Signals not guesses the right kind of targeting

The coffee was already cold when we realized we were aiming at shadows. A small ecommerce team, two laptops, a whiteboard covered in buyer personas that felt like we cribbed them from a brochure. We had spent real money on broad interests and a one size fits all segment called Likely Shoppers. Turns out, they were not. Sales flatlined. Then someone pulled up a report that changed the conversation. People who searched site for return policy and then viewed shipping details converted at five times the average. No guess. A signal. That night we killed the fuzzy targeting and doubled down on people who actually told us what they wanted with their clicks. The next week, the graph moved.

Analysis: what counts as a signal and why it beats a hunch

Most targeting is a vibe. You imagine an ideal customer, pick some interests, add an age range, and cross fingers. That can work if you sell sneakers with massive awareness. For the rest of us, signals get results. A signal is something a person does that ties to intent. It is observable, it is fresh, and you can act on it.

Here are the signals that matter right now across the big platforms that actually spend: Google, Facebook, Amazon getting louder, and the programmatic pipes:

Search intent: queries on your site and on Google. Someone who types best standing desk for small spaces is raising a hand. If you own that phrase and the long tail nearby, you are halfway there. On your own site, build audiences based on terms and follow with matching offers.
Behavior on site: product views with depth, cart starts, checkout field focus, return visitors within 7 days, people who looked at the pricing page for 30 seconds plus. These are not vanity metrics. They predict money.
Email and CRM events: opened three onboarding emails but did not activate, clicked to pricing from a feature update, requested a demo, talked to support about limits. Feed these to Facebook Custom Audiences and Google Customer Match. You can even seed lookalikes from closed won customers with high lifetime value.
App events: trial started, feature used twice, subscription paused, push notifications ignored five times in a row. These are gold for re engagement without wasting reach on happy users.
Context signals: location during store hours, weather for delivery promos, device on WiFi vs on the move, day of week patterns. These are softer than intent but can boost timing.

Signals hit different because they are tied to a decision. Guesses are tied to your hopes. Someone who added to cart and bounced is not the same as someone we think likes our category. That person needs a nudge on shipping time or a reminder with social proof. Your budget should mirror that difference.

Freshness matters. A cart event from 30 days ago is stale. A logged search from this morning is hot. Build windows that map to your sales cycle. For fast purchases use 1 day and 3 day buckets. For B2B, think 7, 14, and 30 day stages with plays that fit where they are in the dance.

Signals also help your creative. You do not need 50 ad versions. You need a few smart messages that line up with what the person just did. For example, if they looked at black backpacks over 100 dollars, lead with that top seller and highlight delivery promise. If they read a comparison post, show a testimonial that answers the same doubt. Message to move the next step. Not a generic brand ad.

The ad platforms are leaning into this too. Facebook keeps improving Custom Audiences and lookalikes. Google is stitching in store visits to AdWords. Amazon keeps adding knobs to its ad console and sits on more purchase data than anyone. Even with all of that, your first party signals are the backbone. You own the pixel, the logs, the CRM. That is your unfair edge.

Risks and tradeoffs you should consider

Signals are not magic. They break or mislead if you are not picky. A few things to keep an eye on:

Noise vs intent: time on page is not intent if the tab was idle. Scroll depth can lie. Favor actions that show effort like start checkout, search, download, or request a quote.
Attribution fog: if you target people already on the edge of buying, you can cannibalize organic sales and brag about fake lift. Use holdouts and geography splits to get a clean read.
Data freshness: stale audiences waste money. Sync daily at minimum. For high volume, aim for near real time on the hottest signals.
Privacy and policy shifts: Safari is moving to limit tracking across sites. Ad blockers keep growing. Get consent, be clear in your cookie notice, and store less than you think you need.
Platform black boxes: lookalikes and similar audiences are great but you do not control the guts. Always keep a baseline with simple, explainable segments you can sanity check.
Creative fatigue: retargeting can feel creepy fast. Frequency caps and burn rules save your brand and your CPA.
Data quality: event misfires are common. Double fires and missing parameters produce phantom conversions. Test with your own clicks before you scale.

There is also the people side. Not everyone in the room loves the idea of tighter tracking. Respect that concern. Be human in your targeting. Offer value. Make opt out easy. Do not follow someone around the web for weeks after a single bounce.

Decision checklist: are we using signals or guesses

What exact action predicts conversion for us in the last 60 days and by how much
Which of those actions can we track today with a clean event and an audience we can export to our ad buys
How fresh is our data and how often do we refresh audiences
Do we have consent and clear disclosures for the data we are sharing with ad platforms
What is our holdout plan to measure true lift not just last click or view through stories
What is our budget split between prospecting lookalikes and intent heavy retargeting
What are our burn rules after a purchase or after X impressions without a click
Do we map creative to the signal with at least three lanes new prospect warm evaluator near close
What is our fallback if a platform limits tracking or our pixel breaks
Who owns the QA of events, naming, and data accuracy every week

Action items: move from guessing to signals in 30 days

Map your funnel events: write down the top ten actions users take before conversion. Pull the numbers for conversion rate by action to rank them.
Clean your pixel and app events: standardize names, send key parameters like value and product id, and kill duplicate fires. Validate with real clicks while watching your network tab.
Build three core audiences: hot visitors last 3 days with cart or pricing views, warm visitors last 14 days with product depth, and past buyers for cross sell after 30 days. Keep sizes above a few thousand to avoid delivery issues.
Seed one lookalike from your top 10 percent customers by lifetime value not just recent buyers. Back it with exclusions for recent visitors to keep prospecting clean.
Draft matching creative: one set for hot with shipping and social proof, one for warm with benefits and use cases, one for cold with a simple promise and proof.
Set frequency and burn rules: cap hot at 3 to 5 per day for a week then pause. Exclude buyers for at least 30 days unless you sell a refill.
Plan a lift test: hold out 10 to 20 percent of your best audience by geography or by random split. Track sales in both groups. Keep it simple and run for two weeks minimum.
Review weekly: pull a short report that shows spend, reach, CTR, conversion rate, CPA, and lift for each signal based segment. Kill what drags. Fund what compounds.
Document consent: update your cookie banner, privacy page, and platform data sharing settings. Keep screenshots and dates. Make it boring. Legal will thank you.
Iterate one signal at a time: add or tweak only one key signal each week. That way you learn what moved the needle.

You will know you are out of the guessing business when your calendar looks different. Less time arguing about personas, more time refining the actions that proved they matter. It is not as shiny as a brand new audience template, but it stacks wins. Signals tell you where to aim and how hard. Guesses burn your budget and your patience.

Right now, platforms keep rolling out tools that reward this approach. Facebook is making Custom Audiences simpler to sync. Google keeps tying ads to store visits for retail. Safari is tightening tracking so first party data becomes even more important. The wave favors teams that build around clear actions and clean consent.

If your spend is stuck and your CPA keeps wobbling, start here. Pick one signal, wire it up, map a message to it, and run a clean test. Then repeat. That cold coffee night turned our account from flat to steady. No magic. Just signals, not guesses.