Batch, Not Continuous

07 Sep, 2026

Batch, Not Continuous

The vat does not care about your quarter

There is a moment in every brewery tour when someone in the group, usually wearing a lanyard from a conference they are avoiding, asks the guide whether the fermentation could be sped up. The guide always smiles. It is a patient, well-rehearsed, slightly sad smile, because the guide has heard this question roughly nine thousand times and knows exactly what it means. It means I have been trained to believe that all durations are negotiable.

They are not. Lager conditions cold for weeks because that is how long the yeast takes to clean up after itself. You cannot halve it by hiring a second brewer. You cannot halve it by hiring twelve. You cannot escalate to the yeast. There is no stakeholder alignment session that produces a faster ester profile. The tank contains a biological process with an irreducible duration, and the only thing your urgency changes is your mood.

Corporate planning, meanwhile, is built almost entirely on the opposite assumption - that work is a fluid, that fluids flow at rates, and that rates respond to pressure. Add people, add hours, add attention, add a war room, and the stream moves faster. This model is not stupid. For a large amount of real work it is correct. The problem is that we apply it universally, to everything, including the things that are shaped like a tank rather than a pipe. And then we act surprised, every single quarter, forever.

Two shapes of work

Almost all work sorts into one of two shapes, and the shape determines what management can actually do.

Stream-shaped work has a rate. Reviewing a thousand contracts, migrating ten thousand records, answering a support queue, writing tests, tagging data. Throughput scales, imperfectly and with overhead, when you add capacity. If you double the people you get something meaningfully better than nothing. The finish date is a function of resourcing, which is why resourcing conversations feel productive.

Batch-shaped work has a duration. A thirty day reliability soak test takes thirty days. A three month audit observation window takes three months, and no volume of weekend heroics will retroactively generate evidence that a control was operating in April. A new hire becomes genuinely useful somewhere between month four and month nine, depending on the codebase and how many people interrupt them. Trust between two teams that recently blamed each other in writing accrues at the speed of repeated behaviour, which is a phrase that has never once fit inside a sprint.

The distinction is not about difficulty or importance. It is about where the clock lives. In stream work, the clock is inside your organisation and you can influence it. In batch work, the clock is outside your organisation, in biology or law or physics or human memory, and your only remaining decision is when to start.

The four shapes, since two was a simplification

Real portfolios contain more than two species, and conflating them is where the forecasts go to die.

The last two are the interesting ones because they are the two that look identical in a planning tool and behave nothing alike in reality.

Maturation is fermentation proper. Concrete curing. Whisky. Cheese. A brand becoming trusted. A reorganisation stopping being the only thing anyone talks about. The work happens during the wait, but not by anyone you employ.

Accrual is waiting for the world to supply raw material. An experiment reaching statistical significance needs users to show up. A case study needs a customer to actually achieve the result you promised. A hiring pipeline needs candidates to exist. Accrual is partially elastic, which is precisely why people mistake maturation for it and start proposing solutions. You can buy traffic to close an experiment sooner. You cannot buy time to make the cement set. Knowing which one you are staring at is worth more than any velocity chart ever produced.

The doubling test

There is a diagnostic that takes eleven seconds and that almost nobody performs before committing to a date.

Ask two questions. First - if I doubled the team tomorrow, what happens to the finish date? Second - during the waiting period, who exactly is doing the work?

If the honest answer to the second question is "nobody, the world is doing it," you are holding a batch, and every subsequent conversation about resourcing is theatre. Pleasant theatre. Well attended. Utterly non-causal.

The reason this test is skipped is not ignorance. It is that the answer is unwelcome. "We can go faster if we prioritise it" is a sentence that gets you a promotion. "The duration is fixed and the only lever was the start date, which was four months ago" is a sentence that gets you gently excluded from the leadership offsite.

Green beer

Here is the part that planning systems get most catastrophically wrong. They assume value accrues roughly linearly with elapsed time, so that shipping at eighty percent of the duration yields something like eighty percent of the value. Under that assumption, cutting the timeline is a trade. Slightly worse product, considerably earlier, net positive, ship it.

Batch work does not work like that. A lager pulled early is not eighty percent of a lager. It is green beer - buttery, sulfurous, tasting faintly of creamed corn and disappointment. It is not a lesser version of the product. It is a different product, one that damages the relationship with whoever drinks it. English handed us the phrase half-baked and we decided to use it only as an insult rather than as a warning about the shape of value curves.

The corporate versions are everywhere once you can see them. The security certification pursued on a compressed timeline, producing a document that technically exists and convinces no serious buyer. The onboarding compressed from three months to three weeks, producing an engineer who can close tickets and cannot tell you why the system is built that way, which is a liability that matures for years. The culture change programme with a twelve week timeline, which is not a shorter culture change but rather an expensive new source of cynicism. The reference customer approached before they got results, converting a future advocate into a mildly irritated logo.

Shipping a batch early does not harvest less value. It frequently harvests negative value, while also consuming the input, which was the only one you had.

What actually breaks the forecast

The deep failure is not that batch work makes forecasts optimistic. It is that batch and stream work have inverted variance structures, and blending them makes your aggregate uncertainty not merely large but the wrong shape.

Stream work has high variance in duration and low variance in quality. You do not know if the migration will take five weeks or eleven, but you are fairly confident the records will end up in the right place. It is late in ways you can see coming, in ways that show up in a burndown, in ways a project manager can escalate.

Batch work is the mirror image. Low variance in duration, high variance in quality. The whisky will be ready in twelve years, with remarkable precision. Whether it is any good depends on decisions made before the cask was sealed, and you will not find out until the end. Batch work is not late. Batch work is early, in ways you can taste.

So when you average these two together in a single portfolio model with a single confidence interval, you produce a number that is meaningless in a specific and dangerous way. Your batch items look reliable, because their dates do not slip. Their dates do not slip because the durations were never elastic, and because when reality pressed on them, the thing that gave way was quality, silently, off-ledger, in a place your dashboard cannot render. You then reward the batch teams for predictability and interrogate the stream teams for slippage, which is roughly the opposite of the correct managerial response, and you do it with total confidence, quarter after quarter, because the chart is green.

Stirring the fermenter

Every fermenter has a lid, and the lid is not decorative. Open it to check on progress and you introduce oxygen, and oxygen introduces oxidation, and oxidation introduces cardboard. The act of observation degrades the product.

Corporate batch work has exactly the same property, and we have institutionalised lid-opening as a management ritual. The daily check-in on the thing that takes six weeks regardless. The request for a status update on the audit observation window, which is going exactly as fast as a calendar. The new hire pulled into a critical delivery in week three because they seemed underused, thereby destroying the very maturation period that was going to make them valuable. The reorganisation that needed eight quiet months to settle, subjected instead to a pulse survey every fortnight, each one reminding everyone that they are unsettled.

Some processes are damaged specifically by being watched. This is not an argument against measurement in general. It is an argument that measurement has a cost, that the cost is highest during maturation, and that "let us just get a quick update" is not free merely because it is short.

Idleness is a shape, not a sin

The reason batch work gets abused is not intellectual. It is aesthetic. Batch work looks like nothing is happening, and organisations are optimised to punish the appearance of nothing happening.

A tank sitting quietly in a cold room has no story, no update, no slide. It generates no evidence of effort, and effort is the currency most organisations actually pay in, whatever the strategy deck claims about outcomes. So the manager, who is not malicious but is very much observed, finds something to do to it. Stirs it. Warms it. Moves it to a different room where it can be seen. Adds a second brewer for optics. Reports progress.

This is how a company can be full of intelligent people who all individually understand that trust takes time, and yet institutionally behave as though it takes a workshop.

Buy more tanks

Here is the good news, and it is genuinely good, because the alternative to speed is not surrender.

Breweries have known the answer for centuries. You do not compress fermentation. You buy more fermenters and you stagger the start dates. Capacity in batch work is measured in vessels and calendars, not in headcount and heroics. The relevant skill is not hustle. It is foresight - the wholly unglamorous discipline of starting things before anyone is asking for them.

Which means the practical moves are these.

• Start earlier, since it is the only lever you have. For maturation work, the start date is the entire decision. Everything after it is weather.
• Run batches in parallel, not faster. Three certifications begun in staggered months, three onboarding cohorts, three reference customers cultivated at once. Pipeline the waiting.
• Front-load all rigor to inoculation. Once the lid is on, your leverage is approximately zero, so every ounce of quality control belongs before the seal. This is exactly opposite to the iterate-as-we-go instinct, and it is why batch work punishes teams that are excellent at mid-course correction and rewards teams that are excellent at preparation.
• Report batch work on a different instrument. Not percent complete, which is a lie for anything maturing. Report start date, expected ready date, and the inputs that were locked at the beginning. Then leave it alone.
• Protect the wait explicitly, in writing. Unprotected waiting is always reallocated to whoever shouts. Give the maturation period a name and an owner whose job is to defend it.

Not all waiting is wisdom

An honest caveat, because this metaphor is unusually easy to weaponise, and someone in your organisation is already drafting the email.

A contract sitting in a general counsel's inbox for three weeks is not aging in oak. A design decision that nobody will make is not fermenting. A queue is not a cask. The overwhelming majority of corporate waiting is not irreducible duration at all - it is a stream-shaped process with a resourcing problem, wearing batch clothing as an alibi. Queue time masquerading as maturation is probably the single most common form of institutional dishonesty available to a competent middle manager, and it is available cheaply, at scale, with no evidence required.

So the test cuts both ways. During the wait, what is happening, and to what? If something is genuinely transforming - yeast working, concrete crystallising, evidence accumulating, a person building a mental model of a system, a market slowly forming an opinion of you - it is a batch, and you should defend it. If the answer is that a document is sitting still, waiting for a human being to have a thought, then you do not have a fermentation. You have a bottleneck, and it has a name, and the name is probably in the meeting.

The calendar is the only honest executive

The deepest reason we model everything as continuous is that continuous work flatters us. It says that outcomes track effort, that will translates into speed, that a sufficiently committed organisation can bend a timeline. It is a theory of the world in which we are the protagonist.

Batch work is a quiet correction. It says that certain things happen at their own pace, that our role is to prepare well and then get out of the way, and that the most valuable managerial act available is often to leave something completely alone for a period that will feel, to everyone watching, like negligence.

The org chart cannot compress time. Not with escalation, not with visibility, not with a cross-functional tiger team, not with a renamed initiative and a fresh set of key results. The whisky will be ready in twelve years. You may of course announce it for next Thursday. The tank will not read the announcement, and the people who taste the result will not care what the plan said.