For the chief operating officer
No figure for how many roles this removes — anyone quoting one before seeing your volume has guessed, and you are the person who would be held to the guess. And we do not take a function without the authority to run it: handing over accountability while keeping every decision produces a slower version of what you have now, plus an invoice.
The volume does not care. It arrives at the same rate whether the roster is full, and on the days it is not, the backlog forms in the morning and is still there at close. Whoever did turn up absorbs the difference, which is how you lose the ones who turned up.
Attrition is the quiet arithmetic underneath all of it. Somebody leaves, the requisition takes weeks, the replacement takes longer to become useful than anybody planned for in the staffing model, and during that whole window the seasoned staff carry double. Then one of them leaves, and the cycle has fed itself.
The seasonal peak is worse, because you have to commit to hiring for it before you know whether it arrives, and you carry the cost either way. Overshoot and you are paying for idle capacity. Undershoot and the backlog runs into the following quarter and drags a service commitment with it.
And when the shortfall bites, what gets dropped is never the visible work. It is the follow-up nobody sees, the second attempt at a contact who did not answer, the note that would have made the next handover clean. None of that shows on a dashboard until several weeks later, arriving as a complaint or a lost account rather than as a staffing figure.
The gap is between the work that arrives and the hours that show up to meet it — and software you have to staff in order to run does not close a staffing gap, it relocates it.
That is why the honest offer here is the work rather than the licence. Give us a bounded function — one queue, one workflow, one shift — and it is operated: our technology and our people, against your service definition, reporting into your operation.
Which is only worth anything if the authority comes with it. An operated function needs the decisions it makes daily to be delegated in writing, with the boundaries drawn explicitly and the exceptions routed back to a named person on your side. Where an organisation wants the function operated but every decision approved, the result is measurably slower than the original process and costs more, and we would rather say that at the start than discover it together in month two.
The boundaries matter as much as the grant. There are decisions we will not take regardless of what is delegated — anything determining a person’s eligibility, employment, credit, housing or benefit, and anything ordering a queue of people, because ordering decides while appearing not to. Those route to your named person every time, and that is a design position rather than a caution.
And no number for headcount. What can be committed to is a service definition — this volume, this response window, this quality standard, measured this way — with the staffing left as our problem rather than converted into a promise you would be held to.
Whether the backlog survives a short roster — measured by queue depth at close on the worst-staffed day of the month.
How long the oldest waiting item has waited — measured by age of the oldest open item, checked daily rather than averaged monthly.
Whether the invisible work happens — measured by second-attempt and follow-up completion rate on contacts who did not answer first time.
How the operation absorbs a seasonal peak — measured by response window held during peak versus the same window in an ordinary week.
Whether supervisors stay off the floor — measured by supervisor hours spent on coaching against hours spent covering absence.
How much of the function depends on one experienced person — measured by whether the workflow holds when that person is away, and what breaks when it does not.
What proportion of items leave the boundary — measured by count routed to your named person, which should be stable and legible rather than growing.
a headcount reduction, or a figure for one. If your finance team needs a number before starting, the honest sequence is a bounded first function measured against your own baseline, and the number afterwards.
The work happens in your systems. Case management, records and ticketing stay where your people already look, so a supervisor checking on an item does not have to learn a second place to check and your own reporting keeps working unchanged.
Telephony and messaging route through what you already run, which matters more than it sounds: a customer should not be able to tell which items were handled by the operated function and which were not, and a separate channel is the first thing that gives it away.
Handover is the part most often underestimated. Where an item leaves the boundary it arrives at your named person with the history attached rather than as a bare ticket, because a handover that loses its context creates the rework it was meant to avoid.
Operating a function means our people and technology touch your customers and your records, which is a larger grant than a software licence and should be examined as one.
The lines below are the ones an operations executive should press on. Where the basis is design rather than attestation, it says so — an operated relationship is a poor place to discover that a word was doing more work than it could carry.
An operated relationship raises questions a licence does not: who is the employer of record, who supervises, what happens to work in progress if it ends, and whose obligation a given regulatory duty remains. Those belong at the start.
The short answer on the last one is that a regulatory duty stays yours. Operating a function does not transfer an obligation you hold, and any supplier suggesting otherwise is describing something that does not exist in your regulator’s framework.
A single queue or a single shift with a clear edge — enough volume that a month of it means something, small enough that ending it costs a scope rather than an operation.
Pick the function that hurts most on a short-roster day, not the one that is easiest to describe. The easy one produces a clean pilot that proves nothing you needed to know.
Write the boundary down before anything runs: what is inside, what routes out, and who receives it when it does. Boundaries described verbally get redrawn under pressure by whoever is most inconvenienced, which is never the person who owns the outcome.
And take your own baseline first. Queue depth at close on the worst day of last month, age of the oldest open item, second-attempt completion rate. Measured beforehand these settle the question afterwards. Reconstructed afterwards they are an argument.
Nobody can tell you that before seeing your volume, your mix and your exception rate, and a supplier who quotes a figure has guessed at something you would be held to in front of your board. What can be committed is a service definition — this volume, this response window, this quality standard — with staffing as our problem rather than your promise. Run one bounded function against your own baseline for a month and the number is then a measurement instead of a sales figure.
Then the honest answer is that you would be buying a slower version of your current process with an invoice attached, and we would rather say so now. An approval queue in front of every decision reintroduces exactly the wait the arrangement exists to remove. The workable middle is a written boundary: routine decisions delegated with the edges drawn explicitly, and a named list of decisions that always route to your person. If that list has to include everything, this is not the right purchase for you.
That is your decision and it is not ours to characterise. What we would say from watching this go well and badly is that the arrangements that hold are the ones where the operated function absorbs the volume nobody had capacity for — backlog, second attempts, the peak — rather than the ones framed internally as a replacement, which tend to lose the experienced staff you most needed to keep during the transition.
You own it either way, which is the actual argument for bounding it hard. One queue, one shift, a written stop condition, and a baseline taken before it starts, so that failure costs a scope and is legible early rather than arriving as a quarterly surprise. If the first function cannot be bounded that tightly, it is the wrong first function — not because the work is unsuitable but because you would be unable to tell quickly whether it was working.
No, and any supplier implying otherwise is describing something that does not exist in your regulator’s framework. The duty stays with you. What an operated arrangement can do is make discharging it more evidenceable — every action producing a receipt naming the actor and the authority relied on, and the decision types your rules reserve to a person excluded by design rather than by policy. That is a stronger position for an examination than most in-house processes can produce, and it is still your duty.