Onboarding has slipped for the fourth time. Sales says it belongs to product, product says it is a services problem, and both are working hard on things they consider more clearly theirs.
Neither team is avoiding it. You created an arrangement in which effort on this task is worth less to whoever supplies it than to whoever does not, and both groups are responding correctly.
Why these three models
The decision is what to do when a responsibility sits between two teams and keeps failing. The features that fire are effort whose cost falls on one party while the benefit is shared, a result you cannot attribute to anyone afterwards, and a failure pattern that repeats in a recognisable shape.
Three lenses. Hidden action produces an equilibrium answer about why effort falls. Monitoring noise produces a complex answer about why you cannot tell who let it slip. Behaviour modes produce a cycle answer about the shape, which you should name before you assign anything. The third goes first in practice, because the shape tells you which of the other two you are actually dealing with.
1. Name the shape before you name anyone
Shared ownership fails in a small number of recognisable patterns, and they need different fixes. Identify yours before doing anything else, because the wrong remedy applied to the right problem makes it worse.
The gap: neither team acts, each believing the other has it. Symptom is silence, and the work simply does not happen until a customer complains.
The overlap: both act, in different directions, and the outputs conflict. Symptom is rework and a customer receiving two answers.
The oscillation: ownership moves back and forth after each failure. Symptom is a reorganisation every quarter and a team that has stopped believing any assignment is durable.
And the drift: one team quietly absorbs it, uncompensated and unacknowledged, until the person doing it leaves. Symptom is that nothing appears wrong until it collapses.
Naming the pattern is more informative than analysing the last incident, because each incident arrives with its own excuse and the shape is what generalises.1 The gap needs an assignment. The overlap needs a boundary. The oscillation needs a decision that is not revisited after the next failure. The drift needs somebody to be told that what they are absorbing has been noticed, before they leave.
Telling them apart takes one question, asked of each team separately: what did you do about it last time. The gap produces two answers describing what the other team should have done. The overlap produces two accounts of work performed which do not match. The oscillation produces a reference to a reorganisation, usually with a date attached. The drift produces one team describing its effort in detail and the other unable to recall the incident. Ask both, never in the same room, and do not accept an answer about intentions.
Getting the pattern wrong is expensive in a particular way. An assignment applied to an oscillation is simply the fifth assignment, and it teaches the team that ownership statements do not survive a bad quarter, so the next one is ignored on arrival. If you have moved this responsibility before, say so when you move it again, and state what would have to be true for it to move a third time. After the second attempt, an assignment that names its own reversal conditions is the only kind anybody believes.
Four failures with four different explanations is one structure with four dates on it.
2. Why effort falls when the cost is shared
The second lens explains the mechanism, and it is not about willingness.
When an outcome depends on effort from two parties and neither bears the full cost of its failure, the effort each supplies is lower than either would choose if they carried the whole consequence. That is an ordinary hidden-action result and it does not require anyone to be shirking.2
Each team is optimising a real objective. Sales is measured on closed deals and onboarding is downstream of their number. Product is measured on shipped roadmap and onboarding is a services activity. Both are behaving well against the scorecard they were given, and the shared item is the one where their private cost of effort exceeds their private share of the benefit.
Which is why exhortation fails and is nevertheless the most common response. A meeting in which both teams agree that onboarding matters changes no one’s incentive, and the agreement is sincere and lasts about a fortnight.
The structural fixes are few and they are all about who carries the consequence. Assign the outcome to one team, with the decision rights and the budget attached, and let the other be a named contributor with a defined obligation. Or, where it genuinely cannot be split, put the shared outcome in both teams’ measures with real weight, so that failing it costs both of them something they care about.
If you cannot move budget or headcount this quarter, and most founders cannot in the middle of a year, there is a partial version that costs nothing. Assign the decision right on its own: name who decides when onboarding and roadmap collide, and put it in writing to both teams. Budget follows later. The decision right is the piece that ends the nine-day wait, because most of that delay is nobody being permitted to choose rather than nobody being available to work.
Expect the team you assign it to push back, and take the pushback seriously in one specific respect. If they ask what comes off their list, they have understood the assignment and you owe them an answer that names something. If they accept it without asking, the assignment has probably not landed, and it will be absorbed quietly by whoever on that team is least able to refuse. Ask them yourself what is being dropped, and get a name for it before the week ends.
What does not work is assigning responsibility without the decision right, which is the most common half-measure. A team told to own onboarding while product controls the roadmap and sales controls the commitments has been given the consequence without the levers, and it will fail in a way that looks like their fault.
3. You cannot see who let it slip
The third lens explains why this persists after everyone has agreed it is a problem.
The outcome you observe is a joint product, and you cannot decompose it. When onboarding slips, the visible facts are consistent with sales having promised something unusual, product having deprioritised a fix, or the customer having been slow to supply data. Each team has a true account in which they behaved reasonably, and the noise is genuine rather than convenient.3
Two consequences follow and both are uncomfortable.
The first is that any judgement you make about who is at fault is a guess dressed as a finding, and the teams know it. A founder who adjudicates these repeatedly, on evidence that cannot support a verdict, is spending credibility on coin flips.
The second is that adding reporting does not fix it. More status updates about a jointly produced outcome produce more descriptions of the same unattributable result. What changes the situation is measuring something each team controls alone: for sales, whether the handover pack was complete on the day of signature; for product, whether the two known blockers were closed. Those are attributable, and they are the parts each team can actually be held to.
There is a further reason to avoid holding either team to the joint number, and it is not only about fairness. A scheme can only be built on quantities that are observable and verifiable, and a target neither party can move alone fails that test, which is why it produces argument rather than effort.4 An outcome nobody controls alone is a topic, not a target.
A worked version makes the substitution concrete. Instead of onboarding completed within fourteen days, which neither team controls, sales is measured on handover packs complete at signature and product on the two named blockers closed by a date. Both are unilateral, both are checkable on a Monday, and together they cover most of what was actually failing. The joint number still gets reported; it is simply no longer what anyone is held to.
Unilateral measures have a failure of their own and it arrives around the second month. A handover pack scored on completeness gets completed, which is not the same as being useful, and product begins receiving packs with every field filled and nothing in them a builder can act on. The guard is to let the receiving team define what complete means, and to revisit that definition once a quarter with the people who actually use it. A measure owned by the party that supplies it drifts toward whatever is cheapest to supply.
There is also the case the split does not cover, which is the customer’s own delay. A good share of nine-day waits are a customer who has not returned a signed form or a list of user names. Neither team controls that, and putting it in either scorecard recreates the problem you just removed. Give it an owner for the chasing rather than for the outcome: somebody sends the reminder on day two and day five, and the clock is stopped while you are waiting. What gets measured there is the chase, not the reply.
The general form is worth stating because it applies well beyond onboarding. When you cannot attribute an outcome, do not try harder to attribute it. Find the upstream action each party controls unilaterally and measure that instead.
What the three say together
- Name the pattern first: gap, overlap, oscillation or drift. Each needs a different fix.
- Assign the outcome to one team with the decision rights and budget attached, not the responsibility alone.
- Stop adjudicating the incidents. The evidence cannot support the verdict and everyone knows it.
- Measure one thing each team controls alone, upstream of the shared outcome.
Where they disagree
The incentive argument and the attribution argument recommend opposite structures.
The hidden-action view says give it to one team outright. Single ownership restores the link between effort and consequence, and any shared arrangement reintroduces the problem however carefully it is written.
The attribution view says that some outcomes genuinely require both teams, that assigning it to one produces a team accountable for something it cannot deliver alone, and that the visible result will be an owner who fails and a contributor who is blameless, which is worse than the ambiguity you started with.
The reconciliation is to split the object rather than the accountability. Almost every jointly produced outcome decomposes into parts that are individually owned: the handover pack, the two blockers, the customer’s data, the first-week check-in. Assign each part to one team with a date, and let the shared outcome be reported rather than owned. What fails is leaving the outcome shared and hoping goodwill closes the gap, which is where most companies sit because it requires no decision.
What none of them contain
None of the three prices the cost of the assignment itself. Giving onboarding to one team takes capacity from whatever they were doing, and the decision is frequently avoided precisely because nobody wants to say what gets dropped. The ambiguity is sometimes a way of not making a resourcing decision.
None of them handles the case where the two teams disagree about what good looks like. Sales may want fast onboarding and product may want thorough onboarding, and no allocation of ownership resolves a genuine disagreement about the objective. That has to be settled first and separately, and assigning ownership on top of an unresolved objective just gives one side the win by default.
None of them handles the case where the boundary itself is wrong. Some responsibilities sit between two teams because the teams were drawn along the wrong lines, and reassigning the task repeatedly is a way of avoiding the observation that the organisation is shaped for a business you no longer run. That is a more expensive conclusion and it is occasionally the correct one.
And one property the ensemble will not produce: the person absorbing the drift is usually the most conscientious person in either team. Fixing the structure removes work from someone who never asked for credit and may read the change as a loss of scope. That conversation is worth having deliberately rather than discovering it in their resignation.
The one action that survives the ignorance: before the next meeting about this, write which of the four patterns you are in, and name one upstream action each team controls alone. That converts an unwinnable argument about whose fault it was into two commitments that can be checked next week.
Who has to move
The founder, because neither team can assign the decision right to itself and both are correctly protecting their own measures. The instinct is to convene both and ask for better collaboration, which produces sincere agreement and no change in incentive, then repeats after the next failure. The cheapest first test is the pattern name plus two upstream measures, which takes twenty minutes and produces something checkable, where the collaboration meeting produces a feeling that lasts a fortnight.
Sources and notes
- Donella H. Meadows, Thinking in Systems: A Primer, Chelsea Green Publishing, and John D. Sterman, Business Dynamics: Systems Thinking and Modeling for a Complex World, McGraw-Hill. The practice of identifying a system’s behaviour over time before explaining any single event is central to both, as is the finding that a structure produces its characteristic behaviour regardless of which individuals occupy it. Used in section 1 for naming the pattern before assigning responsibility.
- Jeffrey Carpenter and Andrea Robbett, Game Theory and Behavior, MIT Press. The hidden-action problem, and the result that effort falls when a party does not bear the full consequence of its own contribution, are developed in the chapters on asymmetric information and contracts. The team-production version, in which a jointly produced outcome cannot be decomposed into individual contributions, is treated in the same volume. Used in section 2, and presented as a structural prediction rather than a claim about anyone’s character.
- Jeffrey Carpenter and Andrea Robbett, Game Theory and Behavior, MIT Press. Imperfect monitoring, and the consequence that punishing or blaming on the basis of a noisy joint outcome imposes cost on parties who behaved well, are developed in the chapters on repeated games. Used in section 3, including the practical corollary that the remedy is to measure an upstream action each party controls unilaterally rather than to add reporting on the joint result.
- Jeffrey Carpenter and Andrea Robbett, Game Theory and Behavior, MIT Press. Mechanism design, and the requirement that a scheme be built on quantities that are observable and verifiable, underpins the distinction drawn throughout between assigning responsibility and assigning decision rights: a party held to an outcome it cannot influence has been given an obligation the scheme cannot support.
A note on a number this article does not give. There is no rule for how much weight a shared measure needs before it changes behaviour. It depends on what else is in each team’s scorecard and how those are rewarded. What transfers is the diagnosis order: pattern, then decision rights, then one attributable upstream measure per team.
Joshua Agonya Pi’Rwot, Founder.