Roll-ups: one row per rival, not eleven mentions

Every social listening tool starts by giving you a list. One row per post, one row per review, one row per comment. That list is honest about one thing: it shows what was actually said, in the order it appeared.

It is also, by default, the wrong shape for the question most teams are asking, which is not "what got said" but "who is this happening to, and how much".

Consider a competitor losing eleven users in one week. Your listener finds all eleven complaints. In the raw list that is eleven rows. In any honest accounting of the week it is one event: one rival, eleven unhappy people. The roll-up is the view that says so.

The raw feed counts posts, not rivals

A mention is a unit of attention. A rival is a unit of decision. The raw feed is denominated in the wrong one.

The distortion is worst in exactly the cases you care about most. A single viral thread about one competitor can generate hundreds of rows, while a slow, steady bleed of customers across five competitors generates dozens. Sorted by volume, the loudest incident wins. Sorted by anything that matters to your pipeline, it might not. One customer writing the same complaint on three platforms is three rows and one person. A press article reposted eleven times is eleven rows and one story.

This is not a failure of the listening tool. The raw list is doing its job. The failure is reading volume as significance.

What a roll-up actually does

A roll-up answers one question: given a pile of signals, which entities do they belong to?

Each signal is a discrete finding from a run: a complaint, a job ad, a review, a comparison thread. The roll-up groups those signals under one row per entity, and the signals stay there underneath, each with its own score, source and text. Nothing is deleted or averaged away. It is a way of reading the same data, not a different data set.

Three groupings exist because three questions do.

Per rival. Who is losing customers, shipping broken features, or hiring for the role that tells you what is broken? The rival roll-up carries the previous window and a per-day series alongside the count, which is the part that earns its keep: a rival with 40 complaints is a fact, and a rival whose complaint rate has doubled since last month is the one to aim at this quarter. A single count cannot say that.

Per account. Which watched company is showing the most demand-shaped activity this week? One company that three listeners have touched is one row, joined across all of them rather than scattered between them.

Per place. For the local side of the product, signals come from Google reviews read over a 90-day window, and you get one row per place rather than one per review, strongest first. Eleven one-star reviews at the same dental clinic is one row with eleven signals underneath.

The practical effect is a change in what you triage. With raw rows you triage mentions. With roll-ups you triage entities. The question moves from "which of these 200 posts matter" to "which of these 12 rivals is bleeding the most, and is it accelerating".

The hard part is entity resolution, not the interface

Any tool can draw a table. The difficult work is deciding that these eleven mentions belong to the same rival, because the internet does not label its entities consistently.

"River Dental" on Google Maps, "River Dental Clinic" on Facebook and "River Dental and Orthodontics" in a forum complaint may be one business, two, or three. A rival that rebranded last quarter appears under two names with no overlap in time. A common name collides with an unrelated business in another city.

Getting this wrong costs you in both directions. Merge two businesses and the roll-up shows a rival that does not exist. Split one business and you dilute a real signal across two rows that each look mild.

This problem has a name in data engineering, entity resolution, and it is older than social listening. Customer-relationship systems fight it daily: the same buyer entered twice with different spellings, one salesperson emailing them twice while another treats them as two deals.

The honest version is that resolution is never perfect. Automated grouping handles the easy cases; the ambiguous ones need a person. Which is why a roll-up that groups mentions once, silently, and never shows its work is a roll-up you should not trust. The grouping has to be inspectable, because the day it is wrong is the day it misdirects your whole morning. Every row here opens to the signals underneath, and every signal opens to its source post.

This is also why the competitor list behind a listener should come from sources rather than from recall. Hand the tool a rival list from memory and the roll-up can only group mentions under the names you remembered; miss a rival and its mentions land nowhere you will look. That is the subject of the previous post in this series, and it is the input this view depends on.

When the raw list is the better tool

The roll-up is a management view. It is not always the right one, and a tool that pushed you into it permanently would be doing you a disservice. Go back to the raw rows when:

1. You are writing the outreach. The roll-up tells you which rival is bleeding. The raw signal tells you what the specific person actually said, in their words, which is what your message has to respond to. Summaries lose the phrasing that makes outreach land.
2. You are checking the tool's work. A roll-up that groups eleven mentions under one rival is only as good as the grouping. When a row looks surprising, drill into the signals underneath. If two businesses got merged, you will see it in the texts.
3. You are investigating a spike. The roll-up says volume doubled for one rival. The raw list tells you whether that was one viral thread or thirty independent complaints, which are very different situations.
4. The count is small. Five signals across three rivals do not need a roll-up. Aggregation earns its keep at scale; at small numbers, read the thing itself.

What a roll-up will not tell you

A roll-up is a faithful presentation of the signals the listener found. It cannot fix what the listener never saw. If your queries miss the forum where your rival's users actually complain, the roll-up will show a calm, quiet rival and you will believe it.

The quality report is the counterweight: per-query and per-source yield shows which searches earn their cost and which have never produced a kept signal. A query that returns candidates but never a signal for three runs in a row is disabled automatically and the workspace is notified, so dead searches stop spending the run. The roll-up shows what you found, organised by entity. The quality report shows what you might be missing, organised by search.

A roll-up also will not decide for you. One row per rival makes comparison easy, and easy comparison invites lazy ranking. Eleven churned users at a rival with ten thousand customers is a rounding error; eleven at a rival with fifty customers is a fire. The roll-up gives you the shape of the week. What the shape means is still your judgement, and that is as it should be.

References

• Towards Data Science, "An introduction to Entity Resolution: needs and challenges", January 2025. Walkthrough of record-matching and deduplication in CRM data.
• Socialinsider, "14 Best Social Media Analytics Tools in 2026", updated September 2026 (vendor-adjacent). Cited for the category practice of per-rival comparison views with mention volume, reach and share of voice.
• Planable, "Competitor monitoring tools: 8 picks compared for 2026", updated September 2026 (vendor-authored). Cited for the side-by-side competitor table as a category convention.
• Open Pulse, "Retrieved, not recalled: why your competitor list should come from sources", on the input this view depends on.