Case study · AI Agents & Automation
Audiences built from who actually buys, not a generic interest list
Every audience in the ad account was an interest list copied from a template, broad enough to reach a lot of people and precise enough to convert almost none of them. This is targeting logic rebuilt from the company's own customer data.
- Client
- a B2B growth-stage services company
- Problem
- Ad audiences were broad interest lists that reached a lot of people and converted few of them.
- Approach
- Targeting rebuilt from the company's own customer data: who actually buys, not who might be interested.
The problem
The ad account's audiences were built the way most are: pick a handful of interests that sound adjacent to the business and let the platform's algorithm sort it out. It worked well enough to keep running, badly enough that cost per qualified lead kept drifting the wrong way.
What the account never used was the company's own customer data. There was a real, detailed answer to who actually bought and who churned fast, sitting in the CRM, disconnected from the ad platform that was supposedly trying to find more people like them.
The reality behind the problem
The obvious fix, narrowing the interest list further, usually makes an audience smaller without making it more accurate. An interest is a guess about intent; a closed-won customer is a fact. No amount of interest-list tuning closes that gap.
The harder part was getting customer data into a shape an ad platform could use safely and effectively: cleaned, modeled around what actually predicts a good customer, and refreshed as new data came in, not a one-time export that goes stale in a month.
What was assumed
Better targeting means narrower interest lists.
What Stallwart asked
What if the audience was built from who already buys, refreshed automatically as that changes?
The desired outcome
The target was audiences grounded in real customer behavior, kept current without a manual export every quarter.
- Customer data modeled into attributes a targeting system can use, not raw CRM fields.
- Lookalike audiences built from closed-won customers, not generic interests.
- Audiences refreshed automatically as new customer data comes in.
- A visible record of which audience is driving which result, not one merged number.
The system we designed
A pipeline that pulls customer data from the CRM, models it into the attributes that actually separate a good customer from a poor-fit one, and builds lookalike and custom audiences from that model instead of a static interest list.
Audiences refresh on a schedule as new customer data arrives, so targeting keeps pace with who the company is actually converting, and each audience's contribution to results stays visible rather than blended into one account-level number.
How it works
- 01
Model
Turn CRM data into attributes that predict a good customer, not raw fields.
- 02
Build
Construct lookalike and custom audiences from that model.
- 03
Refresh
Update audiences automatically as new customer data arrives.
- 04
Attribute
Keep each audience's contribution visible, not blended into one number.
What changed
The account now targets people who resemble actual customers, not a generic interest profile, and that targeting updates itself as the customer base changes instead of going stale.
The team can see which audience is actually contributing, which changed how budget gets argued about internally: from opinion to a specific number per audience.
- From static interest lists to audiences modeled on real customers
- From a one-time data export to a refreshed pipeline
- From one blended account number to per-audience visibility
In practice
Growth Lead, a B2B growth-stage services company
The team noted that budget conversations used to be arguments about instinct. With per-audience results visible, the conversation became a quick look at a specific number instead of a debate about which interest list felt right.
Ownership & handover
What you receive, and keep.
- The customer-data pipeline and audience-modeling logic, documented and handed over.
- The ad account and its audience structure stay yours throughout.
- No ongoing dependency on Stallwart to keep audiences refreshing.
What we learned
The best targeting signal most companies have is their own closed-won data, and it is almost never connected to the ad account trying to find more of those customers.
Narrower is not the same as more accurate. An audience built from real customer behavior beats a smaller version of a guess.
This work connects to
Frequently asked
What customer data does this need access to?
Whatever your CRM already holds on closed-won and churned customers. We scope exactly what is needed, and nothing more, during discovery.
How often do audiences refresh?
On a schedule matched to how fast your customer base changes, typically weekly to start, adjusted once we see the real pace of new data.
See targeting built from your own customer data
Tell us what your current audiences look like, and where the leads that convert actually come from.