Case study: Account Intelligence for a Supplier That Sells Into Pharma

On 12 October 2026, a query to the ClinicalTrials.gov public API returned 8,681 interventional studies that were recruiting and had an industry lead sponsor. Of those, 3,285 were tagged Phase 1, 2,890 Phase 2 and 1,785 Phase 3, with trials spanning two phases counted in both. The API dated its own data snapshot 9 October 2026, three days before the query.

For a supplier that sells to pharma and biotech companies, that registry is one of the main public records of what its accounts are doing. Account intelligence for life-science sales turns records like it into a decision: which account, which product, which message, and when. That is the question underneath searches for healthcare commercial intelligence platforms that provide real-time market intelligence and identify funding announcements for prospect targeting.

We built a commercial marketing hub that answers it for a global life-science supplier. It researches the supplier's pharma accounts, recommends products to the account manager and builds the commercial package. It prioritises products in line with the supplier's strategy, produces the outreach strategy and tracks the results.

The situation: customers whose plans are only partly public

The supplier's buyers are pharma and biotech companies, and their needs follow their pipelines. A company taking a programme into a larger trial needs different things from one that has just stopped one.

Many commercial intelligence cases in our library are built around a different kind of account: the hospital, the health system or the prescriber. Medtronic's case study on Definitive Healthcare's site describes data on US hospitals and integrated delivery networks, with decision-makers' contact details and prescribing behaviour. Its only stated result is that Medtronic "identified new contacts, new markets, and new opportunities". As our library entry notes, there is no figure, baseline or date. It shows provider intelligence in use, and says nothing about selling into a pharma company's laboratory or plant.

A pharma account's signals sit in trial registries, securities filings and company announcements. Each runs on a legal clock and has a blind spot.

Two blind spots matter most. First, the US registration rule defines an applicable drug trial as one "other than a phase 1 clinical investigation". Many are registered anyway, as the Phase 1 count above shows, so a missing record proves nothing.

Second, the public record over-reports survivors. BIO's 2011-2020 clinical success-rate study warns that larger companies may not report failed Phase 1 programmes they consider immaterial. The same study found a Phase 2 to Phase 3 transition rate of 28.9 per cent across 4,933 transitions. A Phase 2 asset on an account's pipeline is more likely to stop than advance. Our post on the Phase 2 gap sets that baseline against others.

What we built: six jobs in one hub

The hub does six jobs, each harder than it looks.

• Research the account. Pipelines, market position and needs are three different questions. The first two have public evidence. The third is an inference, and should be presented as one.
• Recommend products to the account manager. It goes to a person, not the customer, because the account manager knows what no public source holds: the last conversation, the contact who left, the renewal date.
• Build the commercial package. Our post on agentic MLR review argues that the costly defects in pharma promotion sit in the claims library, not with the reviewers. A supplier is not under drug-promotion rules, but the lesson carries over: a package is only as accurate as the approved product statements behind it.
• Prioritise products in line with the supplier's strategy. Explained below.
• Produce the outreach strategy. Who hears what, through which channel and when. Channel choice is where the law bites hardest.
• Track the results. Without this, none of the other five can be judged or improved.

Why the ranking answers to strategy, not propensity

A propensity model predicts who is likely to buy, learned from who bought before. For a supplier that creates three problems.

It learns the old portfolio. Products with long sales histories score well. A new product has no history at all, even when the strategy needs it to grow.

It scores the customers who would have bought anyway. Uplift modelling estimates the effect of the action instead. Gutierrez and Gérardy's review notes that it "crucially relies on randomized experiments". A propensity score cannot tell an account manager whether the call changed anything.

It feeds on itself: recommendations shape what gets sold, which becomes the next training data. Jiang and colleagues analysed these feedback loops in recommender systems in 2019.

Strategy carries what the sales history does not: which products to grow, which to retire, where capacity and margin sit. A strategy-led ranking can also be argued with. The account manager can ask which priority put a product first.

The same order of work shows up in Evaluate's competitive-intelligence case for a top-20 pharma company, dated 11 March 2026. According to Evaluate, the programme first agreed success metrics and expected catalysts with stakeholders, then ranked events by "impact, likelihood, and timing". Its consultants also attended eight congresses. Our library entry records only qualitative impact, with no figures. The ranking rules came from the business before monitoring began.

What it achieved, and what to measure instead of a headline number

We are not publishing performance figures for this hub. For a buyer, the measurement plan is more useful, because published numbers in this category often measure something narrower than they appear to.

IQVIA's Alerts case, dated 25 April 2025, is a fair example. For a large pharma client launching an antipsychotic, daily alerts reached sales teams as little as 48 hours after a clinical event. The page lists a cut in lead-to-engagement time from five weeks to 48 hours. That is a speed figure. The page does not show what the engagement achieved, such as any change in prescribing, and the client's results sit in a separate download. Our library entry files it as launch alerting, which is what it is.

These are the measures we would apply to any account intelligence system, ours included.

The fifth row is the one most often skipped, and the only one that shows whether the hub changed sales. Without a held-out group, a rise in orders could simply be the market.

Acceptance needs its own warning. A high acceptance rate can mean good recommendations, or an account manager who has stopped checking. The reader-study evidence in our post on human oversight shows how far expert judgement drops when a suggestion is wrong. Recorded rejection reasons are the cheapest guard against that.

What makes it work, and where it stops

"Real-time" is bounded by the record. A recruitment status change can legally reach ClinicalTrials.gov 30 days late, and a Form D 15 days after the first sale. Even the API snapshot was three days old when we queried it. A hub cannot be faster than its sources.

Needs are inference. A Phase 3 start or a funding round suggests what an account may buy next, but does not say so. The useful output is a stated inference with its evidence beside it, for the account manager to test in conversation.

The strategy has to exist in written, ranked form. If the priorities are vague, the ranking will be too, and that dependency sits with leadership, not software.

The rules it has to satisfy, position current at 12 October 2026

A pharma company's pipeline is not personal data. The name and work email of a scientist or buyer there are, and most of the law below attaches to them.

Under the GDPR, Recital 47 says direct marketing "may be regarded as carried out for a legitimate interest". The EDPB's Guidelines 1/2024 set three cumulative conditions: a legitimate interest, necessity, and a balancing test the individual does not win. On 12 October 2026 the EDPB page still lists version 1.0, adopted for consultation in October 2024. The Court of Justice held on 4 October 2024 in Case C-621/22 that a commercial interest can qualify, if the processing is strictly necessary and passes the balancing test.

Two GDPR duties catch account research in particular. Article 14 requires contacts whose data came from other sources to be told within a month at the latest. If you contact them sooner, the first message must carry the notice. Article 21 makes objection to direct marketing absolute, and it must be flagged at the first communication.

Email is governed separately, by Article 13 of the ePrivacy Directive as each country transposed it. The proposed ePrivacy Regulation was withdrawn, with notice published on 6 October 2025, so the 2002 Directive still applies.

Three rows need comment. CAN-SPAM's definition of commercial email turns on primary purpose and has no carve-out for business recipients. The tracking row matters for results: counting email opens in the EU needs consent unless an exemption applies. Orders and recorded conversations avoid the problem. Finally, net impression and fair balance, the standards in our post on OPDP letters, apply to prescription-drug promotion, not to reagents sold to a laboratory.

One product rule may still bind. If the catalogue includes reagents labelled for research use only, 21 CFR 809.10(c)(2)(i) exempts them from full diagnostic labelling only while they are not represented as effective diagnostics. Outreach that implies diagnostic use undoes that.

What this means in practice

If you want the same result, the order matters more than the tooling. These steps run from longest lead time to shortest.

1. Write the strategy down first. Ranked product priorities, an owner for each, and a review date. This needs leadership agreement, so it takes longest.
2. Settle the legal basis before collecting a single contact. Document the legitimate-interest assessment and map the email rule in each country you sell into. Draft the Article 14 notice and the objection wording for first contact.
3. Agree the measures and set aside held-out accounts before launch. A comparison group cannot be added afterwards.
4. List your public sources with their legal clocks and blind spots. Require a date and a link on every pipeline statement.
5. Pilot on a sample of accounts with the account managers who own them. Record every rejected recommendation and the reason.
6. Track results through orders and recorded conversations. In the EU, do not count opens with pixels unless you have consent.
7. Judge uplift after a full buying cycle for the products concerned, not after the first month.