Emails and enquiries
Understands what the sender needs, extracts order numbers and client details, and routes the case to the right person or system.
AI automation for work that doesn't fit a template
Standard automation needs clear rules. AI automation handles what doesn't have them – emails, PDFs, free text and exceptions.
What an AI workflow looks like
The AI step is usually just one of several. The rules around it are what make the workflow reliable.
- 01
Something comes in
An email, an invoice, a form submission or a file in a folder triggers the workflow. It can also run on a schedule, such as every morning at seven, or respond to an event in another system, such as a new order or an updated case. We build the workflow in n8n, Power Automate, Make or custom code, depending on what you already use.
- 02
AI interprets the content
The language model reads the content and extracts what's needed in a fixed format: a client number, an amount, a case type or a short summary. This is where AI makes a difference, because the content doesn't follow a template. An email can be worded in a hundred different ways and still be about the same thing.
- 03
The rules take over
The workflow checks the details against your systems and puts them in the right place. Does the client exist? Does the amount match the order? This part is standard automation, with fixed rules that always produce the same result. AI interprets the content, but the rules determine what happens to the result.
- 04
A person handles exceptions
If the AI is uncertain or something stands out, the case goes to a person rather than moving on automatically. They can see how the AI interpreted it and approve, correct or reject it with a click. These exceptions also show where the workflow needs adjusting, so more cases can run automatically from start to finish.
Not everything needs AI
If a task follows fixed rules, standard automation is cheaper, faster and more reliable than AI. So that's what we build instead. AI is only needed when the content varies – free text, PDFs and emails worded in different ways.
- Extracting data from supplier invoices that arrive as PDFs in ten different formats/ AIThe format varies. AI reads the content; the rules check the amounts.
- Sending an order confirmation once payment is recorded/ AutomationSame event, same email. A standard workflow is enough.
- Summarising the week's client feedback by theme/ AIFree text from hundreds of clients. Only a language model can handle that.
- Pasting the same figures into two spreadsheets every month/ NothingKeep just one. The problem is duplicated work, not the tool.
n8n, Power Automate, Make or custom code
We choose tools to suit your environment, not our personal favourites. What you already have matters, because that's where the workflow will be maintained. The AI step can use ChatGPT, Claude or Gemini – whichever best suits the task and your agreements.
n8n
For workflows that span multiple systems and need to run on your own server. Includes built-in support for AI steps.
Power Automate
When everything is already in Microsoft 365. The workflow sits in the same environment as your emails, files and permissions.
Make
For simpler workflows between cloud services, where getting started quickly matters most.
Custom code
When performance, security or testability requirements are high, or when the workflow needs to become part of your own system.
Tasks where AI automation makes a difference
Documents and invoices
Extracts data from PDFs, invoices and delivery notes in different formats, then enters it into your accounting or business system.
Written content
Drafts replies, product copy and summaries in your tone of voice, ready for a person to review before they're sent or published.
Classification
Sorts cases, leads or client feedback by content and sentiment, so the most important items come first.
What happens when AI gets it wrong?
An error should never go unnoticed. That's why we build workflows that check the AI's interpretation before acting on it, and send anything uncertain to a person. That way, you can trust the workflow even when the AI occasionally gets it wrong.
01Responses in a fixed format
The AI step can't respond freely. It fills in predefined fields – client number, amount, date, category – so the rest of the workflow can check each detail separately.
- Predefined fields instead of free text
- Missing details are immediately visible
- The same format, regardless of model
02Checks against your systems
Does the client exist? Does the amount match the order? Does the date make sense? The workflow checks the AI's response against the data already in your systems before saving anything.
- Checks against your CRM and business systems
- Sense checks on amounts and dates
- Nothing is saved unless it passes the checks
03A queue for exceptions
Anything that fails a check goes into a queue for human review, with the AI's interpretation alongside the original. Approving or correcting it takes just a few seconds.
- The AI's interpretation alongside the original
- Approve or correct with a click
- Corrections improve the workflow
04Logging and improvement
Everything is logged, so we can see where the workflow runs into uncertainty and improve it over time. Often, a clearer instruction or an extra check is all it takes for more cases to run from start to finish.
- A log of every run
- Exception statistics
- Ongoing adjustments
What is AI automation?
AI automation means automated workflows where one or more steps use AI to read, interpret, assess or write. The rest of the workflow follows fixed rules. The AI step handles the content that varies – free text, PDFs and attachments in different formats – while the rules make sure the result ends up in the right place.
If the task already follows clear rules, there's no need for AI. Standard automation is enough, and it's cheaper and more reliable.
Frequently asked questions about AI automation
Can't find the answer? Book an AI review or call us on 070-953 53 99.
AI review
30 minutes, one of your workflows
We review a task that's taking up your time and give you an honest recommendation: AI, standard automation or no automation at all.
- A short written recommendation for next steps afterwards
- Meet the people who will build it
- Free of charge
