Five Automations Worth Building First (and Three That Aren't)

You have a list of forty things you could automate and no way to tell which one pays for itself.

Everything on the list looks plausible. Everything has a template somewhere. And every article about it is a longer list, which is precisely the wrong medicine, because the problem was never a shortage of ideas.

So this is a short list. Five worth building, three worth leaving alone, and the reasoning underneath so you can judge the forty-first yourself.

What makes one worth building.

Three things have to be true at once, and most people only check the second.

It runs often. Daily beats weekly beats monthly, by a much wider margin than people expect, because the cost of building is paid once and the return arrives every time it fires.

It eats attention rather than time. This is the one that gets missed. A task that takes twenty minutes of concentration costs you more than one that takes forty minutes of half-watching a progress bar, because concentration is the thing you have least of and cannot buy more of.

It fails visibly. Pick something where a mistake announces itself. Your first automation is also how you learn what these things do wrong, and you want that lesson delivered loudly and early, not discovered in a spreadsheet in November.

And one honest note before the list, because it is the thing nobody says. Automation does not make money. It returns hours. Those hours only become money if you have something to sell and were short of time to sell it. If you have hours and no offer, automating your admin will give you more hours and no offer.

The five.

01 · Sorting the incoming post. Messages arrive, the system works out what each one is actually asking for, routes it to the right place, and drafts replies to the repetitive ones. It runs every day without being asked, it takes back the attention tax of opening things to find out what they are, and when it misfiles something you know within an hour. This is the best first build for almost everybody.

02 · Meetings into action items. A transcript or a set of notes goes in. What comes out is a short summary plus a list of who owes what by when, filed wherever you keep tasks. The value is not the summary. It is that the commitments stop living in somebody's memory of Tuesday.

03 · New enquiry, researched and drafted. An enquiry arrives. The system gathers what is publicly known about the company, compares it with what your good customers look like, and puts a first-draft reply in front of you with the useful facts attached. You are not automating the selling. You are removing the twenty minutes of looking things up that stands between the enquiry and your reply.

04 · Money landing in your records. An invoice is raised or a payment goes through, and the customer record, the deal and the sheet all update themselves. It is the dullest thing on this list and quite possibly the highest value, because errors here are expensive in a way that errors in a draft email are not, and nobody has ever enjoyed doing it by hand.

05 · The unfinished checkout. Only if you sell online. Somebody fills a basket and leaves, and a written, specific, not-desperate follow-up goes out, with the result recorded where you can see whether it worked. It is narrower than the other four and it earns its place because the line between building it and seeing an effect is the shortest on the list.

The free AI Agents pack. Over two thousand ready-made n8n workflows, which is where I would start for any of these five rather than from a blank canvas. Plus a workbook on the twenty building blocks that cover most of them, the Agent-Plus Prompt Builder for writing agent instructions, and the checklists I run through before letting anything work on its own. Download the AI Agents pack →

The three that are not worth it.

These are categories rather than particular templates, because the categories are what keep tempting you back.

Anything that produces content nobody asked for. Scheduled generators that write posts, make videos, assemble digests. They work, technically. The output is real and the audience is imaginary, and you will spend more time reviewing what it made than you ever spent not having it. If nobody is currently waiting for this thing, automating its production does not create demand for it.

Anything replacing a task you do once a month for five minutes. Six hours of building and permanent maintenance, to save an hour a year. The arithmetic never works, and it feels productive the whole time you are doing it, which is what makes it dangerous rather than merely useless.

Anything customer-facing before you have a place to say no. Not because it is a bad idea, but because it is the right idea in the wrong order. Build the version where you approve the message first, run it that way for a fortnight, then decide what may go out unattended. That order costs you two weeks. The other order costs you an apology email to your entire list.

Be honest about the effort.

Each of these five is roughly an evening to build and a second evening to make it survive real data. That second evening is not optional and it is not in the tutorials, because tutorials use clean examples. Yours will meet a contact with no surname, a payment in the wrong currency, and a message that is three messages forwarded together.

There is no version of this that is a weekend and then income. There is a version that is two evenings and then one recurring irritation gone permanently, which is a smaller promise and an actual one.

What the five will not do.

They will not make you look bigger than you are. Customers can tell when a reply was assembled rather than written. The five above are chosen so that the machine does the sorting and the fetching, and you keep the sentences that a person reads.

They will not fix a process you cannot describe. If you cannot write down how you currently decide which enquiry matters, automating it will make that inconsistency faster and give it an air of authority.

They will not run themselves forever. Connections expire and other people's software changes underneath yours. Budget an hour a month for looking at the things you built, or accept that they will stop and you will find out from a customer.

And none of them is an agent in the strict sense, which is worth knowing before somebody sells you one. Four of the five are ordinary automation with some judgement bolted in at one step, and that distinction changes what you should pay for it.

One small thing to do before the next post.

Pick one of the five. Write down its trigger in a single sentence, starting with the word when.

When an email arrives in the shared inbox. When a payment succeeds. When somebody submits the contact form.

If you cannot finish that sentence, you have found the real problem, and it is not a technical one. Everything downstream of a trigger is assembly. The trigger is the decision.

If you want the whole thing in one place.

If you would rather have the complete system than one piece at a time, that is what I wrote AI Agents Bible for. Five books in one, with Custom GPTs, Make.com and n8n each built step by step, thirty worked examples you can copy, and the parts most guides leave out, about cost, reliability and knowing what not to let run on its own.

Both this and the prompt engineering book also come as one 12-in-1 volume, if you would rather have the whole system in a single place.