The question I get most often from operations leads is that they know they should be automating things and they don't know where to start. The honest answer is smaller than you'd think and sooner than you'd think. Here's the longer version.
The first move isn't picking a tool. It's writing down every task somebody on your team does more than ten times a week. That list usually ends up somewhere between thirty and fifty items and it takes about an afternoon to put together. Most of them will look boring, which is the point.
Then score each one on three things: how much time it costs per week, how much it varies (does it look the same every time?), and how much judgement it needs. The ones worth automating are high on time and low on the other two. Forwarding emails and categorising expenses come out near the top. Legal review never does.
Start with the boring ones. The first automation I build for a new client is almost always something nobody is excited about, usually invoice routing or syncing a calendar to a CRM. The reason is that boring automations have clean success criteria. Either the email got to the right person or it didn't. You want to build trust before you spend any.
Use the tools that already exist first. Zapier, Make and n8n handle most high-volume low-judgement workflows for a fraction of what custom code costs. The point to move to something custom is when an off-the-shelf tool stops fitting, usually because the workflow has too many branches or it touches data the SaaS provider can't see.
Where AI is worth adding is the moment in a workflow where someone has to read something and then decide. Read this email and work out which team it belongs to. Read this invoice and work out the category. Read this ticket and work out if it's urgent. Models are good enough now that a small custom layer on top of a workflow tool can turn minutes into seconds.
Where I'd be careful is anything touching money, contracts or customers without a person in the loop. Even at 99% accuracy, the remaining 1% generates enough work to wipe out what you saved. Keep the approval gates at the edges and let automation handle the volume in the middle.
The number to watch isn't how many things you've automated, it's how many hours you've got back. A team that's automated forty things and saved no time has automated the wrong forty things. A team that's automated three and got back half a day a week has done it right.
The hard part is knowing when to stop. Once a team starts automating, the temptation is to keep going until everything is covered. Most processes have a layer worth touching and a layer underneath it that isn't, and knowing where that line sits is most of what I bring to this kind of build.
If you want to go through this for your own business, that's what the scoping call is for. We'd talk through your highest-volume workflows and work out whether any of them are worth building something for.