Short answer
Randomly split your target list into a test group that gets the program and a control group that gets no paid ABM, then compare opportunity rates after a full sales cycle. The difference is the lift you can claim. Plan the size first: with 800 test and 200 control accounts, a lift from 4% to 6% isn't statistically clear. Use bigger groups, longer periods or an earlier metric such as engaged accounts.
Design in five steps
- Clean the list to one row per company with the target account list cleaner.
- Split it randomly within each tier, so both groups have the same mix of size and fit.
- Keep the control dark in every paid channel: exclude it from LinkedIn company lists, DSP account lists and retargeting.
- Keep everything else equal: sales outreach, email and events run as normal for both.
- Compare after a full cycle: opportunity rate, pipeline per account and win rate.
A worked example, and the size trap
| Group (fictional) | Accounts | Opportunities | Opportunity rate |
|---|---|---|---|
| Test | 800 | 48 | 6.0% |
| Control | 200 | 8 | 4.0% |
| Lift | 16 incremental | +2.0 points (+50%) |
A 50% lift sounds decisive. It isn't: a standard two-proportion test gives z ≈ 1.1, well short of the 1.96 needed for 95% confidence. With eight opportunities in the control, chance alone could explain the gap.
How to get a clear answer
- A bigger control: a 50/50 split of the same 1,000 accounts gives far tighter estimates than 80/20.
- A longer period: more opportunities on both sides.
- An earlier metric: engaged accounts or meetings happen more often than opportunities, so differences show sooner.
- Pool several quarters of the same design before drawing conclusions.
What to report
| Line | Example |
|---|---|
| Incremental opportunities | 16 (range wide; not yet significant) |
| Incremental pipeline | 16 × average deal value |
| Cost per incremental opportunity | Program cost on the test group ÷ 16 |
| Next step | Extend one quarter, move to 50/50 split for new accounts |
Use the result to calibrate the ABM ROI model: replace "influenced" deals with incremental ones.
Questions
What is a holdout test in ABM?
A control group of target accounts that is deliberately left out of paid ABM, so you can compare its results with accounts that received the program.
How large should the control group be?
Often 20–50% of the list. A bigger control gives a clearer answer but delays revenue from those accounts, so agree the trade-off with sales.
Can sales still work control accounts?
Yes. Keep sales activity the same for both groups; only the paid program differs.
What if the result isn't significant?
Report the direction and the range, extend the test or move to an earlier metric such as engaged accounts or meetings.
Facts checked against the sources below on 19 October 2026. Ad platforms change quickly: if something here no longer matches what you see, tell us and we will update it.
