A pilot on 200 accounts goes well. Coverage is high, several accounts move, sales books meetings from names they recognise. The obvious next step is to run the same thing on 2,000 accounts.
It usually does not work, and the reason is not that ABM does not scale. It is that the pilot's results depended on properties that quietly disappear at ten times the size, and nobody writes down what those properties were.
What actually made the pilot work
Four things, typically, and only one of them is about the software.
- Frequency was high enough to matter. A full budget spread over 200 accounts produced ten or more impressions per person per month. That is enough for recognition to form.
- The list was hand-built. Someone chose those 200 accounts and could justify each one. Fit was high because a human filtered it.
- Someone was watching. At 200 accounts one person can notice that a segment is under-delivered, that a creative is not landing, that an account went quiet.
- Sales knew the names. Reps recognised the accounts, so a signal produced a follow-up. This is the property most often lost at scale and least often mentioned.
The three things that break
Frequency, if the budget scales linearly
Ten times the accounts at ten times the budget holds frequency constant, which sounds fine and usually is not available. What actually happens is a budget increase of two or three times, which cuts frequency per person to a third of the pilot — often below the level at which the pilot worked at all. The programme then reports wider reach and worse outcomes, and gets blamed on the list.
List quality, because 2,000 accounts cannot be hand-picked
The second thousand accounts are, almost by definition, a worse fit than the first two hundred. If they are treated identically, average performance falls, and the average is what gets reported.
Attention, because nobody can watch 2,000 accounts
The pilot's careful management was a person, not a process. At scale the person cannot do it, and unless the watching has been converted into thresholds, alerts and a defined follow-up owner, the programme runs unsupervised.
Tiering as the scaling mechanism
The answer is not to scale the programme. It is to run several programmes with different economics, and to be explicit that they are different.
| Tier | Accounts | Treatment | Success metric |
|---|---|---|---|
| 1 — strategic | ~100 | Full sequence, function-split creative, high frequency, named sales owner | Meetings, opportunities |
| 2 — growth | ~400 | Coverage and depth stages, moderate frequency, threshold-triggered outreach | Accounts reaching 2+ functions |
| 3 — volume | ~1,500 | Coverage only, low frequency, inbound follow-up | Cost per account reached |
Separate campaigns, separate budget floors, separate reporting. Tier 3 will look terrible on Tier 1 metrics and that is correct — it is not trying to do Tier 1's job. The mistake is a single blended report in which Tier 3's volume drags the average down and someone concludes the whole programme stopped working.
A sequence that works
- Establish the pilot's real economics. Frequency per person, cost per account reached, cost per account moved. These are your reference numbers.
- Scale one tier at a time. Take Tier 1 from 100 to 200 before adding a Tier 3 at all. Confirm the economics hold.
- Add tiers, not accounts. New accounts enter as a new tier with its own budget and its own metric, so they cannot dilute the tier that is working.
- Automate the watching before you need it. Score thresholds, alerts and a named follow-up owner have to exist before the list is too big to watch, not after someone notices nothing is being followed up.
- Re-check coverage monthly per tier. Under-delivery to a tier is the standard failure mode of a scaled programme, and it is invisible in aggregate reporting.
Teams that do this end up with three programmes that each work on their own terms. Teams that scale the list and hope end up with one programme that used to work.