Coordination Without Consensus: Why Your Functions Tell Four Different Stories About The Same Evidence

On 19 October 2021, the market learned something Novavax's own investors had not been braced for. Since roughly the previous March, the company's public and investor narrative had described its COVID-19 vaccine candidate as broadly on track, right up until that day's reporting pointed at an internal reality on manufacturing and regulatory progress that did not match. According to contemporaneous reporting and the subsequent securities filings, the share price fell from around $160.55 to $136.86 by the next session, a drop of $23.69 a share [1]. A securities class action, Sinnathurai v. Novavax, later settled for a $47 million fund, court-approved on 23 May 2024 [1].

Same programme. Two stories.

The easy read is that someone lied, or someone was incompetent. Occasionally that is true. Far more often it is not, and reaching for it stops you seeing what actually happened, which is more ordinary, more structural, and which no alignment meeting on earth would have prevented.

This is not a cross-functional communication problem

Reach for the easy read and you will diagnose a communication problem. The reflex says people are not talking to each other, that a silo has opened up somewhere. So book the offsite, hire the facilitator, circulate the comms plan. (The alignment-workshop industry has a great deal riding on you believing exactly this.)

Here is what that diagnosis misses. The regulatory lead and the commercial lead in that room are not failing to communicate. They have read the same data, agreed the same facts, and still walked out narrating two incompatible stories about what those facts mean. Call it "narrative fracture": one evidence base, several internally coherent and mutually contradictory accounts of what it is telling you.

If your last three alignment meetings produced firm agreement in the room that had quietly dissolved by the following Tuesday, the meeting was never the fix. Something underneath it is doing the work, and it is stronger than any meeting.

Why it happens: your functions are not reading the data, they are building it into a story

The mechanism has a name in organisational research: sensemaking. Karl Weick described it as an ongoing accomplishment, one that "emerges from efforts to create order and make retrospective sense of what occurs" [2]. Faced with ambiguous or incomplete information, a group has to build a plausible story it can act on. Nobody hands it a shared reality ready-made. Weick's study of the Mann Gulch fire, where a wildland firefighting crew's shared understanding of their situation collapsed under ambiguity, is the hard version of the point: organisational sense can fall apart faster than the people inside it can build a new one.

Drug-development evidence is ambiguous almost by design. A Phase II readout is a smudge of signal, noise and confounding that four professional communities will each bring into focus differently.

• Regulatory builds the story that can survive a label negotiation.
• Commercial builds the story that supports a forecast and a price.
• Clinical builds the story about mechanism and biological plausibility.
• Medical affairs builds the story a prescriber will accept at the bedside.

Each story is competent, and each is honest. And the currency is plausibility rather than accuracy: people settle on the account that lets them act, not the one that is provably correct. Studying actual cross-functional healthcare improvement teams, Jordan and colleagues found that the sensemaking narratives that result "tend[s] toward the nonlinear, with multiple story tellers/creators contradicting and interrupting" [3]. None of that is a sign of a broken team. It is what four expert frames sound like working one dataset at once.

I sat in a programme review some years ago where the regulatory lead and the commercial lead walked the room through the same Phase II readout, back to back. Same trial. Same tables. For a good few minutes I honestly could not tell they were describing the same study. Neither was wrong, and neither was spinning. They had built two defensible structures out of one pile of numbers, and nobody had handed them a shared object to build against.

So stop auditing your team for who is "right". Audit for which frame each story was built in. That is what tells you where the next divergence will land, before it lands.

The fix is a shared object

Here is the counterintuitive part, and it is also the oldest finding in this literature. In 1989 Susan Leigh Star and James Griesemer, studying how zoologists and administrators cooperated at a natural history museum without ever sharing a worldview, wrote a sentence that belongs pinned above every cross-functional meeting: "Consensus is not necessary for cooperation nor for the successful conduct of work" [4].

They named the mechanism that makes that possible the boundary object: something "both plastic enough to adapt to local needs... yet robust enough to maintain a common identity across sites" [4]. Boundary objects have "different meanings in different social worlds", but a structure "common enough to more than one world to make them recognizable, a means of translation" [4].

Read that last phrase again. A means of translation. You do not need your four functions to converge on one story. You need one artefact concrete enough that all four stay anchored to the same underlying facts while each translates it into its own language. A 2022 systematic review by Terlouw and colleagues traced the concept across twenty-five healthcare applications, and found it coordinates multidisciplinary professionals who do not, and need not, think alike [5].

So the goal of your next cross-functional session is simple: an object that survives the disagreement intact.

"But surely more cross-functional alignment is always better?"

Let's be honest about what that instinct is really asking for. Most team-science research does show that shared mental models and mutual trust improve performance, and the intuitive next step is to conclude that more alignment must be strictly better: get everyone to agree, lock the story, move on. It is a fair challenge. It is also exactly where the approach turns on you.

Kaba and colleagues, in 2016, put the worry on record: strong team orientation and mutual trust, they found, "may increase the risk of 'groupthink' and group conformity bias, which may lead to poorer decisions" [6]. Push a team hard enough toward agreement and you do not get a better decision. You get a quieter room, and a decision nobody left in it feels able to challenge.

Which is precisely why a boundary object beats a meeting. The meeting manufactures consensus, and consensus manufactured under deadline is where groupthink breeds. The object does the opposite. It coordinates without demanding that anyone surrender their read, so the regulatory scepticism and the commercial optimism stay visible, side by side, both answerable to the same facts. Star and Griesemer said it decades ago: consensus was never the point.

So if your alignment process ends with everyone nodding, be suspicious. Somebody's read got sanded down to get there. A healthy shared artefact leaves the disagreements legible, not dissolved.

The artefact you already have: the TPP, and why it usually isn't one

All of which stays abstract until you name the artefact, and in drug development it already exists. The Target Product Profile is the boundary object made checkable: the document that, in principle, holds every function's story against one agreed set of desired attributes and the evidence needed to earn them. If you want the mechanics of building one, we have covered that already. The harder question is why the TPP so rarely does the job it could.

Start with the aspiration. A 2024 systematic review by Ibnidris and colleagues concluded plainly that "we do not support the idea that a TPP mainly serves single unitary purposes: indeed, they rather try to serve cross-functional aims" [7]. That is the TPP working as a boundary object: one structure, many functions, each reading it in its own language.

Now the reality. Tyndall, Du and Breder, drawing on FDA records, examined 2,138 NDAs and BLAs approved between 1999 and 2015. Just 91 of them, about 4.3%, referenced a target product profile anywhere in the approval record [8]. The figure counts documented references in the approval record, not private internal use, so the true working number is unknown and may be higher. Even so, the artefact built to keep functions anchored to one evidence base is, on the visible record, absent from roughly nineteen in twenty approved programmes.

Then the honest caveat, and it matters. The FDA's own founding guidance on the TPP has sat in draft since 2007 and, as far as this research could establish, was never finalised [9]. Read its actual scope and it is narrower than the cross-functional artefact the academic literature describes. It frames the TPP around "the specific studies that will supply the evidence for each conclusion that is a labeling concept" [9]: regulatory and R&D work, the disciplines that write a label. It says nothing about commercial, market-access or medical-affairs use. So be clear-eyed: the genuinely cross-functional TPP is something sponsors and researchers built on top of a document that, on its own founding text, was never scoped that broadly. The aspiration is real. It is just not the thing the guidance ever promised.

What does the fuller version look like when it works? Talabardon and colleagues, in 2025, describe a cross-functional team at Novartis that "defined key decision points or 'tollgates' in a medicine's lifecycle where the patient perspective must be integrated", running from R&D through to "final commercialization decisions" [10]. That is a boundary object earning its keep across the whole lifecycle, from first-in-human to the medical-affairs handoff at launch. And the failure mode is just as well documented: Kieffer, in 2024, noted that oncology Global Safety Teams are "not universally tasked with the development of the risk section of the...TPP", despite owning exactly that expertise [11]. Hand the pen to one function and the boundary object quietly decays into that function's private file.

Scale the same logic up a level and you reach the Integrated Evidence Plan, the programme-level boundary object to the TPP's asset-level one a sibling argument this blog has made at length, and not one I will re-run here.

A TPP that regulatory files and pen-holds alone has stopped being a boundary object.

It is a filing.

Where this argument's evidence stops

One caveat I will not bury, because the sharper readers will have reached it already. Boundary-object theory and sensemaking are well-evidenced: in organisational science, in natural history museums, in healthcare improvement teams. They are not, as far as the published literature goes, evidenced specifically on biotech cross-functional evidence narratives. Nobody has run that study. Applying these frameworks to your clinical, regulatory, medical-affairs and commercial functions is a reasoned extension of established theory, not a finding somebody has demonstrated on drug-development teams and written up. I think the analogy is strong, and the mechanism has transferred cleanly across every setting it has been tested in. That said, hold it as a hypothesis about your own programme, not as proof. A hypothesis is precisely the kind of thing you can test.

The Monday test: boundary object, or filing?

So test it. Open your current TPP, the live version and not the one from the last board deck, and run four checks. A TPP every function can actually open and read is the kind of discipline InovaCS is built to support. But no tool does the discipline for you. Start here:

• Can your commercial and market-access leads open the current TPP unaided and point to the single assumption their forecast rides on? If they cannot find it, or it simply is not there, you are holding a filing.
• Is it genuinely co-owned, or does one function hold the pen? If safety, or regulatory, or anyone else owns a section in isolation, Kieffer's failure mode is already forming.
• When two functions describe the same readout differently, do they reconcile against the artefact, or against each other in a meeting? Reconciling against each other is how you buy consensus at the price of groupthink.
• Does the current version preserve the four different reads, or has it been flattened into one agreed line everyone signed off? A healthy boundary object keeps the disagreement legible. If yours reads as unanimous, go and find out what got quietly dropped to make it so.

References

1. Novavax investor-narrative divergence, surfaced by a 19 October 2021 Politico investigative report on manufacturing/regulatory problems (which the company disputed). The share-price move — ~$160.55 (19 Oct close) falling to $136.86 (20 Oct close), a $23.69 (14.76%) single-session drop — is as stated in the consolidated amended complaint, Sinnathurai v. Novavax, Inc., No. 8:21-cv-02910-TDC (D. Md.). The $47 million settlement fund and its final court approval on 23 May 2024 are confirmed in Novavax SEC filings (Form 10-K, FY2024) and the Labaton Keller Sucharow / Pomerantz settlement notice.
2. Weick KE. (1993). "The Collapse of Sensemaking in Organizations: The Mann Gulch Disaster." Administrative Science Quarterly, 38: 628–652. Fetched in full via US Forest Service reprint; quoted definition ("reality is an ongoing accomplishment that emerges from efforts to create order and make retrospective sense of what occurs") verified verbatim in the paper.
3. Jordan ME, et al. (2009). "The role of conversation in health care interventions: enabling sensemaking and learning." Implementation Science. PMID: 19284660. https://pubmed.ncbi.nlm.nih.gov/19284660/
4. Star SL, Griesemer JR. (1989). "Institutional Ecology, 'Translations' and Boundary Objects." Social Studies of Science. Quotes at pp.388 and 393. Accessed in full via JSTOR.
5. Terlouw G, et al. (2022). Systematic review of boundary objects in healthcare/design (25 papers included). PMID: 35113023. https://pubmed.ncbi.nlm.nih.gov/35113023/
6. Kaba A, et al. (2016). Team orientation, mutual trust and the risk of groupthink and conformity bias. PMID: 26995480. https://pubmed.ncbi.nlm.nih.gov/26995480/
7. Ibnidris A, et al. (2024). Systematic review of Target Product Profiles and their cross-functional aims. PMID: 39075460. https://pubmed.ncbi.nlm.nih.gov/39075460/
8. Tyndall, Du & Breder. (2017). Target product profile references among 2,138 NDAs/BLAs approved 1999–2015 (91; ≈4.3%). PMC5478920. https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5478920/
9. FDA. "Target Product Profile — A Strategic Development Process Tool." Draft Guidance for Industry and Review Staff, March 2007 (Federal Register notice of availability, 30 March 2007; still listed as draft in the 8 November 2017 FR notice — no finalisation located). Quoted text verified against the guidance document itself. Remains DRAFT; does not endorse cross-functional/commercial use.
10. Talabardon et al. (2025). Cross-functional patient-perspective "tollgates" across a medicine's lifecycle at Novartis. PMID: 40281373. https://pubmed.ncbi.nlm.nih.gov/40281373/
11. Kieffer et al. (2024). Oncology Global Safety Teams and ownership of the TPP risk section. PMID: 38411854. https://pubmed.ncbi.nlm.nih.gov/38411854/