18 November 20257 min read
The pitch for outsourcing is usually framed as capacity: you get more people working on your backlog. What is rarely said out loud is that more people working on your backlog means more of your existing people's time spent deciding what those people should do and whether what came back is right.
That time is real work. It has a cost, it comes out of the highest-value hours you have, and it appears in no budget because it appears on no invoice.
What the time is actually spent on
- Specification: turning a backlog item into something someone outside your building can execute without guessing.
- Question handling: the questions that arise anyway, and the cost of each one being asynchronous rather than a turn of the head.
- Review: reading the work, deciding whether it meets the definition of done, and writing the feedback if it does not.
- Unblocking: access, credentials, environments, a decision only you can make, a dependency on a team that has not been told.
- Planning and prioritisation: the recurring ceremony that keeps the queue full, which is now a queue for two groups instead of one.
- Relationship work: the calls, the escalations, the quarterly conversation about how it is going.
Notice that only two of these are about the provider being external. Specification and review are work you would do anyway for an employee. The difference is that with an employee the specification can be thinner, because the missing context is filled in by someone who sat through the last two years of the company.
The load is front-loaded, and people forget that
In the first weeks, almost everything is a question, because nothing has been asked yet. The context that an experienced team member carries silently has to be transferred piece by piece, and every transfer costs one of your people an interruption.
This curve descends, sharply, if the engagement is stable. It resets, entirely, when people are replaced. That is why continuity is a cost term and not a nicety, and why a provider's staff turnover deserves a question during selection even though it feels like an odd thing to ask about.
Overlap hours change the price of a question, not the number
Working hours that overlap with the provider do not reduce how many questions get asked. They reduce what each unanswered question costs. A blocking question raised at ten in the morning and answered at eleven costs an hour. The same question raised into a time zone that wakes up after you go home costs a day, and it costs it every time.
This is the whole of the time-zone tax, and it is paid in your management time rather than in the rate. A shared working day is not a comfort feature. It is the mechanism by which the coordination cost of an external team stays within the same order of magnitude as an internal one.
Who carries the load is a commercial choice
Staff augmentation means the provider supplies people and you supply the management. Managed delivery means the provider supplies both, and prices accordingly. The rate difference between them is not margin, it is the cost of the coordination layer, moved from your side of the ledger to theirs.
Which is better depends on a number only you can produce: what your internal management hours cost, and whether you have them to spare. A team whose lead is already the bottleneck should be extremely suspicious of the cheaper model, because the saving is denominated in the one resource they do not have.
Outsourcing does not remove management work. It decides who does it, and the rate difference between the two models is the price of that decision.
Measure it for two weeks
You do not need a study. Pick two ordinary weeks and have everyone who touches the engagement note, roughly, how long they spent on it and in which of the six categories above. Round to the half hour. Total it, apply your own internal cost per hour, and annualise.
The result is usually larger than expected, and its distribution is more interesting than its size. If most of it is specification, your backlog is not written for an external reader. If most of it is review, your definition of done is doing no work. If most of it is unblocking, the problem is access and dependencies, and that is the cheapest of all of these to fix.
When the number tells you to change model
If measured management time is consuming a meaningful share of a senior person's week and the engagement is expected to run for a year or more, the arithmetic frequently favours paying a provider to carry that layer. If the engagement is short, or the work is genuinely well specified in advance, keeping the layer in-house is cheaper and more controllable.
Either answer is fine. What is not fine is choosing between them on the rate card alone, because the rate card is precisely the document from which this entire cost has been removed.