The referral moment

Why does the moment come before the program?

Because the program is plumbing and the moment is the product. A referral program's only real job is to shorten the distance between a person wanting to tell someone and them actually telling them. If you do not know when the wanting happens, the ask ends up in account settings, the footer and a monthly email, which is the arrangement almost every underperforming program has.

The seven steps below are in the order that saves time. Steps 1 to 3 take about two weeks and almost no engineering. Step 6 is the one everybody starts with.

Step 1: what counts as the value event?

One event, in the database, with a timestamp. The moment the product's promise is kept, not the moment somebody signed up or paid.

• A reporting product: a report exported or shared outside the company.

• A marketplace: an order delivered and not returned within the window.

• A consumer app: the first completed session of the thing the person came to do.

• A service with onboarding: the first week where the customer ran the process without you.

What goes wrong here. The team answers "when they see the value", which is a feeling rather than a query. Keep pushing until the answer is a table and a column. If the honest answer is that no such event is logged, that is step 1 and it is a week of engineering, not a program.

One refinement worth the extra day: separate the first occurrence from the second. The first export or first order is often a test. The second is when the person has decided.

Step 2: where does sharing already happen?

Three methods, cheapest first. Do the first one on Monday.

Look at sharing that already happens with no program. Exports, shared links, workspace invites, forwarded reports, support threads where somebody copies a colleague. These are people telling other people without any reward, and the timestamps cluster around the moment.

Look backward from your unattributed signups. Take new signups whose source is direct or unknown, and check whether an existing customer at the same email domain, or in the same city for a consumer product, reached the value event within the previous 14 days. The matches are not proof, but the pattern of what the existing customer was doing just before is a strong pointer.

What goes wrong here. Searching for a single moment when there are two. Many products have a private one, where somebody gets value alone, and a social one, where they show it to a person. The ask belongs at the social one even if the private one is more common.

Step 3: what do customers actually say?

Two questions. Who is the last person you mentioned us to, and what were you doing when you did it. Then: what did you say.

The first question tells you the moment and the relationship. The second tells you the surface. The third gives you the copy, and it will be better than anything written in a marketing meeting, because it is what one person actually says to another about your product. Write the answers down verbatim and use the phrasing that repeats.

Ten is enough. Twenty is better and rarely changes the answer.

What goes wrong here. Asking "would you recommend us" instead. That is a different question, it gets a polite yes, and it tells you nothing about when or how.

Step 4: where should the ask sit?

Three rules, and they are the whole of what most programs get wrong.

Surface. If the moment is in the product, the ask is in the product. An email about an in-product moment arrives after the feeling has passed.

Latency. Within minutes of the event, not in the next send. Willingness to tell someone decays fast.

Shape. Not a form, and not a code. Offer the thing the person is already trying to do, with the sharing built in. If somebody just exported a report to send to a client, offer to send that client a copy with a note. The referral becomes a side effect of a useful action rather than a favor.

What goes wrong here. No suppression. The ask fires at somebody with an open support ticket, a failed payment or a complaint from last week. Build the suppression list before you turn the ask on: open tickets, billing failures, and anyone whose churn score has risen, which comes from the same scorecard as churn you can see coming.

Step 5: what should the referred person land on?

The referred person's first screen decides whether any of this was worth it, and it is the half that gets built last and worst.

They should land on the thing, not on your homepage. The report, the product the friend uses, the specific page the referrer was looking at, with the referrer's name on it and anything that can be pre-filled already filled. A referred person who lands on a generic signup page has been converted back into a cold visitor, and you have spent your customer's social capital to buy a click.

What goes wrong here. The referral link drops people at the top of the funnel, so referred signups convert like paid ones and the program looks like it does not work. This is the single most common reason a well-targeted program produces flat numbers.

Step 6: what should the incentive be, and when do you pick it?

Run the mechanism with no reward for a few weeks first, because it gives you the only clean baseline you will ever get. Then test a reward against it.

If you do add one, pick something that is worth more to someone who likes your product than to a stranger. Dropbox's two-sided reward was storage, which is the product. In Drew Houston's 2010 Startup Lessons Learned deck, that program reported a permanent 60% increase in signups, 35% of daily signups coming from the program, and 2.8 million direct referral invites in the trailing 30 days to April 2010.

What goes wrong here. Cash, or a reward large enough to be interesting on its own. It changes who refers, not only how many, and the people it attracts bring people like themselves. When a referral program is just a coupon has the three checks for telling the two populations apart.

Step 7: what has to be logged before you launch?

Seven events, and one field that matters more than all of them.

• share_offered, share_started, invite_sent, invite_opened, referred_signup, referred_activated, reward_issued.

• And referrer_id written onto the referred customer's own record, permanently, in the CRM as well as the product database.

That last one is the mistake that costs six weeks. Without a referrer stamped on the referred customer forever, you cannot compare referred and non-referred cohorts on retention or revenue, you cannot tell which referrers produce customers who stay, and you cannot do anything the research below suggests. Programs launch without it constantly, because the reward logic works fine without it, and the gap is only noticed when somebody asks whether referred customers are any good. Then it is a re-launch.

What does the B2B version look like?

The moment is often not in the product at all. It is a person changing jobs.

Someone who ran a project with your product, and did well out of it, arrives at a new company with budget, authority and a story. That is the warmest referral available to a B2B company and almost nobody systematizes it. The mechanism is unglamorous: keep a list of champions, watch for job changes, and have the person who worked with them reach out as a person rather than as a vendor, within a few weeks of the move.

There is research worth reading next to this. The German bank study published in the Journal of Marketing Research in 2018 found that the best referrers to recruit were customers acquired at least six months earlier who showed high margins and low churn risk, and that referred customers' retention advantage, around a 40% lower churn hazard, lasted only as long as the referrer stayed a customer. Both findings point the same way: pick who you ask, and keep the people you asked.