The short version
Your first AI system should either remove the constraint stopping the business from growing or materially reduce its largest controllable cost. Start with the constraint. Then, while the business grows towards the next one, work on cost.
Founders do not struggle to find things they could automate. The list is normally endless. The hard part is deciding what is worth automating first.
I learned that from building systems day in, day out as a Head of AI inside a seven-figure marketing agency. My co-founder, Josh Henry, had the same job inside another seven-figure agency. Across our team’s careers, we have collectively built more than 100 AI systems for businesses serving automotive, medical, HVAC, solar, home services and other markets, as well as info businesses.
The pattern is always the same: the best first build is rarely the shiniest idea in the room. It is the one that changes the operating reality of the business.
01 · Start with the actual problem
Why founders choose the wrong first AI project
At one agency, the founder wanted us to replace a perfectly good automated SMS platform with a custom-built version. The existing tool worked. It ran at scale. It cost about $200 a month.
Building the replacement would have taken several weeks and consumed roughly $4,000–$6,000 worth of my time. In other words, we would have spent the equivalent of around 20–30 months of subscription fees to recreate something that was already working—and it would not have removed the constraint in the business.
The real problem was client retention. Communication was too easy to miss, updates were inconsistent and the client experience needed tightening. So I pushed back on the SMS rebuild and automated client success instead. That made communication far harder to let slip and addressed the problem that could actually stop the company from growing.
“The worst first AI project is usually the one you build just so you can say you have an AI system.”Harry Meighan
02 · The order of operations
The constraint-first framework
There are two things I look at when deciding what to automate: the active constraint and the biggest controllable cost. The order matters.
- 01
Find the active constraint.
What is stopping the business from taking on more customers, retaining them or delivering the result consistently?
- 02
Test whether a system can materially relieve it.
The process needs to be understood, frequent enough to matter and measurable before and after the build.
- 03
Fix the constraint first.
This creates room to scale. Otherwise, you are making a non-critical corner of the business more efficient while the real bottleneck remains.
- 04
Then automate the largest controllable cost.
While the company grows towards its next constraint, improve the economics of that growth.
- 05
Ignore attractive distractions.
If a project does not increase capacity, protect retention or improve profitability, it is probably not the first system to build.
| Project type | What it changes | Priority |
|---|---|---|
| Relieves the active constraint | Capacity, retention or delivery quality | Build first |
| Reduces a major controllable cost | Profitability and operating leverage | Build next |
| Does neither | Mostly novelty or a minor convenience | Leave it alone |
03 · Diagnose before you build
How to find the active constraint
Ask a blunt question: if demand doubled next month, what would break first? The answer is normally close to the constraint.
Look for evidence such as:
- Clients leaving because delivery quality is inconsistent.
- A backlog that keeps growing even after adding people.
- A process that depends on the founder to keep moving.
- A team that cannot take on more accounts without quality dropping.
- Slow response times that are costing appointments or renewals.
- A high-value process with constant manual handoffs and rework.
Do not start with the software. Map what happens today, where the work stalls, who touches it and how you will know the constraint has moved. If you cannot describe the current process or measure the outcome, you are not ready to automate it properly.
Three readiness checks
- The underlying service or process is already proven.
- The work happens often enough for automation to matter.
- There is a baseline and a clear target for the new system.
04 · From capacity to profitability
What this looked like inside a $3m automotive agency
In my previous Head of AI role, I worked inside an automotive marketing agency doing roughly $3 million a year. Its SEO operation was the constraint. The team cost about $20,000 a month, performance was inconsistent and weak delivery was contributing to client churn. The operation was effectively capped at around 70 clients.
We automated the SEO workflow across research, strategy, production, publishing, optimisation and reporting. The existing department was replaced by the system and one Head of AI SEO who supervised it. Monthly delivery cost dropped from roughly $20,000 to $5,000, and the operation went on to support more than 100 clients.
That was phase one: remove the constraint. Once SEO delivery could scale, I moved to the largest cost opportunity.
The same agency had a team of ten call-centre representatives. We built a 24/7 AI call centre that handled inbound calls, live transfers and appointment booking. It also produced cleaner operating data than the previous setup. Replacing that cost base saved more than $250,000 a year in salaries, while the improved coverage also helped reduce churn.
Experience note: this work was completed during Harry’s previous in-house role as Head of AI, before Build, Ship & Scale. It is included to show the operating framework behind the team’s current work.
05 · Keep the humans who add value
Not every system should replace a team
Full replacement made sense in parts of the automotive operation because the work was high-volume, repeatable and measurable. That does not mean replacement should be the goal everywhere.
Most businesses do not need to become “fully AI.” They need to make their existing people more productive. A system can prepare research, handle repetitive admin, surface risks, draft updates or complete the predictable 80% of a workflow, while a human owns judgement, relationships and exceptions.
If the process is high-stakes, poorly understood or changes constantly, start with augmentation. Give the team leverage, keep approvals in the loop and earn the right to automate more as the data becomes clear.
06 · A clear “not yet”
What you should not automate first
Leave a project alone for now if it is cheap, already works well and has no meaningful effect on capacity, retention or margin. The fact that a custom AI version is possible is not a business case.
This is also why I would not recommend this process to a small business that is just starting out. If the offer, demand or fulfilment model is not proven, automation can hard-code the wrong process and help you make the same mistake faster.
It becomes valuable when the business is established—usually doing at least $50,000 a month—with a proven operation and a constraint that is genuinely stopping the next stage of growth.
07 · From idea to production
What a serious AI system build should look like
Once you have chosen the right problem, the build still needs discipline. At Build, Ship & Scale, we start by developing a detailed brief with the team that runs the process. That removes guesswork before development starts.
Our developers build the system, our QA team tests it and we deploy it into a UAT environment before it touches production. The people who will actually use it can validate the workflow, data and edge cases safely.
We deliver the first production-ready version of every system within 45 days of coming on board. After launch, maintenance packages keep developers available for bugs, questions and new features for as long as the business wants the support.
Speed matters, but speed aimed at the wrong problem is just efficient waste. Get the order right: constraint first, cost second, everything else later.