Signals and first-party data: feeding the machine

Video lecture

Signals and first-party data: feeding the machine

13 chapters · about 8 min · full transcript

Coming soon

Chapter 1 of 13

Signals and first-party data

What signals are now
Four quality tests
Hashing done right

The narrated lecture is in production

Every chapter is scripted and ready. Browse the chapters and read the full transcript now — the video will appear here when it’s published.

Chapters

Transcript of the narration, chapter by chapter.

0:00 Signals and first-party data

Two brands, same budget, same creative, same platform. One grows, one stalls. The difference is rarely the targeting. It's the signals. In this lecture you'll learn what signals actually are now that platforms control the who and the where, how to test their quality with four simple checks, and how to hash customer data correctly so it matches without breaking privacy rules.

0:27 Why it matters

Why does this matter? Because every platform now has roughly the same algorithms available to every advertiser. The only inputs that differ between you and your competitor are your creative and your signals. Here's an analogy. Two students get the same textbook and the same teacher. One gets homework marked accurately every day. The other gets it marked a week late, with some answers counted twice and some never marked at all. Who learns faster? Your conversion signals are the marking. Accurate, timely, complete marking makes the same algorithm learn dramatically better for you than for a competitor with sloppy data.

1:11 Your edge = what the platform doesn't know

Here's the core idea. When the platform decides who sees your ad, your edge is what you know that it doesn't. Which customers are profitable. Which leads actually close. Which products have margin. Who already bought last month. That knowledge reaches the platform as signals: conversion events, conversion values, customer lists, offline outcomes, product feeds, and hints like search themes. Each one quietly tells the model what good looks like.

1:41 Four quality tests

Now the four quality tests. First, coverage. What share of your real conversions does the platform actually see? Ad blockers, consent refusals and people switching from phone to laptop all chip away at it. Second, accuracy. Is each conversion counted once, with the right value and the right currency? Third, match quality. Can the platform connect the event to a person, using hashed email or phone and click IDs? And fourth, latency. How fast does the outcome arrive?

2:15 Fixing slow signals

Latency deserves a moment, especially in B2B. If a deal takes ninety days to close, the platform gets its lesson three months late, and by then the auction has moved on. The fix is to send earlier stages, like a qualified lead or a meeting held, with a value that reflects how likely that stage is to become revenue. The model learns faster, and it still learns the right thing.

2:45 Simple example: Sharjah shop (illustrative)

Here's a simple worked example. A small online shop in Sharjah sells phone accessories. They check last month: the store shows two hundred orders. Meta reports one hundred and twenty. Google reports ninety. That tells them coverage is limited. They add the Conversions API with deduplication and enhanced conversions on Google, capturing hashed email at checkout with consent. Next month the store shows two hundred and ten orders, Meta reports one hundred and seventy, Google one hundred and forty. The platforms now see more of reality, so they can learn from it. These numbers are illustrative, but the check itself takes ten minutes and you should do it monthly.

3:32 Consent is signal design

Consent isn't an afterthought here. It's part of signal design. In Europe and the UK, Google expects consent signals before it uses data for ads personalization and some measurement features. Customer lists must be collected with proper notice and, where required, consent. So the smart move is strategic. Design your business so that your most valuable signals come through channels people happily opt into, like accounts, loyalty programs and post-purchase surveys.

4:03 Example: Riyadh furniture retailer (illustrative)

Let's look at an example. A furniture retailer in Riyadh sold small accessories online, but the big sofas and dining sets were paid for in the showroom after an online visit. The ad platforms only saw the small online orders, so they optimized for cheap accessories. The fix was capturing email or phone at the showroom with a clear privacy notice, uploading store sales daily as offline conversions, and adding a showroom appointment event with an estimated value. The campaigns shifted toward the journeys that ended in big purchases.

4:42 Hashing done right

How do you send customer data without sending customer data? Hashing. You normalize the value first, trimming spaces and lower-casing an email, turning a phone number into international digits, then run it through SHA two fifty-six. The platform does the same on its side and matches the hashes. Each platform has its own small normalization rules, so read the spec. And never log raw identifiers or send them unhashed. The Python snippet in the lesson shows the pattern.

5:16 Signal health = KPI

Finally, treat signal health like a business metric. Watch your match quality scores, the share of conversions with click IDs or hashed identifiers, how many duplicates are being removed, and the gap between orders in your backend and conversions the platform reports. Set an alert if conversion volume suddenly drops, because a broken tag running for three weeks can undo months of learning.

5:43 Mistakes + try this now

Common mistakes with signals. Sending every tiny event, like page views and button clicks, as conversions, which dilutes what the platform learns. Mixing currencies without the currency parameter. Uploading customer lists without honoring opt-outs. And letting a broken tag run for weeks because nobody watches conversion volume. Try this now: pull last month's orders from your backend and compare them with each platform's reported conversions. Write the three numbers side by side. Then list every event you currently send as a conversion, and mark which ones are real business outcomes. Everything else should be secondary.

6:24 Watch me do it: signal inventory (illustrative)

Watch me do it. Let's build a signal inventory for an illustrative Islamabad online pharmacy. I open a spreadsheet with six columns: event, value source, match keys, latency, consent basis, and a score. Row one, purchase: value from the order total without VAT, match keys hashed email, phone and click IDs, latency seconds, consent for ads required, score four out of five. Row two, prescription upload: no value, and here I pause, because health data is sensitive. We decide this event should not go to ad platforms at all. Score not applicable, and I write the reason. Row three, repeat purchase within ninety days: sent as an offline event nightly with a value, latency one day, score three because only some orders have phone numbers. Row four, add to cart: currently a primary conversion. I move it to secondary. Then I compare last month: backend two thousand orders, Meta one thousand four hundred, Google one thousand two hundred. The inventory tells me exactly where to work next: match keys for repeat purchases, and consent-aware server-side purchase events.

7:41 Recap and next step

To recap: signals are the new targeting, and your edge is first-party knowledge. Test them for coverage, accuracy, match quality and latency. Hash properly and respect consent. Your next step is to build a signal inventory: list every conversion event you send, where its value comes from, which match keys it carries, and how late it arrives. Score each one from one to five, and fix the lowest first.

Why signals are the new targeting

When platforms own the "who" and "where", your competitive advantage is what you know that the platform does not: which customers are profitable, which leads close, which products have margin, who already bought. That knowledge enters the platform as signals. Two advertisers with identical budgets and creative will get very different results if one sends rich, accurate signals and the other sends a noisy pixel.

The signal inventory

Signal type	Examples	How it reaches the platform
Conversion events	Purchase, qualified lead, subscription start	Pixel/tag + server-side API (Meta CAPI, Google enhanced conversions, TikTok Events API, LinkedIn CAPI)
Conversion value	Order revenue, predicted LTV, margin-adjusted value	Event parameters, conversion value rules, offline imports
Customer lists	Past buyers, high-LTV customers, churned customers	Customer Match (Google), custom audiences (Meta), matched audiences (LinkedIn) — hashed, consented
Offline outcomes	Closed deals, store sales, phone orders	Offline conversion import, CRM integrations
Product data	Titles, prices, margins, availability	Merchant Center feed, Meta catalog
Context hints	Search themes, audience signals, keywords	Campaign settings

Signal quality: the four tests

Coverage — what share of real conversions does the platform see? Browsers blocking scripts, consent refusals and cross-device journeys reduce coverage. Server-side APIs and enhanced conversions recover some of it, within consent.
Accuracy — are events deduplicated, correctly valued, in the right currency, and fired once per real conversion?
Match quality — can the platform match events to people? Meta shows an Event Match Quality score; Google shows enhanced conversions diagnostics. Hashed email and phone, plus click IDs (gclid, fbclid via the _fbc cookie, ttclid), improve matching.
Latency — how quickly does the outcome arrive? A B2B deal that closes in 90 days is a weak real-time signal; send earlier proxy stages (qualified lead, meeting held) with values that reflect their likelihood to close.

First-party data is only usable if it was collected lawfully and the user's choices are respected. In the EEA and UK, Google requires consent signals (Consent Mode v2 parameters ad_user_data and ad_personalization) for using data for ads personalization and measurement features. Customer lists must be collected with appropriate notice and, where required, consent; they are hashed (SHA-256) before upload. The Privacy-First Measurement course covers implementation; here the point is strategic: design your data collection so that the most valuable signals are also the ones users have agreed to share, for example through accounts, loyalty programs and post-purchase surveys.

Worked example: a Riyadh furniture retailer

A furniture retailer in Riyadh sells online and in two showrooms. Most high-value orders are paid in store after an online visit. Their ads looked unprofitable because the platforms only saw small online orders.

The fix:

Capture email or phone at showroom checkout with a clear privacy notice compliant with Saudi PDPL requirements.
Upload store sales daily as offline conversions (Google) and offline events via Conversions API (Meta), matched on hashed email/phone and, where available, click IDs captured on web forms.
Add a "showroom appointment booked" event with an estimated value, so the platforms get a faster signal.

Within weeks the campaigns were optimizing toward the journeys that ended in large showroom purchases, not toward cheap accessories bought online.

Hands-on: hashing and normalizing identifiers correctly

Platforms expect normalized, SHA-256-hashed identifiers. Normalization rules differ slightly per platform (check each spec), but the common core is:

import hashlib, re

def norm_email(email: str) -> str:
    return email.strip().lower()

def norm_phone(phone: str, default_country_code: str = "971") -> str:
    digits = re.sub(r"\D", "", phone)
    if digits.startswith("00"):
        digits = digits[2:]
    elif digits.startswith("0"):
        digits = default_country_code + digits[1:]
    return digits  # E.164 digits without '+'; some platforms want the '+', check the spec

def sha256(value: str) -> str:
    return hashlib.sha256(value.encode("utf-8")).hexdigest()

print(sha256(norm_email("  Aisha.Khan@Example.com ")))
print(sha256(norm_phone("050 123 4567")))

Never log raw identifiers, never send them unhashed, and only upload people whose data you are allowed to use for advertising.

Pitfalls

Uploading customer lists without a lawful basis or without honoring opt-outs.
Sending every micro-event as a conversion, diluting the signal.
Forgetting currency: mixing PKR, AED and USD values in one conversion action without the currency parameter.
Letting a broken tag run for weeks; set alerts on conversion volume drops.

How to measure success

Track signal health like a KPI: match quality scores, share of conversions with click IDs or hashed identifiers, deduplication rate, and the gap between backend orders and platform-reported conversions. Improving these usually improves CPA before any campaign change.

Your advantage is the first-party knowledge the platform lacks: profit, qualified outcomes, customer history.
Test signals for coverage, accuracy, match quality and latency.
Normalize and SHA-256 hash identifiers; follow each platform's exact spec.
Consent is part of signal design — collect valuable data through value exchanges users agree to.
Monitor signal health like a KPI.

Build a signal inventory for your business: list every conversion event, its value source, match keys and latency. Score each on coverage, accuracy, match quality and latency from 1 to 5.

Enrol for free to save your progress

Reading is always free. Enrol to keep your place, take the final assessment and earn a verifiable certificate.

Signals and first-party data: feeding the machine