Most businesses reach for AI because it feels more advanced. But it's often like using a katana to cut a watermelon. It looks impressive and it's completely overkill.
What gets missed is that a simple automation is usually cheaper, cleaner and far more dependable. The trick is knowing when to use which, without creating unnecessary mess.
Here's how to think about it.
What AI actually is
In the context of working with it inside your business, there are two ways to picture AI.
The first is the chat experience. You have a text box, you enter a question, maybe which headphones to buy for an upcoming trip, and it thinks through the question, does some research, pulls the information together and gives you an answer back.
That flexibility is the point. AI handles ambiguity and interpretation.
What automation is
Automation is usually a fixed sequence of events that you can define clearly. It's defined by a trigger, maybe some conditions or filters, and then an action that happens as a result.
For example: when a new newsletter form submission is received, a filter checks it isn't one of your own test submissions, then it looks for the person and creates a lead for them.
Because it replaces manual steps you'd otherwise do yourself, it frees you up, either to build more automations or to get out into nature and touch grass.
Automation with AI built in
Automation and AI aren't either/or. You can put an AI step inside an automation and get both: the defined, reliable process of an automation, plus the flexibility to handle ambiguous data or inputs at the AI step.
When automation wins
Automation wins when you have predictable, repeatable workflows with clear inputs and outputs.
Take a slightly more complex example. A deal is won. If the value of that deal warrants an invoice, the automation gets the details of the person associated with the deal, finds or creates a customer for them in QuickBooks, creates an invoice and sends it off.
You can make this more sophisticated without reaching for AI. A simple invoice goes one way. A multi-stage invoice might mean creating multiple invoices and only sending the first. A complex invoice might be split into multiple invoices that get reviewed, customised and sent independently. You can add a step to notify someone to check a particular invoice, and you can build in approval steps, putting a human in the loop who has to approve something before the next step runs.
So automation is best when you want consistency, auditability and predictable cost per run.
A note on auditability: it's technically available with AI, but it doesn't come baked in. It's custom and variable depending on your needs.e following months. It keeps you top of mind so they're more likely to come back when the timing is right.
When AI wins
AI is best where natural language, ambiguity or interpretation is involved.
In Asana, you can forward an email to a project and have that email become a task. Say there's an email from a colleague talking about all sorts of things, with one specific ask buried in it. Asana lets you build a rule using AI: you give it instructions on how to read the task that's created, work out whether it's a forwarded email, and decide what to do with it.
This brings everything together. You chat with AI to create an automation that itself uses AI. The part that matters is the AI step reading the content of the email and acting on it.
When you forward the email through, the task is created with everything from the original email. Then the AI reads the description, works out what's actually being asked, and strips everything else out, leaving exactly the ask and the exact date it needs to be done by. From one forwarded email, the system infers the intent, the urgency and the next step required.
Combining AI and automation
You've seen them separately. When should you combine them?
Use automation to trigger reliably, move data reliably and handle the fixed steps. Only put AI into the automation when interpretation is required.
Back to the newsletter form. Sometimes people use a contact form to solicit their own services: “if you need help with marketing, let us know.” That's not what the form is for. Website spam checks exist but aren't always reliable, and anyone with a contact form knows this is a problem you have to deal with.
So add an AI step that reads the message and decides whether it's a real submission or spam. Then add a filter step after it that reads the AI's output and decides whether to create the lead at all. You could even place this before the find-or-create-person step, so you check whether the record should be processed before you process it.
Here's another example: a sales call workflow. When a call finishes, the transcript is sent to an automation where an AI agent (OpenAI in this case) reviews it against a set of instructions. It adds the information to a tracker spreadsheet, emails the salesperson about how the call went, and creates a follow-up activity in Pipedrive.
There's a very clear workflow that needs to happen, but a lot of interpretation is needed too. The AI reads the transcript, understands the intent, applies the scoring rubric, gives feedback to the salesperson and determines next steps like when to follow up.
Because the model is chosen as a separate step, with no instructions locked to it, you can drop in a replacement model whenever you want. One option is OpenRouter: you sign up once and connect to any model through your account, switching between providers without signing up to each one individually.
The bigger takeaway
Businesses shouldn't be asking “how do we use AI here?” They should be asking “is AI the right tool for this situation?”
The most effective systems aren't the most advanced-looking ones. They're the ones most appropriate for the job, that get it done as well as possible without breaking the bank.
Having worked with AI and automation for years, the honest advice is this: don't go down the AI rabbit hole when automation will do the job perfectly well. You'll spend a lot of time and money getting AI to do something automation handles for a fraction of the price and far more reliably.
If you're at your wits' end, get in touch and we'll point you in the right direction. If you want to see all of this in action, watch the video.
FAQ
Is automation cheaper than AI? Usually, yes. For predictable, repeatable workflows with clear inputs and outputs, automation is cheaper, cleaner and more dependable, with a predictable cost per run. AI costs more and adds variability you don't need for fixed steps.
When should I use AI instead of automation? When natural language, ambiguity or interpretation is involved, like reading an email and working out the actual ask, or reviewing a sales call transcript and scoring it.
Can I use AI and automation together? Yes. Use automation to trigger reliably, move data and handle the fixed steps, and add an AI step only where interpretation is required, such as classifying a form submission as real or spam before creating a lead.
Does automation give me an audit trail? Automation gives consistency, auditability and predictable cost per run out of the box. Auditability is possible with AI too, but it isn't baked in and tends to be custom and variable.
Can I switch the AI model in an automation? Yes. If the model is a separate step with no instructions locked to it, you can drop in a replacement. Tools like OpenRouter let you connect to many models through one account and switch providers without signing up to each individually.