Why this matters
A small shop's audiences usually go wrong in familiar ways: a lookalike seeded from the biggest audience, so it models browsers; one remarketing window and message for everyone; and no exclusions, so prospecting money reaches last week's buyers. None of these produces an error, only a cost per order that never comes down.
This exercise applies three readings. The custom audiences guide covers the sources, windows and the three jobs custom audiences do. The lookalike audiences guide explains how to choose and value-weight a seed. The exclusions and audience overlap guide gives the rule set that gives every person one rung.
The shop and its data sources
Tallgrass Plants is a fictional online shop selling houseplants, pots and plant food. It also supplies a few local cafés and offices on account. The pixel, consent banner and standard commerce events with values all work.
Tallgrass Plants: what data exists
| Source | What it holds | Approximate size |
|---|---|---|
| Customer list export | Retail customers with email, phone, postcode and lifetime spend | 3,400 people |
| Same export, filtered | Retail customers with two or more orders | 1,100 people |
| Trade list | Café and office buyers, mostly work email addresses | 60 accounts |
| Pixel: purchase, last 180 days | Online buyers who accepted tracking | About 1,500 people |
| Pixel: add to cart, last 14 days, no purchase | Basket abandoners | About 250 people |
| Pixel: view content, last 30 days, no add to cart | Product viewers | About 3,000 people |
| Instagram engagers, last 90 days | People who liked, commented or saved posts | About 9,000 people |
| Video viewers | People who watched most of the repotting tutorial videos | About 2,000 people |
- Monthly budget: £1,200. The owners want roughly three quarters on finding new customers.
- Typical buyers reorder plant food and pots after a couple of months.
Building and separating the audiences
Start with the jobs: prospecting needs strangers and, optionally, a seed to point delivery at a good starting group; RemarketingAdvertising to previous visitors or engagers. needs people who showed intent and have not bought; exclusions need one "converters" audience every acquisition ad set subtracts. Then check sizes: a rung too small to deliver is better merged with its neighbour, because tiny audiences drive FrequencyAverage times a person saw an ad. up fast.
For the seed, remember that a Lookalike AudienceAudience similar to a seed list. finds people like the seed, so a big seed of the wrong people produces a confident audience of the wrong people. Under Advantage+ audience a lookalike added to an ad set acts as a suggestion, but exclusions of custom audiences still hold.
Task
- List the custom audiences you would build from the sources, with the window for each, and say which job each one does: remarketing, exclusion or seed.
- Choose the lookalike seed and say why it beats the other candidates. Name one source you would keep out of the seed and why.
- Design the prospecting ad set: who it includes or suggests, and who it excludes.
- Design the remarketing ad set or sets: decide whether basket abandoners get their own rung, given their size, and write the message each rung gets.
- Write the exclusions each ad set needs, including the windows, so that nobody sits in two ad sets at once.
- Say where repeat buyers of plant food and pots should go, since excluding all buyers everywhere would lose them.
Worked answer
Custom audiences.
- Converters (exclusion): pixel purchasers in the longest window in the table combined with the full retail customer list, so buyers who declined tracking are still caught.
- Repeat customers with value (seed): the filtered list of customers with more than one order, uploaded with lifetime spend as the value column.
- Basket abandoners and product viewers (remarketing): the pixel audiences in the table.
- Engagers and video viewers (exclusion from prospecting, possible later rung): Instagram engagers and repotting video viewers.
The seed. Use the repeat customers with value. They are real buyers, weighted towards the ones who make the economics work, and the list is a reasonable size for a seed. The pixel purchasers are a decent second choice but miss buyers who refused tracking. Instagram engagers are the largest and weakest seed: liking plant photos is not buying plants. Keep the trade list out entirely. Work emails match poorly, and cafés buying on account are not the customers the retail shop wants more of.
Prospecting ad set. Broad targeting with Advantage+ audience, the value-based lookalike added as a suggestion. Exclude converters, all pixel visitors in the remarketing windows, Instagram engagers and video viewers, so that prospecting money reaches strangers only.
Remarketing. One ad set, not two. The basket abandoner audience is too small to deliver on its own at this budget, so merge it with product viewers into a single "considered" rung and use several ads: one with care guides and reviews for viewers, one addressing delivery and how plants are packed for people who reached the basket. Exclude converters with a short window so a buyer leaves the rung the day they order.
Repeat buyers. Give them a small retention ad set promoting plant food and pots to the customer list and recent purchasers, excluding only very recent buyers. Keep them out of prospecting.
Tallgrass Plants: the audience plan
| Ad set | Includes | Excludes | Share of budget |
|---|---|---|---|
| Prospecting | Broad, value-based lookalike as a suggestion | Converters; visitors in last 30 days; engagers and video viewers in last 90 days | About £900 |
| Remarketing (considered) | Product viewers in last 30 days and basket abandoners in last 14 days | Converters in last 7 days | About £200 |
| Retention | Retail customer list and purchasers in last 180 days | Purchasers in last 14 days | About £100 |
The windows match across ad sets: the visitor window excluded from prospecting is the same as the window included in remarketing, so nobody falls between the two or sits in both.
Common mistakes
- Seeding from engagers because the audience is big. The model learns to find people who like posts.
- No converter exclusion on prospecting. The largest leak in most accounts, and the easiest to fix.
- Mismatched windows. People drop out of one ad set before the next picks them up.
- Excluding all buyers from everything. Repeat purchases of consumables are lost.
Next, test yourself with the audiences and targeting checkpoint, or return to the beginner Meta Ads level to see the module in order.
