Pharma brand teams buy a lot of “listening.” What they usually get back is a mood: volume up, sentiment slightly down, a word cloud that contains the product name and the disease name. That is not patient unmet-needs research. It is a weather report.
The question a brand lead actually has is narrower. What are patients and HCPs struggling with, in this indication, in language we could put in front of medical? If the answer cannot be pointed at a sentence someone said in public, it should not go in a brand plan.
That is the job IndicationIQ is built for. This note is the method, not the pitch.
Sentiment is the wrong unit of work
Sentiment is a summary statistic. It collapses thousands of posts into a number that is easy to chart and hard to act on. “Negative” can mean a side-effect, a reimbursement fight, a clinic that ran out of stock, or a journalist’s take. Those are different problems with different owners.
Unmet need is a claim: people in this situation lack something specific. Examples that are useful:
- “Nobody explained what titration would feel like in week two.”
- “The only support group in my language is a Facebook page that has not posted in a year.”
- “I cannot find an HCP who has used this combination outside the trial sites.”
Those sentences can go to medical, to patient engagement, to creative, to access. A −0.2 sentiment score cannot.
Public conversation is a biased sample — use it anyway
Anything scraped or collected from public web and social sources is biased. The loudest patients are not the average patients. HCPs who post are not a representative advisory board. Bots, advocates with an agenda, and brand-adjacent accounts all exist.
That is not a reason to ignore the channel. It is a reason to treat it as signal generation, not epidemiology. The output should be a shortlist of needs to verify — in qualitative research, in medical information logs, in advisory boards — not a prevalence estimate you would defend to a regulator.
What public conversation is unusually good at:
- Naming problems that do not appear in the SmPC as patients experience them
- Surfacing workarounds (dosing, devices, access routes) that field teams hear late
- Catching HCP frustration that never makes it into a sponsored advisory
- Showing which competitor narratives are already occupying a need
What it is bad at: incidence, causality, and anything you would put in a promotional claim. Do not.
The unit of evidence is a quote
If a finding cannot be tied to a quote, I do not keep it. Themes without sources are how listening reports turn into fiction. A theme is allowed only as a grouping of quotes, and the grouping has to survive a sceptical medical reviewer asking “where did you get that?”
Practically, that means:
- Collect public posts, comments, forum threads, and similar artefacts in a defined indication and time window.
- Extract candidate needs as short propositions, not keywords.
- Attach the supporting excerpt and a stable reference (URL, date, speaker type: patient, caregiver, HCP, other).
- Deduplicate so the same rant in five places is one need with multiple supports, not five insights.
- Drop anything that is a promotional claim, a rumour about an unpublished trial, or a personal identifier you should not be storing.
The last point matters. This is public conversation, not a CRM. You are looking for classes of need, not named individuals to market to.
Why enterprise listening suites still miss this
The big platforms are built for brand safety, share of voice, and crisis. Those are real jobs. They are not the same as pulling named unmet needs for a single asset or a single launch.
Three structural mismatches:
- Taxonomies are brand-centric. Mentions of the product, the competitor, the disease. Unmet need often appears without any of those strings — “I still cannot sleep on this dose” does not contain your brand name.
- Dashboards reward volume. Rare but decisive needs (a device failure mode, a specific access catch-22) get buried.
- Evidence is a screenshot in a slide. There is no durable claim-to-quote object a cross-functional team can reuse next quarter.
If you already pay for an enterprise suite, keep it for the jobs it does. Run a separate extraction pass for needs. That pass can be software; it can also be a disciplined human with a spreadsheet. The method matters more than the logo on the tool.
What to do with a list of needs
A good output is boring on purpose:
- Need statement in plain language
- Who said it (role, not identity)
- Quote
- How many distinct supports, roughly
- Whether it is already in your medical narrative, a competitor’s, or nobody’s
Then you triage. Some needs belong in patient support. Some belong in HCP education. Some belong in access. Some are interesting and not yours to solve. Some are promotional landmines — true in the wild, not usable in materials.
That last category is why this work sits next to promotional compliance, not instead of it. Finding a need in public does not give you the right to claim you solve it.
Where IndicationIQ fits
I built IndicationIQ because I wanted that extraction loop as a product: indication in, quote-backed needs out, usable by a brand team that does not have a 12-person insights department. It currently lives under Elliogen and will likely become its own company if the usage matches the gap.
If you want the tool, it is at indicationiq.app. If you want help scoping the research question before anyone logs in, get in touch.