One Person, Every Function: What Actually Constrains a Small Business

The instinct when advising a one-person business is to scale down what a larger company does. That produces bad advice, because the constraint is different in kind rather than in degree.

A company with fifty people has specialists. Someone does the books, someone handles customers, someone writes the marketing, and each of them stays in one context for hours at a time. Their constraint is coordination: getting those functions to agree.

A one-person business has no coordination problem at all. Every function lives in one head, perfectly aligned by construction.

What it has instead is switching. In a single afternoon an owner might quote a job, chase an invoice, reply to a complaint, order stock, post something, and try to think about next quarter. Each switch carries a cost, and the cognitive research on task switching consistently finds a measurable penalty in both time and error rate when attention moves between unrelated tasks.

And there is a second, quieter difference. Nobody checks anything. In a company, work passes someone else on its way out. In a one-person business, whatever you produce ships, immediately, in the state you left it.

Those two features, constant switching and no reviewer, determine which tools help. Anything that reduces switching pays off. Anything that produces more output needing review is at best neutral, because the reviewing capacity is the thing that was already at zero.

Sorting where the hours actually go, because owners consistently misjudge this and optimise the wrong bucket.

Delivery. Doing the thing the business sells. Cutting hair, fixing boilers, writing code, cooking food, running the shop. This is the only bucket that directly earns, and for most small businesses it is where the owner wants to be.

Customer-facing overhead. Quotes, enquiries, scheduling, chasing, complaints, follow-ups. Large, fragmented, and interrupt-driven, which makes it the main source of switching.

Back office. Invoicing, expenses, records for the accountant, stock, compliance paperwork, supplier admin. Predictable, deferrable, and therefore usually deferred until it becomes urgent at the worst moment.

Growth. Marketing, pricing, deciding what to offer, working out whether the business is actually working. Important, never urgent, and the first thing sacrificed.

The usual pattern is that delivery and customer overhead consume the week, back office gets done badly at night, and growth gets done never.

Which sets the objective clearly. The point of tooling in a one-person business is not to do the delivery, which is what you are good at and what customers pay for. It is to compress the customer overhead and the back office enough that the growth bucket stops being empty.

That is a modest-sounding goal and it is the one that changes the trajectory of a small business, because a business that never gets to think about pricing is stuck at whatever price it started with.

Sorting the four buckets by whether a tool helps, and how.

Delivery sits on its own. This is the skill the customer is buying, and for most trades and services it is physical or relational or both. Tools help around the edges, and trying to automate the core is usually how a small business loses the thing that made it worth choosing.

Customer overhead is the highest-leverage target, for a reason that is about attention rather than volume. Its cost is not the ten minutes each enquiry takes. It is that each one arrives unannounced and pulls you out of what you were doing. Reducing the number of interruptions is worth more than reducing the time per interruption.

Back office is the second target, and it works differently. Nothing here is urgent, so the win is not speed. It is that the work actually gets done, on time, rather than accumulating into a bad weekend.

Growth is the beneficiary rather than the target. It is what the recovered hours are for, and this is where a model is genuinely useful as a thinking partner, because you have nobody to talk it through with.

The arrow to notice runs from the middle two into the last one.

This is the single most important structural fact about running a business alone, and it changes what safe adoption looks like.

In every professional setting, generated output passes a person before it matters. A junior's draft goes to a senior. A marketing email goes to a manager. A quote goes to whoever signs quotes. The review is social and automatic, and it catches things.

A solo owner has none of that. What you write goes out. What you calculate is what you charge. What you post is published.

And this interacts badly with a specific property of generated text: it reads as finished. A colleague's rough draft looks rough, which prompts scrutiny. Generated output arrives polished, which suppresses it. The confidence of the presentation is unrelated to the correctness of the content, and there is no second reader to notice.

So the review has to become structural rather than social. Three mechanisms that work in practice.

A delay on anything irreversible. Prices, public posts, contract terms and anything going to more than one person waits until the next morning. Most errors are obvious on a second look and invisible on the first.

A fixed check per category, written down, so you are not deciding what to check while tired.

And a hard rule that numbers come from the system that computes them, never from a model, because a wrong number is invisible in a way a wrong sentence is not.

None of that is sophisticated. It substitutes for a colleague you do not have.

There is a measurement problem here that deserves stating plainly, because a solo owner has no way around it and should know that.

A controlled study published by METR in July 2025 looked at experienced open-source developers working on their own repositories, with and without AI assistance. The developers expected to be about twenty-four percent faster with the tools, and afterwards believed they had been about twenty percent faster. Measured against the control condition, they were about nineteen percent slower.

That is one study, in one domain, with experienced people on codebases they knew deeply, and it should not be generalised into a claim that these tools never help. What it does establish, robustly, is that self-assessment of the speedup was wrong by roughly forty percentage points, in the favourable direction, by people well placed to judge.

Why that matters more for a small business than for anyone else. A company can run a comparison across teams. A solo owner has one person, no control group, and no baseline, so the only available evidence is the feeling of having been productive.

The practical response is not scepticism about the tools. It is to prefer changes whose benefit is structural rather than felt.

An automated response that fires at eleven at night either happened or did not. Bookkeeping that is current either is or is not. Those are checkable facts about the state of the business.

Whereas drafting felt faster is exactly the class of claim the study suggests you cannot assess about yourself.

A failure mode specific to small businesses, and expensive in a way that creeps up.

The pattern. An owner adopts a tool for scheduling, another for invoicing, another for social posts, another for the website chat, another for bookkeeping, each with an AI feature and each with a monthly subscription. Individually each looks trivially cheap. Collectively they become a meaningful fixed cost, and more importantly they become a system nobody has designed.

Three costs that follow.

The subscriptions themselves, which for a small business can quietly reach a level that would have bought something substantial.

The integration gap. Data lives in six places that do not talk, so the owner becomes the integration layer, copying between systems. That is switching cost, which was the constraint in the first place.

And the learning cost. Each tool has to be understood well enough to use safely, and a tool half-understood is where the errors come from.

The discipline that works. Prefer fewer tools that do more, even if each is individually worse at its job than a specialist. Prefer tools you already pay for, since your accounting software, your booking system and your email provider have probably added the capability you are about to buy separately. And before adding anything, name the specific recurring task it removes, in hours per week.

If you cannot name the task, the tool is being adopted because it is interesting, which is a legitimate reason for a hobby and not for a business.

One use is genuinely distinctive to running a business alone, and it is not about producing anything.

The hardest part of solo ownership, according to most people who do it, is not the workload. It is having nobody to think with. Every decision about pricing, about whether to take on a difficult customer, about whether to hire, about whether the business is actually working, gets made in your own head with your own assumptions unchallenged.

A model is a usable substitute for a specific, limited part of that. Describe a decision and ask what you have not considered. Ask it to argue the opposite case. Ask what a sceptical accountant would say about a plan. Ask what questions someone with more experience would ask you.

What this provides is not expertise. It is the friction of having to explain your reasoning to something outside your own head, which is most of what a good sounding board actually does. People routinely discover the flaw in a plan while explaining it, and the explaining is the mechanism.

What it does not provide, and should not be trusted for. Knowledge of your market, your customers or your numbers. Any judgement that depends on facts about your specific situation that it does not have. And the counterweight of someone whose own money is at stake.

So the framing that works. Use it to find the question you had not asked. Do not use it to answer the question, because the answer depends on things it does not know about your business.

An order that follows the constraints identified in this lesson rather than the marketing.

First, whatever reduces interruptions. An automated acknowledgement to enquiries so you are not answering at nine at night. A booking link so scheduling stops being a conversation. These do not save much time per event and they save a great deal of attention, which was the actual constraint.

Second, the back office that keeps slipping. Receipt capture, categorisation, invoice generation, the records your accountant asks for. The gain is not speed. It is that it happens on time instead of accumulating.

Third, drafting the things you write repeatedly. Quotes, follow-ups, standard replies, the descriptions you post. With a structural check, since nobody else will look at them.

Fourth, the thinking partner, for the decisions you would otherwise make alone and unchallenged.

And deliberately last, or never: anything customer-facing that runs without you. A chatbot that answers questions about your work, an agent that quotes, automated posting that goes out unread. Not because these cannot work, but because they fail in front of customers and you will not be there when they do.

The pattern underneath, worth carrying into the next lesson. Start where a mistake is invisible to customers and reversible by you. Move outward only once the checking habit exists. Most owners do the reverse, because the customer-facing tools are the ones being advertised.