Here is a pattern I have now watched play out at a dozen companies. A team launches a named-account programme on LinkedIn. Three months in, the numbers look acceptable: click-through rate above whatever benchmark the agency supplied, cost per click drifting down, a trickle of demo requests. Six months in, sales is unconvinced and the budget is being questioned.
The team responds by working on creative. New hooks, new formats, a document ad, a video. The metrics improve slightly. Sales remains unconvinced.
Nobody has checked whether the ads reached the list.
The metric that hides the problem
Campaign Manager reports an average frequency. Say it tells you 6.2 impressions per member over the last month. That sounds like a healthy account-based programme: everyone on your list is seeing you roughly weekly.
An average of 6.2 is also what you get when 30% of your audience sees you twenty times and 45% sees you not at all. Those two distributions are indistinguishable in the platform's reporting, and they are completely different programmes. One is building recognition across a buying committee. The other is annoying a small group of people while the accounts you actually wrote the strategy for have never heard of you.
Every average in ad reporting conceals a distribution. In demand generation that rarely matters. In ABM it is the entire question.
The reason this goes unnoticed for so long is that every downstream metric behaves normally. Your CTR is computed over people who saw the ad, so it tells you nothing about the people who did not. Your cost per click is computed the same way. Your conversion rate is computed on clicks. The entire measurement chain is conditioned on delivery having happened, so an unreached account is invisible at every step.
Why delivery concentrates
This is not a bug in the platform. It is the platform working exactly as designed, for a goal that is not yours.
An ad auction optimises for your stated objective at the lowest available price. If you ask for impressions, it buys the cheapest impressions inside your targeting. If you ask for clicks, it buys the impressions most likely to produce a click. In both cases it will find a subset of your audience that is cheap or responsive, and it will keep going back to them, because that is what "efficient" means.
Four things make this worse in a named-account setup specifically:
- Tier mixing. Put enterprise and mid-market in one campaign and the auction will drift towards mid-market, where the audience is larger and the competition for attention is cheaper. Your most valuable tier starves inside a campaign that reports fine on average.
- Small audiences with tight filters. A 400-account list narrowed to four job functions and director-plus seniority might be 6,000 people. That is small enough that the auction can saturate a responsive core within days.
- No frequency cap. Without a cap there is nothing stopping delivery concentrating. The cap is not there to protect the viewer from fatigue; it is there to force the budget outward across the list.
- Match rate decay. The list you uploaded eight months ago is not the list the platform is targeting today. Companies get acquired, rename and change domains, and unless someone reads the match report, part of your list quietly stopped being targetable.
How to measure coverage properly
Coverage is a question about your list, so it has to be computed from your list, not from the delivery data. That distinction is the whole trick. Start with every row you intended to reach, then join delivery to it — not the other way round.
Four numbers, reported per tier, are enough to run a programme:
| Metric | Definition | What a bad number looks like |
|---|---|---|
| Match rate | Rows the platform could resolve, over rows uploaded | Below ~75% on a clean list |
| Reach | Accounts served ≥1 impression in the window | Below ~70% of matched accounts |
| Committee breadth | Distinct job functions reached per account | Median of 1 |
| Cost per account reached | Spend ÷ accounts served | Varies 3× or more across tiers |
The last one is the most useful and the least used. If Tier 1 costs three times more per account reached than Tier 3, that is not a failure — enterprise audiences genuinely cost more. It becomes a failure when the two tiers share a budget, because the auction will resolve that price difference by serving Tier 3 instead.
Fixing it
In roughly the order of how much they matter:
- Split tiers into separate campaigns with their own budget floors. Nothing else on this list has anything like the same effect. It removes the auction's ability to arbitrage between your strategic accounts and your cheap ones.
- Cap frequency. Somewhere between six and ten impressions per person per month for most programmes. The exact number matters less than having one.
- Re-read the match report monthly. Not at setup — monthly. Fix the rows that failed. Most failures are trading names where you needed a registered domain, and they take minutes to resolve.
- Cut the list before cutting frequency. If the budget will not cover the list at the frequency you need, reach 150 accounts properly rather than 400 accounts occasionally. The second option produces a report that looks fine and a pipeline that does not.
What gets worse first
You should expect every headline efficiency metric to deteriorate when you fix coverage, and you should say so out loud before you make the change.
CPM goes up, because you stop buying the cheapest impressions. CPC goes up for the same reason. CTR goes down, because the people who were seeing you fifteen times a month were the responsive ones, and you have just spread delivery to people encountering you for the first time.
What should improve is the number of accounts with multiple functions engaged, and eventually the number of accounts your sales team recognises when you name them. Those move slower and they are the point.
This is why the metric change has to be agreed before the structural change. Teams that fix coverage while still reviewing on CPC reverse the fix in month two, having proven to themselves that the old setup was better.