Every assessment carries two perimeters and only one of them is written down. The first is the scope you were contracted to assess, which arrives with the purchase order and lists the services somebody already suspects are the problem. The second is the set of things you have to look at to say anything useful about the first. Most reviews collapse the two, because the second looks like work nobody has paid for.
A review of an outsourced IT service desk at a large European public-sector organisation kept them apart, and printed the difference on the page that set out its method. Four service streams were formally in scope. Ten were examined.
Two lists, drawn at different widths
The four contracted streams were the services the re-tender was expected to cover. That is the right basis for a scope of work and a poor basis for a research design, because a support estate does not experience itself in contract units. A user with a laptop, a phone, a printer and an application problem has one experience of being supported and no view at all about which supplier or which cost centre owns each piece.
The method page recorded both numbers on the same picture. Four streams marked as in scope, six more marked as additional, and the whole set of ten standing as the population the assessment method would be applied to. Nobody had to reconcile the two perimeters afterwards, because the document never pretended they were the same list.
The same page shows several independent lines of evidence converging on one assessment before any option was drafted. Widening the population of services examined adds nothing to that set. It reuses the lines already running.
The two classes of extra, and why the distinction matters
The six additional streams are usually described as bonus comparators. They were not all comparators, and the difference decides what each one is good for.
Three of them were adjacent demand. Mobile telephony, printing, and a support line for senior management. These are services the same people contact somebody about in the same working day, often without knowing which queue they have landed in. Looking at them tells you what the future scope of a support service should be, because it shows you the contacts a narrowly drawn desk will keep handing on to somebody else.
The other extras were service desks sitting outside the IT organisation, with their own users, their own contracted staff and their own definition of an acceptable wait. Those tell you something different, which is what a well-run desk looks like inside this particular institution, under its procurement route and its staff regulations. That half of the argument is already written up in How to Benchmark an Outsourced Service Desk Against Your Own Internal Desks, and I will not restate it here.
The point I want to add is that these two classes answer different questions and a review needs both. Adjacent demand shapes the scope you should buy. Internal desks calibrate the standard you can hold a supplier to. A review that collects only the second gets a benchmark and no scope argument. A review that collects only the first gets a scope argument nobody believes is achievable.
Where each class of extra is visible
A separate page mapped the whole end-user computing estate, twelve elements in all, against three categories: in scope, not in scope, and not managed by the IT organisation. Four elements carried the in-scope marking, and they correspond exactly to the four contracted streams.
That page finds you the first class of extra and only the first. Print services and mobile devices are on it, sitting one box away from the in-scope markings, so the adjacent demand worth examining can be read straight off the diagram. Nothing on that page names the internal desks. They belong to other parts of the institution, they run on their own budgets, and an estate map drawn by the IT organisation has no reason to contain them.
That is the substantive point about the third category. Its presence concedes, on the face of the scope document, that the technical boundary and the organisational boundary are different shapes. Finding the second class of extra is therefore a question about the organisation chart rather than the estate, and somebody has to go and ask it.
The same page carried a standing instruction that every support area be reviewed for impact before any new supplier agreement was signed. Read alongside the four-of-twelve marking, that instruction is an argument for the wider evidence perimeter written into the scope document itself.
A contract fixes what you have to report on. It does not fix what you are allowed to look at, and those two perimeters are almost never the same width.
What the extra six actually cost
Very little, and the reason is structural rather than a matter of goodwill. By the time the additional streams were examined, the assessment method existed, the interview programme was running and the market questionnaire had been issued. The marginal cost of an extra stream is another set of conversations and a column in the working papers.
It is not zero. It spends attention during the weeks when attention is scarcest, and it produces material the client did not commission. A finding about a service outside your scope of work can read as a criticism of a manager who never agreed to be assessed. The way through that is procedural: ask for permission to look, not for permission to report, and be explicit in the pack about which findings came from outside the paid perimeter.
What you get back is not primarily a comparison. It is the ability to describe the demand a support service actually carries rather than the demand its contract happens to name, and that description is what the next contract has to be drawn around. A review that can only see inside the current perimeter can recommend a better version of the current perimeter.
The perimeter decides the option space
The part I find most useful is one I have to mark as my own reading rather than as something the source states.
The sourcing options at the end of that review include one that folds the desk together with the rest of end-user support and buys the lot as a single wider scope. The services that option consolidates are the adjacent ones: desk-side support, hardware, print, mobile devices, collaboration and messaging, access management. An option like that is easier to put in front of a committee when somebody has already been into the services being folded in. Had the evidence stopped at the four contracted streams, the consolidation option would have been a proposal about services nobody in the room had looked at.
The evidence perimeter set at the start therefore fixes the ceiling on the recommendations available at the end. That is a stronger reason to over-collect than comparison is, and it is invisible at the moment you make the decision, which is why the decision usually goes the other way.
The same choice on an AI service desk review
I see this scoping decision being made badly right now, on assessments of whether to put an automated tier in front of a human desk.
The business case for that tier is built almost entirely on adjacent demand. Password and access requests, printing and device problems, mobile provisioning, and the large volume of requests that arrive as text and are not IT requests at all, from facilities, from HR, from finance. To an automated tier they are the same shape of contact. To a scope of work drawn on the IT service desk contract they are several different owners, most of whom were never consulted.
An assessment scoped at the contracted queue will produce a deflection estimate for that queue, a business case sized to it, and an automation boundary that stops precisely where the old outsourcing agreement stopped. The contacts most amenable to automation will sit just outside it, in the services nobody examined, and the second phase to reach them will need its own funding case a year later.
Three questions settle it early. Which adjacent queues does this population contact in the same week. Which of them are already running with a mechanism worth copying. And which findings in the final pack will have come from outside the scope of work, so the reader knows what they are looking at.
Figures and findings are as recorded on the method and scope pages of a current-state review of an outsourced IT service desk at a large European public-sector organisation, prepared to support a possible re-tender of that service. The counts are the review's own: four service streams formally in scope against ten examined, and twelve end-user computing elements mapped across three scope categories with four marked in scope. That review delivered a diagnosis, a market study, sourcing options and a transition plan, not a let contract or a moved desk. The wider scope of one of the sourcing options is the review's own; the reading that the evidence perimeter caps the option space is mine. Splitting the six additional streams into adjacent demand and internal comparators, reading the estate map as the place the first class is visible and the organisation chart as the place the second class is found, the argument about marginal cost, and everything said about automated tiers are ours.
“A contract fixes what you have to report on. It does not fix what you are allowed to look at, and those two perimeters are almost never the same width.”
Get in touch
Put RealAI’s applied-AI team on your hardest data problem.
We help enterprises move from pilots to production: sovereign models, governed data, and agents you can audit. Start with a value-first assessment.