and the design usually goes wrong in the same four places.
By Jerry Feng, Growth at Happenstance. Updated August 28, 2026.
The short version
A referral motion is a habit inside meetings you already hold, not a page on your website. Count what you get today, attach the ask to an existing meeting, write down what counts as a referral before you reward one, and track refusals next to referrals. Formalize it only after the manual version has survived a quarter.
Why most referral programs die in their second quarter
The economics work. Schmitt, Skiera and Van den Bulte tracked almost 10,000 accounts at a German bank over 33 months and found that “the lifetime value of referred customers, measured over a six-year horizon, was 16 % higher, on average, than that of non-referred customers with similar demographics and time of acquisition” (2013), with churn 18 percent lower. Their reward was 25 euros against a 40 euro value difference, a 60 percent return over six years by their own conservative estimate.
Referred customers stay longer, and a program can pay for itself on retention alone, which makes the failure rate worth explaining.
Four things kill these programs, and none of them is the arithmetic. Nobody owns the process, so names arrive and sit. The reward is designed by people who will never receive it, which is how you end up offering a gift card to a VP who isn't allowed to accept one. The only thing measured is referrals received, a lagging number that can't tell you whether anybody asked. And the program gets built before anyone has proved the manual version works, so there is nothing to automate from.
The same paper has the warning in it. Rather than the “all in” approach, the authors write, a referral scheme should be targeted so that attractive prospects are the ones pulled in, and monitored to check acquisition costs don't exceed the value that follows. A program that treats every customer as an equally likely referrer is the version that produces nothing and then gets quietly switched off.
Doing this with Happenstance
Knowing which customer can reach which prospect is the piece a referral motion needs and no CRM holds. A CRM records the relationships your company has bought. It has no view of the relationships your customers and your own team already carry. Happenstance searches those: everyone connects Gmail, Outlook, Google Calendar, Google Contacts, LinkedIn, Twitter and Instagram, groups pool those networks, and the whole pooled set answers questions in plain English. Ask “Who do we collectively know at Series B fintechs in New York?” and the answer names the people, who holds each relationship and the evidence behind it.
That turns the quarterly round from a broadcast into a shortlist. Instead of asking twelve customers to search their own memory, you ask three of them about one name each. Happenstance is free. The free plan is the whole product, with usage limits on chat and web searches beyond your network. Pro raises the limits: unlimited web searches and results, higher chat limits, and CSV export.
The team-wide version of the same question is account mapping, and Happenstance for sales covers how a revenue team runs it.
The quarterly round
This is the one message in the whole motion that goes out without a name in it, and it converts worse than an individual ask by design. Its job isn't to produce referrals. Its job is to find out which customers are willing, so the individual conversations that follow go to the right ten people.
The quarterly round, sent to one segment
Subject: The once-a-quarter question
Everything after a reply is one to one. The single ask, with the name in it and the risk handled, is how to ask a customer for a referral, and the version that runs inside a live deal is how to ask for a referral in sales. For the accounts nobody can reach through a customer at all, the starting point is how to warm up cold leads.
What Happenstance can see
Email is read as headers only. Message content and attachments never reach our servers. Yes. Your connections are searchable only by you, the friends you accept, and the groups you join, and sharing is always mutual. Your public profile shows a constellation of named people you know by default, which you can turn off anytime. We never share, sell, or use your data to train AI models. Happenstance is SOC 2 certified and DTI Trust Level 2; the full list of what is stored and for how long is on the security page.
Frequently asked questions
How do you build a customer referral program?
Run it by hand first. Pick the twenty accounts most likely to say yes, put the ask into a meeting that already exists on the calendar, and record every ask and every answer in one place for a quarter. What you learn in that quarter is what the program should automate. Building the mechanics first is why most of them never produce anything.
Why do most customer referral programs fail?
Four reasons, in order. Nobody owns the loop, so referrals arrive and nothing happens to them. The reward is designed for the finance team rather than the referrer. Nothing is measured except the count of referrals received, which hides the fact that nobody is asking. And the program gets built before the manual version has proved anything, so there is nothing to automate.
Should you pay customers for referrals?
Cash works in consumer products and is awkward in B2B, where the referrer often can't accept it and their employer may not allow it. Access, influence over what you build, and a real introduction of their own tend to be worth more to a senior referrer than a gift card. Whatever you choose, publish the terms rather than improvising them per deal.
What should a referral program measure?
Three numbers: how many asks were made, what proportion produced a name, and how many referrers heard the outcome. Referrals received is a lagging number that tells you almost nothing about whether the motion is running. Refusal rate is the early signal, because a rate near zero means nobody is asking.
How many referrals should a customer base produce?
There's no useful benchmark, because it depends on how many of your customers have peers doing the same job elsewhere. Set the target off your own baseline instead. Count what arrived unprompted last quarter, then aim to double the number of asks rather than the number of referrals, because asks are the part you control.
When should a referral program be formalized?
When the manual version is producing more names than one person can follow up on, and not before. Formalizing means published terms, a defined reward and a named owner. It's worth doing at that point precisely because the volume has become a coordination problem, which is the only problem a program structure solves.
Make your own luck
Join 500,000+ professionals already searching their whole network.
Continue with Google