Every review of an outsourced function eventually needs something to compare the function against. This one built its comparison in two sections. The first carried three precedents from other organisations: a transport operator that had consolidated its end-user computing under one supplier, a healthcare research body that had moved onto a modern service management platform, a pharmaceutical manufacturer whose knowledge sat scattered across its support teams. The second carried two comparators from inside the same organisation being reviewed.
The second section is the one that produced anything you could act on.
One of the two internal comparators was another service desk entirely. It faced customers rather than staff, it sat inside its own directorate alongside two sibling units, and by the review's own structure it was the model rather than the patient: the slide is headed with what the IT desk should learn from it, not the reverse. The other was the facilities and switchboard line, outsourced like the IT desk, whose contract printed a reward mechanism with real numbers in it.
The argument this piece exists to carry is narrow and, once stated, hard to argue with. An internal comparator beats a market benchmark when the constraints are shared. Same employer, same procurement route, same pay bands, same labour market, same appetite for risk in the legal department. Anything the sibling function has managed to do, you can do, because every excuse available to you was also available to it.
The challenge
Look at what the three outside precedents could actually tell a reader who had to make a sourcing decision.
The transport precedent reported a significant reduction in incidents after the whole end-user estate went to a single supplier who owned the value chain. Directionally useful, and no more: no baseline, no denominator, no cost. The healthcare research precedent reported operating cost cut by half over two years on the back of modern tooling and heavy use of cloud collaboration. Same problem, larger claim. The pharmaceutical precedent named a real diagnosis worth having, that knowledge sharing across support teams was sporadic with no consolidated place for knowledge and service artefacts to live, and then stopped: four quadrants of that slide were marked "not in scope of this engagement", among them pricing and the service level and KPI quadrant.
So the outside section could not answer the two questions the review had been commissioned to settle: what to pay for this, and what to measure it on.
Meanwhile the IT desk's own scoreboard was in the state you would expect of a function nobody had a comparator for. One-call resolution reported at 47 percent against a contract target above 65 percent. Close to 95 percent of incidents logged at the lowest priority. The tool could hold three priority values; the contract had written its targets against five. Incident acknowledgement and incident resolution carried no measured actuals at all. The self-service portal took 15 of roughly 4,620 contacts in a month. Agent utilisation, the measure nearest to productivity, appeared in neither the service levels nor the KPIs.
The approach
The internal comparison ran along three lines, and each one returned something the outside precedents had not.
On experience, the customer-facing side stated the operating principle plainly: fix problems through the process, because on the IT side the working model was that the loudest escalation got the fastest fix. The review recorded a director routing a problem through a business application team because the desk could not resolve it, and described personalised handling for senior users as needed and absent. That is not a soft finding. Every bypass is an unlogged ticket, and a demand picture assembled from tickets is missing precisely the cases that were hardest.
On contract terms, the sibling facilities line printed the mechanism. Speed to answer contracted at 94 percent of calls inside 20 seconds. Recording the incident in the system within 20 minutes. Accuracy of incident categorisation named as a KPI in its own right. A reward regime on two independent clocks: an SLA bonus worth 5 percent of contract value for the period under review, generally settled half-yearly, with a clause dividing it equally between the supplier company and the agents themselves; and a monthly balance of 50 credit points against which each error deducted a pre-defined 10 to 20 points, with a further bonus for holding the balance and a penalty for running it negative.
On compliance, the customer-facing side had documented its workflows roughly two years before the review, and it had begun breaking its audits into three domains: process, product and content. Auditing the process tells you the workflow ran. Auditing the content tells you the answer in the knowledge base was right. Almost nobody scores the second one, which is why knowledge bases quietly rot while the process dashboard stays green.
One detail deserves to be read against the review's own recommendation. The customer-facing desk was staffed with 10 internal people and 11 contracted from an outside supplier, 21 in all by arithmetic rather than by any figure printed on the slide. The model desk was majority contracted. Whatever separated the two desks, it was not employment status. It was documented process, commitments that could be measured, and a reward that reached the person doing the work rather than stopping at the supplier's revenue line.
The outcome
What this engagement produced was a current-state picture, a comparison and a set of recommendations ahead of a re-sourcing decision. No desk was rebuilt, no contract was re-let, and the sibling desk's superiority rests on the practices it had and the IT desk lacked, not on a measured head-to-head. Nobody had ever put the two desks on one scoreboard. That absence is the most repeatable part of the whole finding.
Three things would be done differently now.
The comparison was assembled from interviews, which means it captures what managers remember rather than what their systems did. Both desks ran ticket tools and both sat behind telephony that counts things automatically. Process mining over those event logs gives the same comparison with a source attached and a refresh cadence, and it settles arguments that interviews only relocate. The same review put two different volumes for the same month on one slide, 4,620 contacts counted across the channel table beside a chart showing 5,784 incidents, and nothing reconciling them; two slides later it called 844 escalations about 20 percent of the incidents logged, which implies a third base near 4,200. Write the denominator dictionary before the first chart, or every ratio downstream is a guess with decimal places.
Second, the unit sitting beside the customer desk whose entire job was learning from user behaviour is the part worth copying, and it is a data-readiness decision disguised as an org-chart decision. A permanent team that owns interaction data, its definitions and its lineage is the precondition for anything later: for model deployment against a resolution target, for straight-through processing of the routine request classes, for the early and cautious language-model pilots now being run over support content. Our Platform work starts at that inventory rather than at the model, because a pilot without it produces a demonstration and a pilot with it produces a baseline.
Third, the incentive problem is still open, and the source is admirably honest about it without meaning to be. One section raises rewarding a supplier for answering more questions as a gaming hazard that has to be managed carefully so value for money is measured properly. A later section proposes output metrics of exactly that shape as the reward basis. That is not carelessness. Output metrics are the only things cheap enough to contract on and the easiest thing in the world to optimise against. The same trap is set today wherever an automated first line is paid by contained contacts rather than by resolved ones.
The best available comparator for a function you are about to re-source is usually already on your payroll, filing under a different directorate, and nobody has thought to ask it anything.