Ping Tree Optimization Guide
· LeadSalesHub
Note: this post was carried over from the previous site during the marketplace rebuild. The URL is preserved; the content is due a refresh.
A ping tree is a queue of buyers tried in order until one accepts. Simple to describe, and full of interactions that are easy to get backwards.
Ordering is not just price
Sorting buyers by price descending maximises revenue per lead in isolation. It does not maximise revenue overall, because the highest bidder is frequently the one with the tightest filters and the lowest accept rate.
The quantity that matters is expected value: price multiplied by the probability that buyer accepts this particular lead. A buyer at $42 with a 12% accept rate is worth less than a buyer at $28 with a 70% accept rate, and a tree ordered purely on price puts them in the wrong sequence.
Timeout budgets compound
Each buyer in the tree costs time. If you try eight buyers sequentially with a generous per-buyer timeout, a lead can spend a long time in flight — and every second of it is a second the consumer is cooling off.
Two things help:
- Tighten per-buyer timeouts. A buyer who cannot answer in a few seconds is not going to be the one who converts this lead anyway.
- Cap tree depth. Past a certain position, expected value is low enough that the latency cost exceeds the revenue.
Concurrency helps more than either, but it changes the semantics — you are now bidding rather than waterfalling, and you need to handle multiple acceptances.
Capacity interacts with ordering
A high-price buyer with a daily cap will exhaust it early in the day, after which the tree behaves completely differently. If you tune ordering against morning data and evaluate against full-day data, the numbers will not reconcile.
Look at accept rates by hour before concluding anything about ordering.
Duplicate windows are a hidden reorder
A buyer's duplicate window means some leads can never be sold to them regardless of position. If a large share of your volume is repeat traffic, your effective tree is shorter than your configured tree, and the buyers at the bottom see far more volume than the configuration suggests.
What we would measure
- Accept rate per buyer, per hour of day.
- Expected value per position in the tree.
- Time-in-flight distribution, not just the mean.
- Share of leads reaching the last position.
If that fourth number is high, your tree is too long or your top-of-tree filters are too tight.