"Intent" has become one of those words that means whatever the vendor selling it needs it to mean. That is a problem for buyers, because the underlying signals differ enormously in what they can support, and a team that treats them as interchangeable will over-trust one and under-use the other.
Two different things wearing the same word
Third-party intent data infers interest from behaviour elsewhere on the internet — content consumption across publisher networks, search patterns, review-site activity. Its strength is that it can see accounts that have never encountered you. Its weakness is attribution: it is usually probabilistic at the account level, often noisy about which company a signal belongs to, and it cannot tell you whether the person reading about your category was a buyer, a competitor, or a student.
First-party ad engagement is what happened with your own campaigns. Its strength is certainty: you know exactly which creative was served, to which company, to which job function, when. Its weakness is scope — it only sees people you paid to reach.
Neither is better. They answer different questions, and the failure mode is expecting the certainty of the first from the coverage of the second, or vice versa.
What ad engagement is genuinely good at
Three things, and they are worth more than they are usually given credit for.
Detecting committee formation
Nothing else in your stack sees this as clearly. When a second job function at the same account starts engaging within a few weeks of the first, something has happened internally: someone forwarded something, or a problem got raised in a meeting. It is a directly observed event, not an inference, and in most programmes it is the strongest available predictor of a real opportunity.
Ranking within a known list
You already decided these 400 accounts are worth pursuing. Engagement tells you which twelve to call this week. That is a much easier question than "who in the world is in market", and it is the question sales actually has.
Telling you what an account has already seen
Underrated, and instantly useful. A rep who knows the account has seen the migration-risk piece four times can open on migration risk. This changes the quality of a first conversation more than any score.
What it cannot see, and will not tell you it cannot see
Being explicit about the limits is what keeps the signal trustworthy:
- Accounts you never served. An account with zero engagement might be uninterested or might have never been delivered an impression. These are opposite conclusions and they look identical in a score.
- Off-platform activity. A committee can research you thoroughly through your website, your documentation, a peer recommendation and three review sites without ever engaging with an ad.
- Negative interest. Someone reading your content carefully because they have chosen a competitor and are writing the justification document engages exactly like a buyer.
- Timing. Engagement tells you an account is interested. It does not tell you their budget cycle, and in a lot of B2B, timing is most of the outcome.
Combining sources without fooling yourself
Keep the axes separate. Three of them, reported separately, and combined only at the point of prioritisation:
- Fit. Firmographic and technographic. Should this account buy? Stable, slow-moving.
- Reach. Have we actually got in front of them, and how broadly? A precondition, not a signal.
- Engagement. What did they do with what we showed them? The behavioural signal.
Third-party intent, if you buy it, is a fourth axis and belongs at the top of the funnel: it is best used to decide which accounts to add to a list, not to rank accounts already on one. Once an account is on your list and you are advertising to it, your own engagement data is both more precise and more current.
The reason to keep them separate rather than blending into one number is diagnostic. A high-fit account with low reach is a media problem. A high-reach account with low engagement is a creative problem. A high-engagement account with no sales follow-up is a process problem. Blend the three and every one of those becomes a mid-range score with no indication of what to do about it.