Confidence without receipts
Recommendation engines assert relevance with total confidence. Almost none measure whether their picks outperform chance on your data, because the measurement could embarrass them.
Wave ยท Content Lift Shadow Testing
Wave proves its own recommendations. Against random, every day.
Every content recommendation Wave makes runs alongside a shadow pick: a random selection from the same candidate pool. Wave measures which drove more downstream engagement, and its ranking must beat chance, visibly, before Wave upgrades its own algorithm.
Wave~
Accounts
People
Buying Groups
Schedule
Content
Insights
Settings
Insights › Recommendation lift
Recommended vs shadow arm
Recommended23 engagements
Shadow pick15 engagements
Recommended arm leading
Readout unlocks at minimum population
Test arms
Ranked pick, shown to the seat
Random pick, logged and never shown
Quick answer
Content Lift Shadow Testing is Wave's built-in measurement layer for its own content recommendations. Each time Wave fills a recommendation slot for a buying-group seat, it also records a random pick from the same eligible candidate pool. Both are tracked against downstream engagement, and the gap between them is the lift Wave's ranking generates above chance. Operators see the lift readout per tenant, and Wave upgrades its ranking algorithm only when measured lift clears a pre-registered bar on a sufficient population.
Key capabilities
- Every recommendation paired with a random shadow pick
- Both arms drawn from the same candidate pool, a fair test
- Downstream engagement measured for both, continuously
- Per-tenant lift readout operators can inspect
- Algorithm upgrades gated on measured lift, not claims
- Per-recommendation breakdown of which signals matched
Last updated: July 2026
The problem
"AI-powered" is a claim. Lift over random is a measurement.
The recommendation category sells sophistication and shows nothing. Wave treats its own picks as a hypothesis and runs the test that could prove them wrong.
Case studies are someone else's account
A lift number from another company's data says nothing about your library, your buyers, or your motion. Borrowed proof is not proof.
Unfalsifiable systems never improve
If a recommender cannot fail a test, it never has to get better. Without a control arm, there is no way to know whether the algorithm adds anything at all.
How Wave does it
A control arm on every recommendation, automatically.
Shadow testing runs inside the recommendation flow itself: no experiment setup, no analyst project, no holdout to negotiate.
- 01
Two picks per slot
When Wave fills a recommendation slot for a buying-group seat, it records two assets: its ranked pick and a random pick from the same eligible pool. Only the ranked pick is ever surfaced; the shadow exists purely for measurement.
- 02
A fair test by construction
Both arms draw from the same theme-matched candidate pool for that seat. Whatever gap emerges is attributable to the ranking itself, not to differences in the content being compared.
- 03
Engagement settles it
Wave measures downstream engagement against both assets over a fixed attribution window. Engagement with the shadow pick reflects chance; engagement with the ranked pick reflects the algorithm.
- 04
The readout, and the gate
Operators see measured lift per tenant in a color-graded dashboard, with population size alongside so small samples never masquerade as proof. Wave upgrades its ranking engine only when lift clears a pre-registered bar on a sufficient population.
Where it fits
The proof layer under Wave's content recommendations.
Shadow testing instruments the recommendations Wave already makes for buying-group seats, and it complements the platform's person-level holdout experiments.
Every recommendation
Wave
Lift readout
Every recommendation Wave makes already carries a visible breakdown of which signals matched and why the winning asset ranked first; shadow testing adds the outcome measurement on top. It is one half of Wave's accountability layer: AI Lift Experiments run control and treatment holdouts to measure what Wave's live CRM writeback changes, while shadow testing measures the content ranker specifically, continuously, without holding anyone out. The recommendations it measures serve the same seats Next Up Content serves, so the lift you see is lift on the motion you actually run.
Works alongside Next Up Content, AI Lift Experiments, The full Wave platform.
Why Wave is different
Every vendor sells the algorithm. Wave sells the measurement.
The honest question for any recommendation engine is: what is your lift over random? Wave built the mechanism to answer it, per tenant, on live data.
Most tools
Wave
Vendors describe algorithm sophistication in the abstract.
Wave measures its picks against random from the same pool and shows the lift, per tenant.
Recommendations are take-it-or-trust-it, with no visible basis.
Every recommendation carries a breakdown of which signals matched and why it won the slot.
Algorithm changes ship on the vendor's schedule, justified by release notes.
Wave's ranking upgrades are gated on measured lift clearing a pre-registered bar on your data. The system can fail the test; that is the point.
Measuring AI value means building your own experiment.
The shadow arm runs inside the recommendation flow, always on when the ranker is active. No setup, no analyst project.
FAQ
Questions buyers ask about Content Lift Shadow Testing.
What is Content Lift Shadow Testing in Wave?
A built-in measurement layer for Wave's content recommendations. Every recommendation is paired with a random shadow pick from the same candidate pool, both are tracked against downstream engagement, and the gap between them is the lift Wave's ranking generates above chance. The readout is visible to operators, per tenant.
What exactly is a shadow pick?
A randomly selected asset from the same eligible pool the ranked recommendation came from, chosen reproducibly and recorded alongside the recommendation. It is never surfaced to anyone; it exists so engagement with it can serve as the chance baseline.
Why measure against random instead of a fixed baseline?
Random from the same pool is the honest control: it holds content quality, topic match, and audience constant, so any engagement gap is attributable to the ranking itself. A fixed baseline would let the algorithm take credit for the quality of the pool.
Does shadow testing change what buyers or reps see?
No. The recommended asset is always the ranked pick. The shadow is instrumentation: recorded, measured, and never served to anyone.
What does the lift readout show?
Measured lift per tenant, color-graded, with the size of the measured population alongside so small samples are visible as exactly that. It also surfaces when a tenant's accumulated evidence is sufficient for Wave to upgrade the ranking algorithm.
What happens if lift does not clear the bar?
The current ranker stays in place and keeps accumulating evidence. Upgrades happen on measurement, not on a schedule, and the bar is pre-registered rather than moved after the fact. The system can fail the test; that is the point.
How is this different from AI Lift Experiments?
They are complementary. AI Lift Experiments run control and treatment holdouts at the person, account, or committee level to measure the impact of Wave's live CRM writeback. Shadow testing measures the content ranker specifically and runs continuously without withholding anything from anyone.
How do I see the lift readout on my own data?
Book a 20-minute walkthrough. We will show the lift readout, the per-recommendation signal breakdown, and how the measurement accumulates on a live tenant.
See it on your data
See the measurement other tools cannot show you.
Book a 20-minute walkthrough. We will show the shadow-testing readout live and how Wave accumulates proof that its recommendations beat chance.
Request a demo