Choosing a first AI automation project is where many founders go wrong. They want something impressive, something they can describe at a barbecue. An AI that writes all their marketing. An AI that handles complex client negotiations. Something with a bit of theatre to it.
Then it half works, produces output nobody trusts, and quietly gets abandoned six weeks later. The founder concludes AI is overhyped and goes back to doing everything manually.
The projects that stick are dull. Appointment reminders. Quote follow-ups. Sorting the inbox. Chasing the review request nobody remembers to send. Tasks so repetitive that nobody in the business wants them, and so well-defined that the AI cannot really get them wrong.
This post covers how to pick that first task, how to score it, and what actually happens after you cross it off.
Key Takeaways
- Your first AI automation project should be the most repetitive, lowest-judgement task in the business, not the most impressive one.
- Boring tasks have clear inputs and clear outputs, which is exactly what makes them reliable enough to trust without supervision.
- MIT research found 95% of AI initiatives return nothing, mostly because they start with technology instead of a specific process.
- Score every recurring task as fully automatable, assisted, supervised, or human-only. Start at the top of the fully automatable list.
- Lead response is usually the highest-value boring task: 78% of deals go to whoever replies first, and most businesses take hours.
- One task automated properly builds more internal belief than five half-finished clever projects.
- Track the percentage of recurring tasks the system handles. 20% to 30% is the point where you feel it in your week.
Why the Clever Project Always Fails First
The clever project fails for a simple reason. It asks the AI to make judgment calls that nobody in the business has ever written down.
Take “automate my sales emails.” Sounds like one task. It is actually forty decisions: which prospects get which angle, when to push and when to back off, what tone suits a builder versus a dentist, when to stop chasing. You have been making those calls on instinct for years. None of it exists anywhere except your head.
So the AI guesses. It produces something that reads plausibly and is subtly wrong. You correct it. You correct it again. After the fourth round, you think, faster to just write it myself, and you are right. You were always going to be right, because the task was never defined well enough to hand over.
The MIT NANDA research on enterprise AI pilots found that roughly 95% of these initiatives produce no measurable return. The failure pattern is consistent, and it is not the technology. It is starting with what the tool can do instead of what the business actually needs handled.
A boring task has none of that ambiguity. “Every quote that has not had a reply in three days gets a follow-up message” is not a judgement call. It is a rule. You could teach it to a school leaver in four minutes. That is precisely why the AI will run it for a year without you checking.
What Makes a Task Boring Enough to Be a Good First Project
Four tests. A task that passes all four is your starting point.
It happens on a schedule or a trigger
Something predictable sets it off. A new lead lands. Three days pass. A job gets marked complete. The month ends. If the task only happens when you decide it should, it is not ready yet, because “when I decide” is the part that lives in your head.
The inputs are already in a system
The information the task needs exists somewhere a computer can read it: your CRM, your accounting file, your booking calendar, your inbox. If the AI would need to ask you for context every time, it is not automating anything. It is just adding a step.
The output looks the same every time
A confirmation message. A follow-up text. A record updated. A summary written to the same place in the same shape. Consistency of output is what lets you stop reviewing it after the first fortnight.
Nobody enjoys doing it
This one is not sentimental. The tasks people resent are the ones they do late, do inconsistently, or quietly skip. Those gaps are costing you money right now, and they are the gaps an automation closes permanently.
Run those four tests across your week, and you will usually find the same suspects: lead response, quote follow-up, appointment reminders, review requests, invoice chasing, and getting new job details from the enquiry into the system. Not one of them is exciting. All of them leak money when they slip.
The Task Audit: How to Pick the One
Do not automate from memory. Founders consistently misjudge where their time goes, usually by a wide margin.
Sit down and list every recurring task in the business. Yours and your team’s. Daily, weekly, monthly. Most businesses land somewhere between fifty and a hundred, which is itself a useful shock.
Then score each one into four buckets:
Fully automatable. Clear trigger, clear rule, clear output. The system does it start to finish and you never see it again unless it breaks.
Assisted. The AI does roughly 80%, and you steer. Drafting, first-pass research, summarising. Good second-wave candidates.
Supervised. The AI does nearly all of it, but a human signs off before anything goes out. Anything client-facing with money or reputation attached tends to live here for the first few months.
Human-only. Judgement, relationships, negotiation, hiring. Leave these alone. Trying to automate them is how the clever project gets started.
Now take your fully automatable list and sort it by hours consumed per month. The task at the top is your first AI automation project. Not the one you find interesting. The one at the top.
One caution on scoring: be honest about what is genuinely rule-based. The temptation is to mark something fully automatable because you want it gone. If you cannot write the rule in two plain sentences, it belongs in the supervised bucket for now.
Lead Response Is Usually the Right Boring Answer
For most founder-led businesses, the task at the top of that list is responding to new enquiries.
It is about as dull as work gets. Someone fills in a form or rings the office. Somebody replies, asks a few standard questions, and books a time. No creativity required. No judgement beyond a simple qualification check.
It is also where the money leaks fastest. Research on inbound lead handling consistently shows that around 78% of deals go to whoever responds first, and most businesses measure their response time in hours rather than seconds. Your leads are not comparing your service to your competitor’s. They are picking whoever gets back to them while the problem is still front of mind.
An automated first response does the same four things every time: acknowledges the enquiry within seconds, asks the qualifying questions, offers available times, and writes everything into the CRM. Predictable. Repetitive. Exactly the profile you want in a first project.
The related pattern is the dormant database. Every business sitting on a few thousand old contacts has revenue in there it has already paid to acquire. James, a finance broker I worked with, had 319 contacts his team had written off entirely. A multi-touch reactivation running over SMS and email recovered $49,000 from that list. The work itself is mind-numbing, which is the entire reason nobody had done it.
If the phone is the leak rather than the form, the equivalent boring project is call handling. Every call answered, qualified, and logged, including the after-hours ones. Dr Claire’s practice went from missing a large share of inbound calls to missing none, and booked appointments rose 44%. Not a clever automation. A phone that always gets picked up.
What Happens After the First One
Here is the part founders do not anticipate. The first boring automation changes how you think more than it changes your calendar.
Week one, you check it constantly. Week two, you check it daily. By week four, you have stopped thinking about that task entirely, and something shifts. You start noticing other tasks the same way, cataloguing them as you do them: this is repetitive, this has a clear trigger, this could go too.
That shift is worth more than the hours you recovered. It is the difference between “I should look into AI sometime” and actively hunting for the next thing to hand over.
The second automation is faster to build than the first, because the groundwork is already there: the system knows your business, your data is connected, the plumbing exists. The third is faster again. This is why starting boring beats starting clever. Boring gets finished, and finished compounds.
Track it with one number: what percentage of your recurring tasks does the system handle now? Start at zero. At 20% to 30% you feel it in your week, genuinely, not as a spreadsheet abstraction. Somewhere past halfway, the business starts running on something other than your attention.
There is a ceiling on what disconnected automations can do, and you will hit it. At some point the individual tasks need to sit inside something that knows the whole business: your context, your numbers, your clients, your priorities. That is the AI brain and AI workforce layer, and it is a different conversation. But you do not start there. You start with the follow-up message nobody sends.
Conclusion
Your first AI automation project should be so boring that describing it at a barbecue would be rude. Clear trigger, clear rule, clear output, and a task everybody in the business quietly resents.
That is not a compromise. It is the whole strategy. Boring tasks are the ones the AI handles reliably enough that you stop supervising, and the ones you stop supervising are the ones that actually give you your week back. Clever projects consume attention forever and give nothing back.
List your recurring tasks. Score them honestly. Take the dullest, highest-volume item on the fully automatable list and finish it properly before you touch anything else. Then do the next one. The interesting work arrives later, once the boring work is running itself.
Ready to Find Your First One?
If you are not sure which task to start with, that is the conversation to have. Book a 30-minute discovery call, and we will work through your week, find the tasks worth automating first, and you will leave knowing exactly where to start. No pitch, no pressure.
Worth reading next: what to track at the 90-day mark once your first automation is live, and how AI agents actually work for small business if you want the wider picture before you commit.
Frequently Asked Questions
What should my first AI automation project be?
The most repetitive, highest-volume task in your business that has a clear trigger and a predictable output. For most founder-led businesses, that is responding to new enquiries, following up on unanswered quotes, or sending appointment reminders. Avoid anything requiring judgement you have never written down. If you cannot describe the rule in two plain sentences, pick something else first.
How long does a first AI automation project take to set up?
A well-scoped single task typically takes days rather than months. Lead response or appointment reminders can be running inside a week when the data already lives in a CRM. The delays are almost never technical. They come from unclear rules, missing access to systems, or nobody deciding what the automation should actually do when an edge case appears.
How do I know if a task is worth automating?
Multiply the time it takes by how often it happens, then ask whether it slips when the week gets busy. A ten-minute task done twenty times a month is over three hours, and those are usually the tasks done inconsistently. Anything that leaks revenue when skipped, such as lead follow-up, is worth automating even at low volume.
What is the difference between AI automation and regular automation?
Regular automation follows fixed if-this-then-that rules and breaks the moment something unexpected arrives. AI automation handles variation: it reads a message written in plain English, works out what the person wants, and responds appropriately. For a first project, you often want something closer to the rule-based end, because predictability is what builds trust early on.
Should I automate something myself or get help?
Try one yourself if you have the time and curiosity, as the learning is genuinely valuable. Get help when the task touches customers, money, or your CRM, because a broken customer-facing automation costs more than it saves. Most founders do best with the first one built properly by someone else, then run the rest of the audit themselves.
What if my first automation does not work?
Diagnose it before abandoning it. Most first attempts fail for one of three reasons: the task was not as rule-based as it looked, the data it needed was not accessible, or nobody defined what happens in unusual cases. All three are fixable. What is not fixable is concluding AI does not work for your business off a single badly scoped attempt.
How much of my business can realistically be automated?
Across a typical founder-led business, 60% to 70% of recurring tasks can be automated or heavily assisted within six months. That does not mean 60% of the work disappears. It means the repetitive layer moves off your plate, leaving judgment, relationships, and growth, which is the work you started the business to do.
About Octavius
Titus Mulquiney is the founder of Octavius AI, where he builds AI brains and AI workforces for founder-led businesses stuck running everything out of their own head. Twenty years in marketing, ex-Sony product manager, ex-GM Zeal NZ. Based in Auckland, working with operators across NZ, Australia, and the US. Connect on LinkedIn.