Why Good Services Fail Even When Every Part Works
Most service failures do not look like failures from the inside. Every team reports against its targets, every screen tests well in usability sessions, and every policy has a defensible reason behind it. The person using the service still waits without knowing why, repeats information they have already given, or gives up somewhere in the middle. The failure is real, and almost nobody in the organisation can see it from where they sit.
If every part passes, where is the failure
A service is not a single object. It is a sequence of interactions spread across channels, teams, systems and often several organisations, assembled over years by people solving different problems at different times. The online form was built by a digital team. The letter template belongs to communications. The eligibility rules sit with policy. The case management system was procured under a contract signed before most of the current staff arrived.
Each of those parts can be assessed on its own terms and pass. The form is accessible and tests well. The letter is legally accurate. The eligibility rules are applied consistently. The system meets its availability target. Nothing on that list is wrong, and no individual on that list is careless.
What none of those assessments looks at is the space between them. The failure is rarely inside a component. It is in the join.
What gets dropped in a handover
Consider someone applying for financial support after a change in circumstances. They complete the application online, reach a confirmation screen, and wait. Two weeks later a letter arrives asking for a document the form never mentioned, because the assessment team works from a checklist the digital team was never shown. They send it. A different assessor picks the case up and asks for the same document again, because the first request was logged in a system the second assessor does not use.
Nothing in that sequence involves anyone doing their job badly. Each handover is simply a point where context falls away: what the person has already told us, what stage they are at, what they were promised and by whom. Systems are generally good at carrying records forward. They are much worse at carrying understanding forward.
Handovers between organisations are harder again, because there is often no shared record at all. A hospital discharge into community care, a council and a housing provider, a bank and a debt advice charity: in each case the person becomes the only continuous thread, and the work of joining things up quietly transfers to them. Loban and colleagues, in a 2021 cross-case analysis of multi-stakeholder partnerships in two Canadian provinces published in Health Science Reports, treat those cross-organisational partnerships as something to be deliberately built rather than assumed. That framing is the useful part. Coordination between organisations is a thing to design, not a thing to hope for.
Why silence is never neutral
A person submits a claim and receives an acknowledgement. Then nothing, for six weeks. Internally this may be entirely normal: the case is in a queue, progressing, well within the standard turnaround. Externally it is indistinguishable from having been forgotten.
People do not experience a gap in communication as an absence of information. They fill it, usually with the least favourable explanation available, and then act on that explanation. They ring the contact centre. They submit a second application in case the first one failed. They chase through a different channel and create a duplicate record. The contact centre handles those calls politely and quickly and records them as demand met, which is true, and tells nobody anything about why the call happened.
The silences in a service are almost never designed. They are the residue of a process that was mapped from the organisation's side, where waiting looks like nothing happening. From the other side, waiting is the most active part of the experience.
Who owns the whole thing
Ask who is accountable for the end-to-end experience and the answer is usually a name with no authority attached, or a governance board that meets monthly and cannot change a letter template. Every team owns its part properly. The whole belongs to nobody in particular, which means improving it is nobody's day job and always somebody's favour.
This is not only an org chart problem. Vargo and Lusch's service-dominant logic argues that value is co-created through interaction among actors rather than delivered by a firm to a passive customer, and later work in that tradition replaced producer and consumer language with actor-to-actor networks. Read practically, that says something uncomfortable. If the outcome is produced jointly by several organisations, several teams and the person themselves, then no single actor can be made responsible for it in the way a manufacturer is responsible for a product. A model that assumes one owner will keep describing a service that does not exist.
What coordinates actors when there is no conductor is what the service research literature calls institutions, or institutional arrangements: the humanly devised rules, norms and shared meanings that let many parties work to the same purpose without central control. That reading is an interpretation built on the 2018 Journal of Service Management editorial on service-dominant logic, service ecosystems and institutions, not a quotation from it. The practical implication is that shared rules, shared language and shared standards are real design material.
Platform businesses are often held up as the demonstration of this, and a 2018 chapter on tight and loose coupling in evolving platform ecosystems examines Airbnb and Uber in exactly those terms. They are also read, just as seriously, as arrangements that concentrate value with the platform and push risk outwards. Both readings are available, and this piece does not need to settle which is right.
The gap between the measure and the experience
Organisations measure parts because parts are what they can control. Average handling time. First response within the standard. Form completion rate. System availability. Each team optimises the number it owns, and the numbers are honest.
A person does not experience any of those. They experience elapsed time, effort and certainty: how long the whole thing took, how many times they had to explain themselves, and whether they ever knew where they stood. Four teams can each meet a five working day standard and the applicant still waits close to a month, with no breach recorded anywhere. The arithmetic of the whole is invisible to everyone measuring a part.
Measures also shape behaviour, which makes this worse than a reporting gap. A target on call handling time discourages the long call that would have resolved the underlying problem. A target on completion rate discourages telling people early that they are not eligible. When the measures describe the parts, the parts get protected and the whole gets neglected.
What a team can actually do
None of this requires a reorganisation, which is fortunate, because reorganisations rarely fix joins. It requires deliberately looking at what the organisation is not set up to see.
A few things tend to be worth the effort:
walk one real case end to end, including the phone calls, the letters and the waiting, rather than reviewing each part separately
name the handovers explicitly and give each one an owner, even when the two sides report to different directors
measure elapsed time and repeated contact alongside the existing operational measures
treat every silence as a design decision: say what happens next, roughly when, and what to do if it does not
make handovers carry context, not just data, so the next person knows what was already asked and promised
agree the shared rules and language between organisations early, because that is what coordination actually runs on
Journey mapping remains the practical instrument for most of this, and thinking in terms of ecosystems and institutions complements it rather than replacing it. One shows the path a person takes. The other explains why the path is shaped that way and who would have to agree to change it.
The uncomfortable conclusion is that a service can be full of competent, well-run parts and still fail reliably. Quality in the components is necessary and it is not sufficient. What decides whether the thing works is the connective tissue: the handovers, the silences, the shared understanding and the accountability for the whole. That tissue is almost always the part nobody was asked to design.