Formwork

A waterfall is an ordered list of buyers. The lead is offered to each in turn and the first one who accepts it gets it. The only decision you make is the order — and the obvious answer, highest payout first, is frequently the wrong one.

The number that actually matters

What a position in the list earns you is not the payout. It is the payout multiplied by the chance that buyer takes the lead.

A buyer paying $45 who accepts 15 percent of what you send is worth $6.75 per offer. A buyer paying $22 who accepts 55 percent is worth $12.10. Putting the $45 buyer first feels right and costs you money on most leads, because the lead spends its first and best offer on someone who will probably decline.

Except that is not quite the whole calculation either, because a decline is not a loss. The lead continues down the list. If your $45 buyer declines, you still get the $22 buyer's offer immediately after.

So the real ordering rule is simpler than expected value: order by payout, but only among buyers who will realistically say yes. A buyer with a 15 percent acceptance rate costs you almost nothing to try first, because 85 percent of the time you just move on and nothing is lost.

What actually costs you to try first

Three things.

Time. Every buyer you try is a timeout you might have to wait out. Four buyers at ten seconds is potentially forty seconds before the lead reaches anyone who wanted it. In verticals where speed-to-contact drives conversion, a lead that took forty seconds to place is measurably worth less than one placed in five.

Data exposure. If you are posting full leads rather than pinging first, every buyer in front of the eventual winner receives a copy of the lead's contact details. Ordering a low-acceptance buyer first means most leads get handed to somebody who did not want them.

Quota and cost. If your outbound calls are metered — by your platform, by rate limits at the buyer's end, or by your own infrastructure — offers are not free.

That gives you the practical version of the rule: put high-payout buyers first when you are pinging, and be much more careful about it when you are posting.

Acceptance rate is not a constant

The most common mistake in waterfall management is setting an order once and treating those acceptance rates as facts. They are not. They move with:

Buyer capacity. A buyer who accepted everything in January has a full pipeline in March.
Time of day. Many buyers only take leads while their call centre is staffed, and simply reject everything outside those hours.
Duplicate windows. If a buyer suppresses anyone they have seen in 30 days, your acceptance rate against them falls as your own audience overlap grows.
Lead quality drift. A creative change upstream can shift who qualifies without anything in your routing changing at all.

The fix is not a smarter algorithm. It is looking at the actual numbers every couple of weeks and moving rows up and down.

Use conditions instead of ordering, where you can

If a buyer only covers four states, do not put them in a national waterfall and let them decline 80 percent of your volume. Put them behind a condition that only sends them leads from those states, where they can sit at the top and win nearly everything.

This is generally the highest-leverage change available. It converts a low acceptance rate into a high one by removing the offers that were never going to work, which improves your latency, your data exposure and your fill rate simultaneously.

The same applies to time-of-day. A buyer who rejects everything overnight should be behind a condition that reflects their hours, not sitting at the top of your list soaking up a timeout on every 3am lead.

Two ordering traps

The default is not a fallback. In most systems, including this one, the default waterfall runs when no condition matched. It does not catch leads from a matched branch whose buyers all declined. If you want a matched branch to fall through to your general buyers, those buyers need to be at the end of that branch's own list too. This catches nearly everyone once.

Disabling is better than deleting. When a buyer pauses, disable them rather than removing them from your waterfalls. They keep their position in every list, get skipped at run time, and slot back into the right place when they return. Deleting means rebuilding your order from memory later.

A reasonable starting order

With no data yet:

Buyers behind tight conditions, where they are the natural fit and will accept most of what they see.
High payout, if you are pinging first.
High acceptance rate, if you are posting full leads.
Everyone else, by payout.
Your lowest-payout buyer who takes nearly everything, at the bottom, as the floor.