Experimentation powered by AI
Constant questions. New growth curves.
Left alone, Google Ads and Meta optimize the same signals until growth flattens.
The loop
Now testing
Revenue over time
Illustrative sequence. Curve shapes, lift and timing do not describe any real account.
Experimentation = asking better questions.
We solve business problems by injecting the variables ad platforms can't see on their own directly into Google and Meta, then measuring what actually changed.
What if we guide the algorithm toward high-value customers instead of sheer volume?
Customer value
What if we isolate the geographies the algorithm quietly ignores?
Geographic efficiency
What if we revive the products the system has abandoned?
Product potential
What if we measure lift, not just performance?
Incrementality
What if we seed PMax with our highest-value customers instead of every past buyer?
If we give Google Performance Max an LXRInsights high-value customer (HVC) seed plus customers predicted to become HVCs, promote the skincare bundle HVCs buy most, and focus budget on the metros with the highest revenue per capita at the hours they shop, then PMax will acquire new customers with a higher first-order AOV and higher predicted 12-month value than our current PMax campaign, at the same or lower CAC.
1LXRI audience signals
Both lists sync from LXRInsights to Google Ads and are added as PMax audience signals.
2Google PMax setup
- Campaign goal
- Acquisition · customer acquisition set to "Only bid for new customers"
- Bidding
- Maximize conversion value with a target ROAS
- Asset group
- Glow Routine bundle, with HVC seed + predictive as signals
- Landing page
- Bundle page, with final URL expansion off
3Ad copy
Vitamin C serum, barrier moisturizer and SPF 30 in one set. Save 15% on your first routine.
- Headline: Your Skin's New Morning Routine
- Headline: Glow Routine Set · Save 15%
- Description: Dermatologist-tested and loved by repeat customers.
4Product bundle suggested by LXRI
Why this bundle: it's the combination HVCs most often buy in their first 90 days, and customers who start with it reorder sooner than single-product buyers.
5Geo targeting by revenue per capita
| Metro | Revenue per 100K people | Play |
|---|---|---|
| Austin, TX | $41.2K | Scale +25% |
| Denver, CO | $35.6K | Scale +25% |
| Charlotte, NC | $29.0K | Test +10% |
| Phoenix, AZ | $16.7K | Hold |
6Launch timing and time of day
Launch: Tuesday at 6 AM local time, so the campaign has a full week of learning before weekend demand.
Peak windows come from when the HVC seed audience places its orders.
Current PMax campaign with all-purchaser signals, same budget and geos
New-customer revenue · First-order AOV · Predicted 12-month CLTV · CAC
6 weeks, or until results reach 95% statistical confidence
Sample experiment for a fictional beauty brand. Audiences, geo figures and timing are illustrative.
Our experimentation evaluation method.
Our experts evaluate which campaign shows better outcomes across the questions that matter to your margin.
Experiment (X) vs control (C)
Statistical confidence is the source of truth.
Not lift, not instinct, not a good first week. AI and human expertise design the test, and only significance decides whether it deserves to scale.
We Paused a “Winning” Campaign — and ROAS Went Up
Most brands would have scaled this campaign. We paused it. Because it was targeting the wrong customers and the data proved it.
More Revenue.
Same Spend.
We’re breaking the black box of AI and empowering you with transparent insights to maximize your budget.
Let's Chat
Our team will contact you shortly