Most large services can be summarised in a dashboard.
Throughput.
Cost-per-case.
Time-to-resolution.
Volume processed.
If the lines move in the right direction, we call it progress.
This is not accidental. It reflects a deeply embedded management model, one that treats services as production systems. The work is decomposed into stages. Variability is managed. The “back office” is protected from the messiness of human interaction. Certainty is engineered through standardisation.
In 1989, operations scholars argued in the Harvard Business Review that service organisations should be designed more like factories1. Their recommendation was explicit: reduce variability by decoupling customer contact from operational processing. Optimise throughput. Measure performance in units.
For transactional problems, this logic is elegant.
If I need to renew a licence, pay a fine, or update my address, I want speed, consistency and minimal friction. A standardised system is a gift. Money in, goods out.
The difficulty begins when we fail to distinguish between transactional and relational problems.
Relational services - social care, housing support, probation, public health - do not “produce” outcomes in the same way a factory produces cars. Their value is co-produced. Stability, trust and behavioural change emerge through ongoing interaction. The user is not an input; they are an active participant in the system.
Factory logic struggles here.
When relational complexity is forced into a throughput model, the system behaves in predictable ways:
It standardises edge cases.
It privileges measurable outputs over lived outcomes.
It optimises the interface rather than the relationship.
It generates demand caused by its own design.
John Seddon described this as “Failure Demand” - Demand caused by a failure to do something, or do something right for the customer2. In factory terms, this is rework. The production line created 100 widgets the wrong size. In social systems, it becomes something else: escalation, mistrust, crisis.
Consider the Windrush scandal.
The policy objective was clear: reduce unlawful residency and tighten compliance. The operational system was built accordingly. Cases were processed against standardised data requirements. Documentary evidence became the unit of truth. Throughput and enforcement activity became measurable indicators of performance.
The National Audit Office (NAO) report on handling the Windrush situation3 exposed that when individuals with complex, decades-long histories entered this system, the factory encountered variability it had not designed for. The response was not to slow down or reinterpret the relationship between citizen and state. The response was to process the case according to specification.
From a factory perspective, the system behaved coherently. It applied the rules it was built to apply. The harm emerged because a relational reality - identity, belonging, trust in the state - was being adjudicated through a throughput mechanism.
This is not isolated to public services.
During the peak of the UK energy crisis in 2023, contractors acting on behalf of British Gas forcibly installed prepayment meters in households that had fallen into debt. The operational objective was straightforward: reduce arrears and secure payment. Targets were set. Processes were defined. Third-party contractors were mobilised.
The system reduced debt exposure. It also generated public outrage, regulatory investigation and long-term reputational damage, far graver - it put vulnerable people in danger45.
Again, this was not a failure of individual morality alone. It was a predictable outcome of optimising for throughput in a context defined by vulnerability and hardship. Poverty cannot be managed through a dashboard in the same way inventory can.
Leadership often selects factory logic because it feels controllable. It promises a linear relationship between investment and output: £X equals Y cases processed. It allows performance to be visualised. It provides the appearance of certainty.
But complex social systems do not guarantee binary relationships between input and outcome. They are adaptive, participatory and contingent.
When we frame service design as a craft that “delivers” outcomes, we risk performing what might be called Delivery Theatre6. Interfaces improve. Transactions accelerate. Digital journeys become seamless. Meanwhile, the underlying relational system remains unchanged.
We have spent the last decade building faster, more beautiful conveyor belts.
In many cases, they are moving in exactly the wrong direction.
The failure of most ambitious transformation programmes is not a lack of user research. Nor is it a failure of intent. It is a logic problem.
We are applying throughput optimisation to systems that depend on discretion, trust and co-production. We mistake efficiency of interaction for effectiveness of outcome. We protect the back office from variability rather than designing for variability as a structural condition.
The Service Factory is not a metaphor. It is a management inheritance. It has shaped how we measure performance, allocate funding and define success.
It works exceptionally well - when the problem is transactional.
When the problem is relational, it does not merely underperform. It systematically erodes the very conditions that make relational services viable.
If a service depends on trust, participation and adaptation, it cannot be optimised for throughput without altering its nature.
The question for service design is not how to make the factory more humane.
It is whether the factory is the right model at all.
https://hbr.org/1989/07/the-service-factory
https://en.wikipedia.org/wiki/Failure_demand
https://www.nao.org.uk/reports/handling-of-the-windrush-situation/
https://www.bbc.co.uk/news/articles/c308vzqj975o
https://www.ofgem.gov.uk/press-release/ofgem-statement-british-gas-prepayment-meter-installations
https://happygeneralist.com/delivery-theatre/