Audience research and signal architecture for paid social

Interest stacking has largely stopped being a targeting lever. Past roughly 50 conversions a week, broad delivery beats hand-stacked interest sets in almost every comparison we have run, and the platform expands beyond your selection regardless of what you picked. What still moves the number is the research that stacking was standing in for: what the buyer was doing in the hour before they went looking, what they were afraid of, and which words they use for the problem. So this engagement produces language, segments and an exclusion map. It does not produce a list of interests. If a list of interests is the deliverable you want, someone will sell you one for a tenth of this price. How the research is done: Ten to twenty recorded interviews with recent buyers and recent non-buyers, run by us rather than by a panel provider. Non-buyers are the expensive half and the half that pays: people who considered you and chose otherwise will state the objection your own customers have already forgiven. Review and complaint mining across sources that carry weight in this market, Trustpilot, Prisjakt, Flashback threads, sport-specific communities, and the Facebook groups where switching advice actually circulates. Search demand read for the language of the problem rather than the language of the category. And sales call recordings where they exist, which in logistics and energy they usually do, and which are the best data in the building. What comes out of it: A language bank. The buyer's own phrasing for the problem, the objection and the outcome, in Swedish and English, sourced line by line so your writer can see where each phrase came from and how often it appeared. A message-market matrix mapping segment against objection against proof, where segments are defined by trigger rather than by demographic. And a trigger calendar: in energy that is fixed-contract expiry, moving house and the first cold invoice of the season; in sports and fitness it is event registration windows; in logistics it is tender cycles and peak capacity planning. Signal architecture, which is what replaced interests: Customer file segmentation by value rather than recency alone, so lookalike seeds come from the top decile of contribution instead of everyone who ever bought. Value-based lookalikes where the platform has enough matched records to build them, and a plain statement when it does not rather than a source list that quietly underperforms. Exclusion architecture, which in energy is the largest source of waste we find anywhere: an existing contract holder shown an acquisition offer costs you twice, once for the impression and again at the renewal conversation. Retention windows set per segment with the reasoning written next to each one. Not included: No media buying, no creative production, no landing pages. We hand the language bank and the matrix to whoever writes, and we will sit in the first briefing to defend the research. After that it belongs to you. No quantitative survey work either: if you need statistically projectable segmentation, that is a different discipline and we will name two firms who do it properly. Who this is not for: Anyone who cannot introduce us to real customers. The interview half is not optional and cannot be substituted with your NPS verbatims or a summary from your account managers. Also not for teams who already have this work and want a second opinion on it; validating someone else's segmentation is not worth what this costs. What you receive: The language bank, the message-market matrix, the trigger calendar, the audience and exclusion map ready to build in Business Manager, the interview recordings and transcripts, and a two-hour session where we take the objections apart with your team in the room.