A few years ago, automating a business process meant buying an enterprise platform, hiring a consultant, and waiting six months. The bill was usually six figures before anything ran in production. That era is over. The tools your team can use today cost almost nothing to run, and the people best placed to build them are the ones already doing the work.

Why the cost collapsed

Three things happened at once. Workflow builders like n8n moved to self-hostable, open-source models, so the licence fee dropped to zero. Cloud compute got cheap enough that a small firm can run a workflow server for a few pounds a month. And AI inference became pay-per-use, meaning you only pay when the automation actually does something.

The result: the infrastructure stack that once cost £200k in platform licences now costs close to nothing. A workflow server, a Google Sheet acting as your queue and database, and a handful of AI API calls replace what used to require a dedicated software team.

The people who should be building these things

Heres the shift that matters more than the price drop. The person who knows a process best is the person who does it every day. They know where the information arrives (an email attachment, a WhatsApp message, a supplier portal). They know the exceptions (the PDF that comes in a different format on Fridays). They know what “done” looks like.

Traditional automation projects pulled that knowledge out of the ops team, handed it to a developer who had never seen the process, and produced something that broke the moment an edge case appeared. No-code agent tools flip that. The ops manager, the logistics coordinator, the office manager, they can now describe the process in plain language and wire it up themselves.

That is not a small thing. It means the automation reflects reality, and the person who built it knows how to fix it when something changes.

What your team could realistically build in a week

Start with read-only or low-stakes tasks. These are the ones where a mistake costs you a minute, not a customer.

  • Data entry from emails and PDFs. An AI model reads an incoming order confirmation or delivery note, pulls out the key fields, and writes a row to a Google Sheet. No one types it in. No one forgets.
  • Status chasing. A workflow checks a list of open jobs each morning, identifies anything overdue, and sends an internal Slack or Teams message to the owner. The human still decides what to do next.
  • Report assembly. Pull numbers from a spreadsheet, format them into a summary, send it to the right person at 8am on Monday. No one has to remember to do it.

None of these touch a customer directly. None of them spend money. If they produce a wrong answer, a human catches it before it matters. That is exactly where you want to start.

What to leave to a specialist

Some processes look simple and arent. If an automation can send an email to a customer, raise an invoice, modify a booking, or trigger a payment, the failure modes are expensive and sometimes irreversible. These need guardrails that take more than a week to get right.

Specifically: idempotency (making sure the automation doesnt run twice and do the same thing twice), spend caps on AI calls so a runaway workflow doesnt burn through your budget overnight, and human approval steps on anything that costs money or touches a customer at volume. We wrote about one £700 mistake that came from skipping those guardrails - the article is worth reading before you wire anything to a payment or comms system: The £700 Automation Mistake.

How to set up your first safe automation

Pick one process. Make it internal. Make it read-only or human-approved at the output stage. A simple n8n workflow for email-to-spreadsheet extraction looks roughly like this in structure:

Trigger: Email received (IMAP or Gmail node)
  |
  v
Extract: AI text node with prompt
  “Extract: supplier name, order number, delivery date, total value.
   Return JSON only. If a field is missing, return null.”
  |
  v
Write: Google Sheets node, append row
  |
  v
Notify: Slack message to #ops-team
  “New order logged - please confirm before processing.” // human approval step

The Slack message is not decoration. It is the guardrail. Until you have run this workflow for two or three weeks and trust the extraction, a human signs off before anything downstream happens.

Once that is stable, you hand the workflow to the team member who owns the process. They can edit the prompt, add a new column to the sheet, adjust the Slack message. The keys are theirs.

What BBL typically does

We build the first automation with you, not for you. We set up the infrastructure, wire the first workflow, add the spend caps and approval steps, and then show the team how to extend it. The goal is that you dont need us for the next one. Some clients come back when a process gets complex enough to need AI judgment rather than just extraction - that is when a specialist saves more time than they cost.

If you want to talk through which process in your business is the right first candidate, that conversation is free and takes about twenty minutes.

See how we work and what a first build looks like →