Dashboards are not proof
Showing you the platform's own predictions is not the same as showing the lift those predictions caused.
Wave · Revenue outcomes
Prove the lift on your own data.
Turn engagement into pipeline you can prove. Wave keeps a control group it leaves alone and a group that gets its predictions, then compares the two in your own CRM every day. It reports the difference with a confidence range and a minimum sample, and it tells you when it is too early to call. The answer to whether it worked is built into the product.
Quick answer
AI Lift Experiments is a Wave capability that runs a control-versus-treatment holdout on your own contacts. In suppressed-holdout mode, Wave withholds its AI writeback from a control set while continuing to enrich treatment, then reports the per-arm lift daily with a guardrail. An observational mode tracks a cohort without suppressing anything. You measure Wave's contribution in your own CRM, with no separate analytics tool.
Key capabilities
- Control-versus-treatment holdout
- Suppressed-holdout and observational modes
- Daily lift readout with a confidence range
- Says 'not enough data yet' instead of guessing
- Withheld writes logged and replayable
- Covers every person-level product
- Run it from day one of a pilot
- Measured in your own CRM, no extra tool
Last updated September 2026
The problem
The hardest question is renewal, not the first sale.
Every CMO who buys an intelligence tool eventually gets asked by finance: what did we actually get? Most platforms cannot answer with a clean controlled comparison.
Holdouts are a project
Building a clean control group by hand, then measuring it, usually means spreadsheets and an analyst's time.
Trust erodes without evidence
Without a controlled result, the renewal conversation quietly becomes a re-evaluation.
How Wave does it
A controlled holdout, baked into writeback.
The holdout is not a report layered on top. It is part of how Wave writes back, so the control group stays genuinely clean.
- 01
Split into arms
Wave assigns a configurable set of contacts to control and the rest to treatment, per person or per account.
- 02
Withhold from control
Control contacts receive no Wave enrichment, so their records look exactly as they did before Wave was installed.
- 03
Measure daily
Wave compares the two groups every day and reports the difference with a confidence range and a minimum sample size. If the sample is too small to call, it says so. No number is printed that the data cannot support.
- 04
Roll it out
When the experiment concludes, Wave can write the held-back predictions so the control set catches up to the enriched state.
Where it fits
Lift you can measure across every product.
This is the proof end of the Wave journey: every play measured against a holdout so the lift you report is lift you can defend. The holdout applies across every person-level Wave product, so the control group is truly un-enriched, not just missing one signal.
Treatment
Wave
Measured lift
Predictions are still computed for control contacts so the counterfactual stays measurable; suppression happens only at the boundary where Wave would write to your CRM. Every withheld write is logged, so you can see exactly what the control group would have received and replay it later. Suppression rides the experiment arm only, never your consent records or contact status.
Works alongside Channel Affinity, HubSpot Enrichment, Motion Traction.
Why Wave is different
The platform brave enough to exclude itself.
Every vendor shows you a dashboard of its own predictions. Almost none will hold a control group against itself, and none we have found publish the result with a significance test. Wave does both, in your CRM, from day one of a pilot. Ask any other vendor for the control group and the confidence range. The pause is your answer.
Most tools
Wave
Most platforms show you their own dashboards.
Wave shows you a controlled comparison from your own data.
ROI is asserted from period-over-period metrics.
Wave reports the difference with a confidence range, and withholds the number when the sample is too small.
A holdout is a manual analytics project.
The holdout is built into the writeback pipeline, with no setup.
You trust the tool for six months, then re-evaluate.
Run a holdout from day one and decide on evidence.
FAQ
Questions buyers ask about AI Lift Experiments.
What is AI Lift Experiments?
It is a Wave capability that runs a control-versus-treatment holdout on your own contacts. Wave withholds its AI writeback from a control set, keeps enriching treatment, and reports the per-arm lift daily, so you can measure its contribution in your own CRM.
What is suppressed-holdout mode?
In suppressed-holdout mode, the control group receives no Wave enrichment while treatment does. The difference in downstream results is the measured contribution of the AI. An observational mode is also available that tracks a cohort without suppressing anything.
Does the control group get touched at all?
No. Control contacts receive no Wave-written fields during the holdout, so their records look exactly as they did before Wave was installed. That keeps the comparison clean.
Can I see what the control group would have received?
Yes. Every withheld prediction is logged with the value Wave would have written, so you can inspect it and replay the writes later if you decide to roll out fully.
Which products does the holdout cover?
It applies across every person-level Wave product, so a control contact is genuinely un-enriched rather than missing a single signal.
Will a holdout slow down my results?
Only for the control set you choose, and only for the period you choose. Treatment contacts are enriched normally throughout.
Can I use this during a pilot?
Yes. You can enable a holdout from day one of a pilot and have a controlled result from your own contacts before you make a purchase decision.
How do I set one up?
Book a 20-minute walkthrough. We will design a holdout against your stack shape and show how the per-arm lift would read on your data.
Is the lift statistically significant?
Wave reports a confidence range with every lift reading and requires a minimum sample before it will call a result. If the groups are too small or too new, the readout says 'not enough data yet.' The full method is in the lift methodology whitepaper, delivered with the security pack.
What if the AI made no difference?
Then the readout will say so, and you will know before renewal instead of after. Wave is built so that answer is possible.
See it on your data
Prove Wave's lift on your own data.
Book a 20-minute walkthrough. We will design a control-versus-treatment holdout and show how the lift would read in your CRM.
Request a demo Download the one-pager (PDF)