Skip to content
ZappushZappush

Decide what trains
the algorithm.

Ad platforms optimise toward whatever you send them. Send one generic Purchase and they hunt the cheapest identical conversion. Signal engineering means sending the outcomes you actually want, new customers, qualified leads, high-value orders, each with a value attached, server-side.

Book a Demo

From $59/month · 14-day free trial · Cancel anytime

What the platform receives

Raw events

Purchase $12

promo, repeat buyer

Purchase $900

first-time customer

Lead —

unqualified form fill

Engineered signals

Purchase

value: $12

New Customer Purchase

value: $900 · new

Qualified Lead

held until CRM confirms

Same store, same day. One version teaches the algorithm; the other confuses it.

1 signal set
engineered once, delivered server-side to Meta, Google, TikTok, and LinkedIn
Value on every event
order value, margin, or predicted LTV, so bidding can tell a $12 sale from a $900 one
New vs repeat
tagged on every conversion, so prospecting trains on genuinely new customers

Meta, TikTok, X, Snapchat, and Google's AI is only as good as the data you feed it.

Tell it to optimise for Purchase, and it optimises for any purchase.

What the platform sees

$12 promo order=$900 first-time customer

Both fire the same generic Purchase event

Junk form fill=Booked demo

Both fire the same generic Lead event

Most accounts fire a single generic Purchase or Lead event, so every conversion looks identical to the platform. A $12 promo order weighs the same as a $900 first-time customer, and a junk form fill counts as much as a booked demo. The bidding algorithm does exactly what it was told: it finds the cheapest, most frequent version of that event, repeat discount buyers and low-intent leads.

The result shows up slowly. CPA looks fine while new-customer acquisition cost quietly climbs, prospecting budget re-buys customers you already own, and reported ROAS drifts further from real, profitable growth each month. Piling on dozens of thin custom events makes it worse, not better: the model learns fastest from a few strong signals, not many weak ones.

The signals you can send.

Each engineered signal is a first-class conversion event with a value attached. Most have a dedicated feature behind them.

New Customer Purchase

First-time buyers, resolved off the profile's purchase history, so prospecting trains on genuinely new customers.

Know more

High-Value Order

Orders above your value threshold as their own event, so bidding can chase your best buyers specifically.

Know more

Order Value / Predicted LTV

A revenue-weighted value on every conversion: raw order value, margin-adjusted, or predicted lifetime value.

Know more

Qualified Lead

CRM stages like qualified and closed-won sent back as conversions, so platforms stop optimising for raw form fills.

Know more

Subscription Start & Renewal

Sign-ups and renewals as separate events, so platforms learn to find subscribers who stay, not one-off trials.

Know more

Profit Value

Contribution margin after COGS, shipping, discounts, and fees, so bidding optimises for what you actually earn.

Know more

What the platforms do with them.

Engineered signals are not a workaround. They are the input the platforms' own bidding levers were built for.

  • Campaigns optimise on your outcome, not the default. New Customer Purchase or Qualified Lead replaces generic Purchase and Lead as the event delivery learns from.

  • Value-based bidding runs on the values each signal carries, per auction, on Google and Meta alike. The engineered value is the bid input.

  • Server-side delivery keeps the signal usable. Hashed identifiers and shared event IDs mean high match quality and no double counting.

The full bidding mechanics, value rules, multipliers, and new-customer goals → Value-Based Bidding

How it works

  1. Decide which outcomes count

    Pick the business events worth optimising for, new customer, qualified lead, high-value or high-margin order, subscription start, instead of defaulting to one generic Purchase. Zappush maps each to the exact platform event and value it should train.

  2. Enrich and weight each signal

    Every event is stamped with a revenue-weighted value, a new-vs-repeat flag resolved off the persistent profile, and category metadata. A first-time $900 order carries a very different weight than a repeat promo purchase.

  3. Deliver server-side to every platform

    Signals fire from your server via Meta CAPI, Google Enhanced Conversions, TikTok Events API, and LinkedIn, deduplicated with a shared event ID and matched on hashed identifiers, so they survive browser limits and arrive with value attached.

Questions, answered.

Isn't sending more custom events always better?
No. Too many low-quality signals confuse the algorithm and slow learning. Signal engineering is about choosing fewer, stronger outcomes, new customer, qualified lead, high-value order, and sending them reliably with value attached, so the model learns faster on what actually matters.
How does this help the platform find high-value customers?
Zappush attaches a revenue-weighted value (order value, margin, or predicted LTV) and a new-vs-repeat flag to every event, then delivers it server-side via CAPI. That lets value-based optimization chase higher-value and net-new buyers instead of treating a $10 promo order the same as a $900 first-time customer.
Does signal engineering work across all my ad platforms?
Yes. The same engineered signal set, deduplicated with a shared event_id and matched on hashed PII, is delivered server-side to Meta (CAPI), Google (Enhanced Conversions / offline import), TikTok (Events API), and LinkedIn, so value and identity stay consistent across every channel instead of being rebuilt per pixel.

Find out what your account is training on today.

Start free in minutes, or see it live on a call.

Book a Demo

From $59/month · 14-day free trial · Cancel anytime