The Risks You Already Agreed to Carry
It is eighteen months after launch. The product works exactly as specified. Every technical commitment held. And the organisation could not sell what it agreed to launch.
R&D has a mechanism for tracking technical risk into that outcome. It's called the risk register. Product Managers (PMs) don't typically have an equivalent document, so on a programme that runs three or more years, trade-offs made months apart compound quietly until the gap between “the product works” and “the business case holds” is exactly this one.
Concordia was the Roman goddess of harmony and agreement, shown holding a pair of clasped hands. Rome honoured her because agreement was scarce, contested, and worth commemorating when it finally arrived. Every Roman politician understood the corollary: concord could be announced, but the clasped hands only told you the fighting had stopped. They told you nothing about whether the argument was over.
Late in a new product development (NPD) programme, most rooms are in concord. That may be the problem.
No single issue is ever big enough
Here's what the agreements look like from a PM's perspective. Same PM in the room each time. Different meeting, different quarter, different context each time.
- Eight months before launch, a comparative sensitivity claim comes off the evidence list. The head-to-head study against the incumbent ELISA takes eleven weeks, but the evidence freeze for the regulatory and claims package falls in nine, and the lab slot to start it is already booked by another programme.
- Six months out, a workflow gap on the liquid handler integration gets absorbed by field applications, who agree to cover it with a training module.
- Ten weeks out, the distributor onboarding package gets cut to a datasheet and a webinar. The technical writer is committed to the instrument launch until the quarter turns, and the PM approves the cut rather than pull the writer off a live launch.
Each decision was reasonable given what the room could see that day. Nobody was wrong on any given day, including the PM, who signed off on all three.
Notice what the three have in common. Each had an owner in the room, usually the PM. What none of them had was a second owner, someone whose job was to hold the three together as one item and ask what they add up to. R&D's risk register won't do that job, and that's not a flaw in the register. A register is built by the people who own technical delivery, so it answers one question well: will the product work. The PM's question is different: will this organisation be able to sell what it agreed to launch. Only one of those questions has a document behind it.
Quick check: list the three biggest trade-offs from the last nine months and see if they appear in the risk register. They typically don't.
How often does this actually cost somebody a launch
Around 35 percent of launched medical and life sciences products fail commercially, and Matt Wilkinson has written on why that rate is closer to a process variable than a fact of life. The best performers fail at 24 percent. Everyone else fails at 46 percent. Same sectors, same buyers, nearly double the rate.
I'm not going to claim compounding trade-offs explain that gap. Nobody has measured it. What I will claim is narrower: whatever else is driving that spread, most candidate explanations at least show up somewhere. Weak market sizing shows up in a TAM model. A technical shortfall shows up in the risk register. Compounding trade-offs don't show up anywhere. That's not proof they're the leading cause. It's proof your organisation currently has no way to check.
Deferring is not the failure. Losing count is.
Sometimes the deferral is right. Trading a support burden for a first-to-market window can be worth every hour of it. What matters is documenting the assumptions around the trade.
Somebody has to answer a specific question: what risk is this organisation knowingly carrying, and did the people with authority to accept it actually accept it? That question separates a deliberate risk from one that settled by default across nine months of individually reasonable meetings.
That's the question a premortem answers.
One premortem exercise, three jobs
Gary Klein's premortem is usually described as a way to surface unknown risks before a plan launches. Run correctly, late in an NPD programme, it does two more jobs most teams never ask of it.
Here's how to run it. Most late premortems fail right here: stipulate a specific deferral, like the liquid handler workflow gap, and you'll get four people restating the workflow gap in four registers, not new information. Stipulate the commercial outcome instead:
It is eighteen months after launch. The product works exactly as specified. We could not sell what we agreed to launch.
Nobody can retreat to whether the assay performs, because you've already conceded that it does. Run the premortem several months before your launch decision, with the people who've been in the programme long enough to remember what was traded away. Everyone writes silently first, then round-robins one item each, then discusses. Don't open with discussion, because the first confident voice sets the frame and the room calibrates to it. Don't invite the decision maker. Their presence collapses the candor the exercise depends on. Their turn comes after, once the list is ready for them.
1. Surfacing risk nobody has named yet
Stipulating failure and asking what happened gives the room permission to raise causes no official meeting has surfaced, because “here is what happened” is forensic and “I object” is political. Apply two filters before anything moves forward. Ask what observation would prove the cause wrong, and discard anything no evidence could contradict. Then ask whether it names a mechanism, an actor with veto power, a document, or a timing constraint: “customers may be conservative about switching reagents” names nothing, while “a core facility director declines the reagent unless we supply lot-to-lot concordance data” names the person who says no and the document they need, which tells you who to call.
2. Retrieving the trade-offs you already made
Job 1 finds things the room didn't know. Job 2 retrieves things the room already knew, and it isn't relitigating them: the stipulation keeps everyone answering what happened, not who was wrong, so naming a conceded item here isn't reopening whether the call was right, it's making that call visible next to the others for the first time. Sort every cause into the new-assumption pile (nobody had raised it before this session, so it gets tested or validated with customers) or the conceded pile (the room already knew, and had already decided to accept it). Writing the conceded pile down here is the first time your programme's full set of trade-offs exists as one document, each item decided on a different day by different people for a different reason, never assembled until now. What goes to the decision maker isn't a verdict on any single call, it's the forward question of whether the organisation still wants to carry the combined risk now that it's visible: sort each item into a timeline if more work will resolve it, a customer if only they can answer it, since asking beats waiting, or a decision-maker recommendation if the answer is a judgement call nobody outside the company can make, then give each accepted item one leading indicator and a review date, since one without the other is a confession and one with it is a control.
| Pile | What it contains | What it is for |
| New assumption | Nobody had raised it before this session | Test it, then validate it with an advisory panel or a few customers |
| Conceded trade-off | The room already knew, and had already decided to accept it | A forward-looking call: does the organisation still want to carry it, now that it's visible next to the others, not a verdict on whether the original decision was right |
3. Why this works: legitimizing dissent in a room that already agreed
This job is what makes the other two possible. A room that's already agreed doesn't lack information, it lacks permission to doubt, and stipulating failure supplies that permission: it cuts confidence roughly twice as much as a pros-and-cons or cons-only discussion (Kahneman, Thinking, Fast and Slow; Veinott, Klein & Wiggins, 2010). One caution: the technique increases how many reasons people generate, not how good those reasons are (Mitchell, Russo & Pennington, 1989), so sorting is still job 2's work, not this one's.
Concordia is a point in time, not a sum
A room in concord has done the work. Months of hard calls, everyone informed and aligned by the end. What no one owns at that point is the sum of those calls.
Run the premortem while there's still time to act on it. Stipulate the commercial failure and grant the technical success. Sort the causes into what you learned and what you already knew. Send the conceded list and its recommendations to the decision maker while you go get evidence on the new assumptions.
No one risk on that list was ever big enough to stop the launch. Together they're big enough to stop the sale, and to leave the business case's revenue forecast unexplained eighteen months after everyone signed off on it being right.
FAQs
If I hand leadership a list of nine things the organisation quietly conceded, do I own those nine things forever?
Only if you walk in alone, so don't. Before the list reaches the decision maker, take it to the functional leader whose area carries the largest item on it, usually commercial or field applications, and ask them to co-sign the recommendation rather than review the list. Two names on a list is staff work. One name is a complaint, and the person who filed it owns it. If no functional leader will co-sign a concession, that tells you something about how real it is.
Why not run this earlier, before the business case is approved, when the investment can still be saved?
Run one there too, because early premortems are valuable. They will not produce this list. At business case approval nothing has been traded away yet, so there is no accumulated position to inventory. The late version depends on there being something to concede and on there still being runway to act. The early one protects the investment. The late one protects the commercial outcome, which is the claim you can still make, the segment that will still qualify you, and the support load you can still resource.
