Automations for Small Businesses

Eighty-nine per cent of small businesses now use AI in some form, up from 36% in 2023. And 73% of the ones using it say they lack the training and resources to do it properly.

That gap is the whole opportunity, and it is a much better one than it sounds, because it is not a technology problem. The tools are cheap and mostly point-and-click. What is missing is somebody who will sit down, work out which twenty minutes of a plumber’s day is being wasted on retyping enquiries, and wire it up.

Sixty-two per cent of small and medium businesses say they plan to hire AI specialists as freelancers within three months. Seventy-one per cent plan to increase freelance hiring generally. They have budgets and no idea who to call.

What the work is, in concrete terms#

Forget “AI consultant”. The jobs people actually pay for are small and dull and specific:

• An enquiry arrives by email or web form and it currently gets copied by hand into a spreadsheet. Now it goes into the CRM automatically, gets categorised, and pings the right person on WhatsApp.
• A dental practice gets forty voicemails a week. Now they get transcribed, summarised in one line, and sorted into “book them”, “call back”, “ignore”.
• An accountant receives receipts as photos. Now they get read, the numbers extracted, and dropped into the right sheet.
• A shop answers the same eleven questions on Instagram every day. Now a bot answers nine of them and escalates the other two.
• A recruiter reads 200 CVs a week. Now they get summarised against the actual brief, with the reasoning shown.
• A charity has fifteen years of documents nobody can search. Now they can ask it questions.

None of that is impressive to a technical person. All of it saves a small business real hours, and small businesses report saving 20+ hours a month and $500–2,000 in costs when this is done properly, which is why they will pay four figures for a build that takes you two days.

What it pays#

One detail worth knowing before you choose a tool: self-hosted n8n work reportedly pays 40–60% more than the equivalent Zapier project, because the client ends up owning the stack rather than renting it. Learning the self-hosted option is the single highest-return hour you can spend here.

Upwork for the hourly range; project and retainer figures are practitioner guides. Sources: Upwork’s own n8n hiring page.

Why retainers matter more than builds#

The one-off build is the obvious business and the worse one. Every serious operator in this space now pairs a setup fee with a monthly retainer, and the reason is not greed, it is that these automations break.

An API changes its response format. A platform deprecates an endpoint. The client renames a column in their spreadsheet. Something that worked perfectly for four months silently stops, and nobody notices until a customer complains that nobody called them back.

So the pitch to a client is not “I will build you this”. It is “I will build you this and keep it working”, and the second half is what they actually need. It also converts your income from project-hunting into something predictable, which is the difference between this being stressful and being a business.

Say it plainly in the quote: “$2,400 to build it, then $400 a month to monitor it and fix it when something upstream changes. If you’d rather not have the retainer that’s fine, but when it breaks it will be a new project at my hourly rate.”

What it costs to start#

Genuinely almost nothing, which is unusual in this section.

n8n can be self-hosted free. Make and Zapier have free tiers with usage limits that are enough to learn on and enough for small client workflows. An LLM subscription is $20 a month. Total realistic outlay to become competent: the price of one month of one subscription, plus the hours.

The costs that do arrive are the client’s, not yours, and you should say so up front so it is never a surprise: the automation platform’s paid tier if volume needs it, plus API usage. Quote your build fee separately from their running costs.

Finding the first client#

The mistake is going after businesses that already know what AI automation is. Those have agencies. Go to the ones that do not know the phrase and do know their own problem.

Start with businesses you already understand. If you worked in hospitality, you know exactly which part of a restaurant’s admin is miserable. That knowledge is worth more than technical skill here, because the hard part is diagnosis, not building.

Lead with the problem, not the technology. Nobody wants an automation. They want to stop retyping enquiries.

An approach that works, adapted rather than copied:

Subject: the enquiry form on your site

Hi. I filled in the contact form on your site on Tuesday to see what happens, and got a reply about six hours later, which is quick for a small firm.

I build small automations for trades businesses. The one that would probably help you most: when someone fills that form in, it lands in your CRM already categorised, and whoever is on call gets it on WhatsApp within a minute. Most of the six hours is someone noticing the email exists.

I can build it for £1,200 and it takes about three days. Happy to show you a working version of the same thing I built for a plumbing firm before you decide anything.

The blank version. Note that the first line costs you five minutes and is the reason the rest gets read:

Subject: [the specific thing on their site you used]

Hi. I [used their form / rang the number / booked online]
on [day] to see what happens, and [exactly what happened,
with the timing].

I build small automations for [their kind of business].
The one that would probably help you most: when [the
trigger], it [the outcome], and [who gets told, how fast].
Most of [the delay you observed] is someone noticing the
[email/message] exists.

I can build it for [price] and it takes about [n] days.
Happy to show you a working version of the same thing I
built for [a comparable business] before you decide
anything.

Everything in that email is checkable by them in under a minute, which is the property that separates it from the automation pitches they already delete. You are not describing a capability; you are describing their Tuesday.

The parts doing the work: you tested their actual process, you named a specific gap, you priced it, and you offered proof rather than a call. Ten of those, sent individually to businesses in one trade, is a realistic first client.

The other route is warmer and faster. Tell everyone you know what you now do, specifically. This work spreads by referral more than any other route on the site, because a business owner who saved ten hours a week tells the other business owners in their trade association. The first client is hard and the fourth arrives unprompted.

What to learn, in order#

1. One automation platform properly. n8n if you are willing to self-host, Make if you want it easier. Not all three.
2. Connecting an LLM into a workflow, summarising, categorising, extracting fields from messy text. This is the bit that turns a 2015-era automation into something that handles real-world input.
3. The specific tools your clients already use. Whatever CRM, spreadsheet and messaging their trade actually runs on. Being the person who knows the tools in one industry beats being generally capable.
4. Document search over a client’s own files. The highest-value thing you can build for anyone with a filing cabinet’s worth of PDFs, and it demos extremely well.
5. Voice agents, last. Answering-service and appointment-booking agents built on the current voice platforms are the most impressive demo in this field and the most likely to embarrass you in production. Learn them once the boring stuff is reliable.

Where this goes wrong#

Building something clever the client cannot run without you. If it breaks and they cannot even describe what broke, you have created a dependency they will resent. Write a one-page plain-English explanation of what it does and hand it over.

No retainer. Covered above, and it is the difference between a business and a series of one-offs.

Charging by the hour once you are fast. The same workflow takes you two days in month one and four hours in month six. Price the outcome. This is the same trap as in selling a skill and it costs people the same money.

Holding the client’s credentials casually. You will end up with access to a business’s email, CRM and customer data. Use their accounts rather than your own where possible, hand over every credential at the end of a build, and never store client API keys in a spreadsheet. If you are working with anything personal, understand that data protection obligations exist and are not optional, see protecting your income for the general shape and get proper advice if you are handling anything sensitive.

Promising to automate something that should not be automated. Turning off the human on a process where a mistake harms a customer is how you lose the client and deserve to.

The adjacent opportunity almost nobody has taken#

Forty-five per cent of consumers now use AI to find local businesses, up from 6% in 2025. And 88% of local businesses have no strategy at all for appearing in AI answers.

That is an unserved market sitting next to this one: helping a local business be the one an assistant names when someone asks for a plumber in their town. It is early enough that nobody can honestly promise results, see what is actually known about AI visibility, which is less than the people selling it claim but it is the most obvious gap of anything in this section, and the businesses already know they have it.

Who this suits#

Practical people who like fixing things and do not need to be on camera. People who have worked in an ordinary industry and know where its time goes. People who want client income with almost no startup cost and no audience.

People who want the work to compound: unlike data work, this builds a client base, a set of reusable templates, and a reputation in a niche.

Who it does not#

Anyone who will not do sales. This is a business with clients, and the first three are found by asking.

Anyone who wants passive income. Retainers are recurring, not passive, something breaks and you fix it.

Anyone unwilling to learn one tool properly. Surface familiarity with five platforms is worth less than real command of one, and clients can tell the difference within a week.

The first month#

1. Pick one platform. Build three automations for yourself, something you actually do by hand every week.
2. Pick one industry you already understand.
3. Build one polished demo of the most common problem in that industry, using fake data. This is your entire portfolio.
4. Write to fifteen businesses in that industry individually, having actually looked at their process, with a price.
5. Take the first job slightly under your intended rate, deliver early, and write the plain-English handover document.
6. Propose the retainer on delivery, not later.
7. Ask that client who else in their trade has the same problem. This is the question that produces client two and three.