A diagnostic across the service operation of an international health insurance group returned one line that reframed everything written after it. The contact centres are sub-scale, under a hundred seats, and supporting multiple stand-alone units in each region will be uneconomic.
That reads like a cost finding. It is a shape finding, and the difference decides which options on the table can ever pay. A hundred seats is a small operation by any benchmark anyone would hold it against. Spread that same small operation across regions, each with its own languages, its own hours of business, its own expectations of what good service looks like, and it acquires the coordination burden of a business many times its size while still buying technology at the price a hundred-seat operation gets quoted. The overheads belong to a global business. The purchasing power belongs to a small one.
Read against headcount, the operation looks like a candidate for consolidation and nothing more. Read against footprint, the shortlist changes.
The challenge
The diagnostic named the tension on the same page as the finding and left it unresolved, which was the right call because preference does not resolve it. Locally relevant service was identified as an enabler of the regional growth strategy. Consolidation would take cost out. Localisation was what the growth case depended on. Both were true at once, and any design that quietly picked one over the other would have been answering a question nobody had settled.
The availability requirement made the same point operationally. Members had to be served around the clock, every day of the year, and the design space was bounded by two end points: follow the sun across sites, or shifts of eight hours at a single centralised centre. Language provision and cultural fit were named as the variables that choose between them. Neither end point is a headcount decision. Follow the sun buys sites. Centralising buys every required language under one roof, on a rota. Both price the footprint, and the number of people employed is an output of that choice rather than an input to it.
Then comes the part that quietly decides where money goes. Of the three components under review, two carried a value at stake and one did not. Digital and real-time service carried eight million pounds, with benefit drivers of increased leads, conversion and renewal rates. Automation carried twenty-two million, resting on a five percent reduction in average cost and a ten percent increase in claims handled per person. The global service model carried no figure. Its stated benefit driver was a productivity increase from customer analytics and agent support tools, and its value was left explicitly open.
That asymmetry is not sloppiness. It is what happens when the components that scale with transaction volume are easy to model and the one that scales with the number of places you operate is not. Volume gives you a unit and a rate. Footprint gives you a fixed cost repeated in every territory, which is the thing a per-transaction benefit case has no column for. A shortlist assembled the usual way funds the two priced items and defers the unpriced one indefinitely, and the unpriced one carries the shape problem.
The approach
The proposed work was three stages across fourteen weeks: six weeks to scope the change and evaluate options, five to design the service model, three to capture implications and plan implementation, with workshops acting as gates between them. The method was hypothesis-based, framing opportunities as propositions to test with the client rather than researching them to exhaustion. Stage one produced evaluated options and a dependency map, not a build recommendation, which for an operation of this shape is the correct order.
The automation thinking held the same discipline. Targets were to be agreed process by process against best practice, rather than as one organisation-wide number. The reason matters more than any single target. The benchmark range of fifteen to thirty percent came from personal lines, where volume is high and competitive pressure constant. Importing a scale business's answer into a sub-scale one is how automation programmes fail politely: the target lands, the volume behind it does not, and the payback is never quite found.
The warnings attached to that view were more useful than the headline, and most of them were cautions. Automating the front of a claim too aggressively pushes cost downstream rather than removing it. Starting a process on incomplete information is a false economy, because entering all the data at the door outweighs the extra work it takes. And automation depends on the pricing behind the rating engine being accepted, because underwriters who do not accept it find workarounds to a lower rate or softer terms, damaging the result and the efficiency together.
Underneath all of it sat a data problem no amount of design would have fixed. The diagnostic found that the operation could not continuously gather information from all customer touch points, and that only a few satisfaction scores were captured at all. The ambition for the digital component, meanwhile, was an experience covering more than ninety percent of member interactions, and the recommended way to find self-service candidates was analysis of actual enquiries rather than assumption. Those two do not sit together. You cannot mine an enquiry population you are not recording, and you cannot place underwriting, claims handling, the clinical helpline or the assistance desk region by region when the regional evidence is a handful of scores. Instrumenting those touch points is a precondition of the design rather than a later phase of it, and it is where our Platform work begins.
The outcome
What this engagement produced was a diagnosis and a proposed design, priced as options. The eight and twenty-two million figures are estimates of value at stake in a proposal, not savings anyone had banked, and no target operating model was running. The named deliverables were options defined and evaluated, a service model designed across customer and partner segments, cost modelling, a cost benefit analysis and a plan for tracking the benefit case afterwards.
Held against the shape finding, though, the diagnosis does something a headcount review would not. It sorts the options by how they behave as the footprint grows. Seats get worse: every new territory adds a fixed minimum of people, supervision, premises and rota coverage, and none of it is shared with the territory next door. An integrated agent desktop with a single customer view, full interaction history, telephony integration and drafted reply suggestions gets better, because it is bought once and amortised across every place the group operates. Self-service and the digital estate behave the same way. For a business that is small in every place and present in many, that is the only economics that improves with spread.
That is the reframing worth carrying out of this. The question is not how many people the operation employs. It is how many separate places it has to be correct in, and which of the things on the shortlist get cheaper per place rather than more expensive.
Two things we would now do earlier. Process mining over the enquiry and claim event logs replaces the workshop that produces the automation candidate list, with evidence that can be refreshed rather than a view that ages from the day it is agreed. And the early, cautious language-model pilots appearing in document handling and reply drafting belong under the same test. At under a hundred seats in one centre, a drafting assistant saves a rounding error. Applied once across the whole footprint, against an enquiry population that has been instrumented properly, it is one of the few purchases whose value rises with the number of territories. Our Consult engagements open by asking which side of that line an investment falls on.
A hundred seats in one place is a small contact centre. A hundred seats in each of many places is a different business, and it should be priced like one.