Launch Software Watches the First Six Months. The Misses Start Earlier.

On 30 June 2025 IQVIA published the ninth edition of its Launch Excellence study, which ranks innovative launches across eight core markets. Its most durable finding has held since the first edition: more than 80% of launches "continue as they start in the first six months". Between six months and two years, only 13% of non-orphan launches made a major improvement in their ranking. Just 9% of the latest cohort were Excellent, in IQVIA's terms, in two or more countries.

That six-month window is the commercial case for biopharma launch excellence software. If the trajectory is set early, anything that shortens the time from signal to action should be worth buying. This piece asks a narrower question. Which launch failures does the published evidence attribute to things software can see, and which to decisions taken long before the dashboard went live?

How often do launches under-perform, and against what?

The headline numbers disagree because they measure against different baselines. Quoting one without its baseline is how launch post-mortems go wrong.

Deloitte adds a point for anyone buying a tracking tool: products that meet analyst forecasts in year one usually keep doing so. General medicines did worst, with 58% missing. Across the decade, Deloitte wrote, "we do not see a significant change in launch performance, overall".

Trinity's public page does not say whose forecasts it used, so treat its figure as directional.

IQVIA's label is relative by design. To be Excellent, a launch must beat its peers on three tests: early trajectory, sales well above the median, and outperformance against competitors. It is a ranking, not a pass mark, so most launches cannot be Excellent whatever software they run.

The fourth row is the uncomfortable one. Kossmeier and colleagues compared the forecasts companies must file with the Austrian reimbursement authority against actual reimbursed sales. For 55.9% of products, the forecast was more than double the actual figure or less than half of it. Accuracy did not improve across the decade. Three McKinsey authors reached a similar verdict on analyst consensus forecasts in Nature Reviews Drug Discovery in 2013: many are wrong, often substantially.

So part of what gets called launch under-performance is forecast over-performance. A dashboard that reports variance to plan inherits every error in the plan. And the incentive to draw a steep plan is financial. Robey and David, two Pharmagellan consultants, note that a 20% shorter projected time to peak sales can radically change an asset's apparent value.

Why launches miss: the reasons on the record

The most useful public evidence on causes is Deloitte's. Its analysts read the analyst reports on 91 of the 98 launches that missed. They sorted the reasons into six categories, often more than one per launch.

1. Market access. Formulary exclusion, heavy utilisation management, high patient cost-sharing, weak health-economic evidence or larger-than-expected rebates.
2. Poor product differentiation. Prescribers not convinced the product beats the standard of care.
3. Inadequate understanding of the market and customers. Including the inability to target patients and prescribers.
4. Unfavourable clinical findings. Unanticipated trial results leading to a boxed warning or reluctance to prescribe.
5. Low prioritisation or resourcing. Too few representatives, or a weak patient support programme.
6. Unexpected events. Patent litigation, manufacturing delays, supply disruption and, for 2020–2021, the pandemic.

Larger companies missed more often for reasons 2 and 3. Deloitte's explanation is organisational: clinical teams take commercial input too late, or an acquired asset arrives with its choices made. It recommends commercial input on complex therapies as early as phase 2. Smaller companies missed more often on clinical findings.

IQVIA adds stretched health systems, tighter budgets and fewer prescriber interactions. It cites the European Patient W.A.I.T. survey: on average across the EU27, only 46% of 173 innovative medicines approved centrally in 2020–2023 were publicly reimbursed by January 2025. In the US, IQVIA estimates that manufacturers fully funded half of prescriptions for newly launched brands in 2022, up from under 10% in 2018.

The peer-reviewed literature on uptake agrees. A 2014 systematic review by Lublóy of 35 studies named the strongest predictors of early prescribing. They were scientific commitment, high prescribing volume, exposure to marketing and contact with colleagues. A 2021 review by Medlinskiene and colleagues of 66 studies added cost, formulary and reimbursement conditions, guidelines and patient factors.

Read against software, the six categories split cleanly. Only the third is mainly a data problem. Access shows up in data but is decided by payers and health systems. Differentiation, clinical findings, resourcing and most unexpected events are decided by the product, the trial and the budget.

What launch excellence software actually covers

"Launch excellence" is sold as one category. It is really five, with different buyers, data and evidence. IQVIA's white paper lists the AI uses it expects, from faster insight and personalised content to a launch cockpit for early signals. IQVIA sells several of these, which is worth bearing in mind.

Only one layer has peer-reviewed evidence behind its logic. The rest rely on vendor case studies that measure time and visibility, not uptake.

Where AI genuinely earns its place

Targeting by network, not only by volume. The strongest evidence here is not from a vendor. Donohue and colleagues, in PLOS ONE in 2018, studied 11,958 Pennsylvania physicians and three first-in-class drugs. A 10-percentage-point rise in adoption among peers who shared patients raised a doctor's own adoption by between 5.9% and 8.3%. Physicians in the top decile of patient-sharing networks had nearly twice the influence of those in the top decile of prescribing volume.

Most target lists are still ranked by volume, so they miss the people who move their colleagues. Building the network from claims is graph analysis, not generative AI. Language models help upstream, matching one physician across publications, trial registries, claims and CRM records. The study has limits: one state, three drugs, no data on promotion.

Signal synthesis in the first six months. Field notes, advisory-board transcripts and medical information enquiries arrive faster than anyone can read them. Summarising and clustering that text is a task language models do well. Latency matters too. IQVIA reports that daily alerts cut one launch team's time from lead to engagement from five weeks to 48 hours. That measures speed of contact, not prescriptions.

Content throughput. Launch content is a volume problem, and drafting and mechanical pre-checks are where agents help. Our pillar on what agentic MLR review automates sets out what can be automated and who still signs. The one disclosed architecture reaches just over 70% recall, so reviewers still read everything. Claim-level checks also cannot see net impression, the standard OPDP's recent letters turn on. For accelerated approval products, content throughput also sits on the approval critical path, as the table below shows.

What four launch case studies show, and what they don't

Each comes from the vendor's own page. That does not make them false, but nobody independent checked the baseline.

An oncology manufacturer mapped access a year after launch. In the Norstella (MMIT) case, the manufacturer bought claims and transactional records from several vendors and linked them by token. The vendor's page says the data still lacked plan-level detail: the team could see doctors, drugs and denied claims, but not each patient's insurance plan. MMIT matched the records and appended plan, payer, channel, formulary, tier and coverage status, flagging where sources disagreed.

The reported result is visibility, with no uptake figure. The question was asked a year after launch, well past IQVIA's six-month window. And the decisive step was a record join against a payer database, not a model. The data the brand had paid for could not answer its own access question.

A health-system barrier, fixed by account managers. In the MMIT IDN case, a midsize company's oncology therapy had good payer coverage but low adoption. A panel of health-system decision-makers showed where it stalled. At three integrated delivery networks the brand moved onto formulary. At four more, the barriers were workflow: separate patient pathways in the electronic record, competing order sets and extra lab tests. The vendor reports a 17% rise in utilisation over six months.

There is no stated baseline or comparison group, and no split between formulary wins and workflow fixes. What it shows is the Medlinskiene finding in practice: payer coverage is not access. The data located the problem. People fixed it.

An unplanned prescriber group, found within weeks. In the Within3 case, a top-five company launched an acquired chronic-disease therapy in an unfamiliar area. Its plan assumed endocrinologists and primary care doctors. Within3's page, dated 21 October 2025, says referral anomalies and online discussion showed psychiatrists and therapists starting treatment conversations. The company widened medical education and field targeting to reach them.

This is the clearest case of AI synthesis earning its keep. But the outcome is a counterfactual, a delay of six months or more avoided, with no prescription numbers. And the miss it corrected is Deloitte's third category, in an acquired asset, exactly the situation Deloitte flags. The software spotted it quickly. The segmentation was a human decision.

Creative concepts in three days, approval not measured. In the Indegene case, a global company preparing a launch swapped a 12- to 16-week agency process for three days of workshops. The vendor's page reports a creative timeline about 97% shorter, over 100 concepts against the usual 10 to 15, and concept costs about 50% lower. An MLR pre-flight layer flagged regulatory risks, and 11 concepts went forward.

The clock stops at concepts. There is no time to MLR approval, no count of review rounds and no launch outcome. It is the same gap between drafting and approval we found in AI cycle-time claims in clinical development. Faster drafting is real capacity. It moves the launch date only if drafting was the bottleneck.

The failures no software decides

Against Deloitte's six causes, the software in these cases surfaced a targeting miss, located access barriers and sped up content. None touched differentiation, clinical findings, resourcing or supply.

Differentiation is decided in the trial. Deloitte's advice on phase 2 commercial input is about endpoints, indication order and formulation. No post-approval dashboard can make a product more different than its data allow.

Access is decided by payers and health systems. In Europe, joint clinical assessment now covers new cancer medicines. But the HTA Regulation leaves pricing and reimbursement to member states. In the US, funding half of new-brand prescriptions yourself is a commercial choice, not a software output.

Resourcing and launch count are portfolio decisions. IQVIA found in 2013 that the more launches a company made, the lower the share that were Excellent. The ninth edition still finds a weak version of that pattern among top companies. A cockpit can show a field force spread across three launches. It cannot decide which launch gets it.

Forecast belief is governance. The Austrian data show company forecasts running optimistic for a decade without improving. A model can produce a better-calibrated forecast. Whether leadership adopts it is another matter.

Supply is upstream. Deloitte counts manufacturing delays among its unexpected events. As we argued on what serialisation data can predict, demand-side data sees a supply failure only after it happens.

What binds when launch tools touch promotion, patients and payments

Position current at 12 October 2026. Neither FDA nor EMA has issued guidance on AI in commercial launch planning or targeting. FDA's January 2025 draft on AI credibility is scoped to evidence on safety, effectiveness and quality, not commercial operations.

Two practical notes. HIPAA binds the data source, not usually the manufacturer. But a patient-event trigger sent to a representative is only as defensible as the de-identification behind it. Ask the vendor for the expert determination before go-live. In Europe, the Article 14(5)(b) exemption for disproportionate effort is written mainly for archiving, research and statistics, not for building commercial target lists.

What this means in practice

Set the work in lead-time order. Most of it starts before any launch software is switched on.

Phase 2 to early phase 3. Put commercial input into endpoint and indication choices. This is where differentiation is decided, and no tool does it.

About 24 months before launch. Back-test your own forecasts. Compare pre-launch forecasts with actual first-year sales for your last five to ten launches, as Kossmeier did for Austria. If your median error is 30% optimistic, every variance-to-plan dashboard will report a crisis that is partly arithmetic.

18 to 24 months. For new cancer medicines and advanced therapies in the EU, plan the joint clinical assessment dossier now. Orphans follow from 13 January 2028.

12 to 18 months. Build the data spine: claims linked to plan-level coverage, patient-sharing networks alongside volume rankings, and health-system formulary status. Get the expert determination for patient-level data and the Article 14 notices for EU expert profiles. The Norstella case shows the cost of waiting until a year after launch.

9 to 12 months. Measure MLR throughput on real launch content: assets per week, review rounds per asset and time to approval. For accelerated approval products, the first 120 days of materials are due during FDA's pre-approval review. Pilot pre-review on your own past assets before trusting a vendor's recall figure.

3 to 6 months. Choose five to eight early signals, each with a named owner and the decision it triggers. Examples are new prescribers by specialty, formulary status at your top 20 health systems, and time from referral to first dose. A signal without an owner is just a chart.

Months 0 to 6. Review weekly and act inside the window. IQVIA's data say fewer than one launch in five recovers substantially after a poor start. The software's value is the weeks it saves. The launch's value was mostly fixed by the steps above.