You Can’t Refactor a Budget
Batch size is a cached estimate of what a decision used to cost. Nobody has refreshed the cache.
Your quarterly planning cadence is not an engineering practice. It is an amortization schedule for decisions that used to be expensive to make.
Approving capital took months of analysis, so companies approve capital annually. Getting forty teams to agree on sequencing took a two day event, so companies sequence quarterly. In both cases the interval was set by the cost of producing the decision. In both cases that cost is now falling faster than anyone’s calendar.
This is the reason AI-accelerated delivery keeps underwhelming at enterprise scale. Teams do get faster at building. Nothing gets faster at deciding what to build, or at releasing the money to build it. The work piles up behind a gate that opens four times a year.
My last piece argued that context, not labor, is the scarce input to modern delivery. This one is about what sits downstream of that. When the labor constraint loosens, the funding constraint becomes the binding one, and nobody in the organization owns it.
The arithmetic nobody runs
Flow efficiency is the share of an item’s elapsed time during which someone is actually working on it. In large enterprises, on the rare occasions anyone measures it, the number tends to land between ten and twenty percent. The rest is queue. Waiting for review, waiting for a dependency, waiting for a decision, waiting for the next planning event.
Sit with what that implies. If active work is fifteen percent of cycle time, and you make active work twice as fast, total delivery time improves by seven and a half percent. You have doubled the throughput of your engineers and moved the number by less than a rounding error on the quarter.
Amdahl worked this out for processors in 1967. Speeding up one component yields returns bounded by that component’s share of the whole. Org charts obey the same math. The difference is that when you optimize a processor, the next bottleneck is another piece of silicon. When you optimize a delivery team, the next bottleneck is a governance forum with a standing calendar invite.
What PI planning is actually for
SAFe takes a beating in engineering circles. Most of it is deserved on the merits and almost none of it is deserved on the diagnosis. The standard complaint is that PI planning is theater. That complaint is both accurate and beside the point, because PI planning was never primarily an engineering ritual.
Look at where the framework spends its structure. There is a portfolio layer. There is an entire discipline called Lean Portfolio Management. A methodology concerned only with how software gets built would need neither. What SAFe actually contributed was a legible unit of work that finance could fund, and a single scheduled moment when many teams could negotiate sequencing with each other. It solved a capital allocation problem and a cross-boundary coordination problem, then dressed both as an engineering practice so that engineers would show up.
That accounts for the thing everyone notices and nobody explains. Engineers experience PI planning as overhead. Executives experience it as indispensable. Both are describing the same event correctly, because they are using it for different jobs. This is why the argument has run for a decade without resolving, and why every attempt to settle it inside the engineering organization fails. The argument is not located there.
Batch size is a cached estimate
Once the pattern is visible in planning increments, it shows up everywhere. Every cadence in a company is a stored guess about what a decision costs to produce.
Annual budgets assume that building a credible case takes a quarter. Stage gates assume that assembling enough evidence to proceed takes weeks. Quarterly increments assume that aligning many teams requires physically assembling them. A statement of work assumes that scoping and pricing a body of work takes weeks of discovery.
None of these get re-derived. They are set once, usually by someone who has since left, and then they persist, because they live in the calendar and the calendar does not present itself as an assumption. It presents itself as weather.
The second collapse
Here is what makes this urgent rather than merely true. The inputs to a funding decision are getting cheaper at the same time as the work itself.
A business case used to require weeks of human effort. Someone gathered comparables, built a cost model, wrote the options analysis, socialized a draft, absorbed feedback, revised, socialized again. That labor is a real part of why enterprises allocate capital annually. Justifying things was genuinely expensive.
Much of that production cost is now collapsing. Not the judgment. The production. The synthesis, the modeling, the drafting, the option generation, the impact analysis against systems that already exist. If a credible funding case can be assembled in two days instead of six weeks, annual capital allocation stops being a constraint and becomes a preference.
It might be a defensible preference. But a preference has to be defended, and at the moment nobody is being asked to defend it.
The batch I am standing in
Consulting sells in batches, and I would rather say that plainly than ask anyone else to examine theirs first.
A statement of work fixes scope across a multi-month period, priced from an estimate produced by a process whose cost is falling fast. The commercial construct assumes that scoping is expensive, that changing scope mid-flight is expensive, and that a client’s tolerance for ambiguity is lower than their tolerance for a change order. Those assumptions are load-bearing, and at least one of them has already stopped being true.
I went digging through my own firm’s records at a client recently and found a signed change order with a planning increment number in the title. We had not merely adapted to the client’s cadence. We had adopted it as a unit of commerce. Our billing was shaped by their calendar, and their calendar was shaped by a decision cost from over a decade ago.
That is not a knock on the team. It is what happens when a batch size goes unexamined long enough that it stops looking like a choice.
What is genuinely irreducible
Two objections deserve real answers.
The first is that some batching has nothing to do with decision cost. Public companies report quarterly. Capital markets pay for predictability. A CFO who cannot forecast has a governance problem, not a courage problem. You cannot roll planning past the point where numbers can be committed to the street, and that constraint is structural.
So the honest version of my claim is narrower than the slogan. The irreducible portion of batching is much smaller than what most enterprises currently batch, and almost nobody has measured where the line falls. Finding it is the interesting work. Denying it exists is not.
The second objection is that frequent small allocations produce thrash. Fund quarterly and teams chase the quarter. Fund weekly and teams chase the week, and the platform that pays back in three years never gets built. Some commitments have to outlive the enthusiasm that produced them.
That objection resolves rather than refutes, because not every decision deserves the same funding frequency. Consequence and reversibility are the two variables that matter, and I have used them as a two by two in delivery governance for years. Low consequence and easy to reverse: let the team decide, get out of the way. High consequence and hard to reverse: bring the full apparatus, because that is what the apparatus is for.
The same two axes work on money with no modification. Continuous allocation for reversible bets. Slow, deliberate, defended allocation for the ones you cannot walk back. What most enterprises do instead is run one funding cadence across both, which overserves the reversible work, underserves the irreversible work, and then blames the framework for the result.
Three things that do not require a transformation program
Measure decision latency the way you measure lead time. You almost certainly track how long work takes. You almost certainly do not track how long approval takes. Which means the largest queue in your system is the only queue with no instrument on it. Start there, because there is no useful argument about batch size without a number.
Name the person who can re-order the queue on a Tuesday. Rolling planning is not a process change. It is a delegation of authority wearing a process costume. If prioritization only happens when senior people are in a room together, removing the room will not produce continuous flow. It will produce whoever escalates loudest, which is worse than a stale plan because it is invisible.
Separate the gate from the calendar. This is the part SAFe got right and most of its critics throw out along with everything else. A genuine go/no-go decision is valuable. A two day event on a fixed quarterly boundary is one delivery mechanism for that decision, and it is not the only one. Keep the decision. Drop the interval. Then argue about what the interval should be, with data, out loud, where people can see the reasoning.
The estimate has expired
Nobody at your company will stand up and defend the quarterly cadence, because nobody at your company chose it. It arrived. It was reasonable when it arrived, priced against a world where building was slow and deciding was slow, and it held those two things in rough balance.
Building got fast. Deciding did not.
The organizations that pull ahead over the next few years will not be the ones that picked the better framework. They will be the ones that opened the calendar, found a cost estimate sitting inside it, and noticed the estimate had expired.