The decision rule
The best AI investment removes the active constraint first. If the same system also increases capacity and reduces headcount, the business gets a revenue and profit lever instead of a slightly cheaper version of the operation it already has.
Most conversations about AI return on investment begin with the wrong number.
A founder adds up the salaries attached to a process, estimates how much of the work software could automate and treats the difference as the potential return. That calculation can be useful, but it is not where I start.
Before I look at the financials, I find the constraint of the business.
A system that saves 10% or 20% on an expense can improve the margin. A system that removes the reason the business cannot grow can change the size of the company. When the constraint is real and the system is built around it, creating a route to two or three times the business can be more valuable than squeezing another small saving from a department that was not holding growth back.
01 · Diagnose before you calculate
The ROI spreadsheet starts one step too late
Return on investment is normally presented as a formula: expected gain minus cost, divided by cost. The arithmetic is easy. The difficult part is deciding which gain belongs in the calculation.
If you start with a task, the return looks like time saved. If you start with a department, it looks like payroll removed. If you start with the constraint of the business, the return can include more capacity, more clients, better results, stronger retention and a lower cost to deliver all of it.
Those are very different investment cases.
| Starting point | What gets measured | Likely impact |
|---|---|---|
| A task | Hours saved | Individual productivity |
| An expense | Payroll or software cost reduced | Improved margin |
| The constraint | Capacity, results, retention and profit | A different growth ceiling |
The task and expense still matter. They simply sit inside a larger question: what is preventing the company from becoming the business the founder wants it to be? If that is not yet clear, use the constraint-first framework for choosing an AI system before you model the return.
02 · Find the commercial problem
The three questions that reveal the constraint
I use three simple questions at the beginning of the conversation:
- 01
Where is the business now?
Establish the current revenue, profit, client count, capacity, retention and operating model.
- 02
Where does the founder want it to be?
Define the destination in operating terms, not just as a vague ambition to grow.
- 03
Why is the business not there already?
The honest answer to this question normally reveals the active constraint.
Agency owners tend to describe that final answer in a handful of ways:
- We churn too many clients.
- We do not sign enough clients.
- We do not have enough cash flow to reinvest properly.
- We do not get good enough results for clients.
Each answer points to a different system. A sales constraint might justify improving lead response, qualification and follow-up. A retention constraint might call for better delivery, communication or client-success infrastructure. A results constraint might sit inside research, strategy, implementation or the speed at which a team acts on performance data.
The technical answer can be automation, an AI agent, custom software or a combination. The architecture should follow the constraint, not the other way around.
“Where are you now? Where do you want to be? Why are you not there already? The answer to the third question is the constraint.”Harry Meighan
The value of the build comes from closing that gap. Until the gap is clear, any ROI forecast is just a precise-looking guess.
03 · Put a value on capacity
Calculate what the business could do after the constraint moves
Suppose a media-buying team can manage its current client base but needs to hire at roughly the same rate as the agency adds accounts. If a system doubles the team’s capacity without lowering performance, the commercial value is not limited to the hours it saves.
The agency can now support substantially more clients before adding the same payroll. If media buying was the constraint, that new capacity creates room for the entire business to grow.
A useful starting calculation is:
Capacity value
Additional client capacity × average monthly profit per client × expected retained months
Use profit per client, not headline revenue. Keep the assumptions grounded in the agency’s real sales rate and retention. Capacity only becomes valuable as the business fills it.
Then add the direct operating improvement:
Operating value
Payroll removed or avoided − AI usage − hosting − human supervision − maintenance
The strongest systems improve both sides of the calculation. They create room for more profitable revenue while reducing the cost and complexity required to deliver it.
04 · A real operating example
How an SEO operation moved beyond its 70-client ceiling
The clearest example from my previous Head of AI role was an SEO department inside a home-services marketing agency doing more than $3 million a year. You can also read the full SEO operations automation case study.
The service line was capped at roughly 70 clients. The human team was underperforming, and taking on more accounts meant adding more people to an operation that was already becoming harder to control. As the department pushed beyond its comfortable capacity, tasks were forgotten, mistakes increased and the quality of service dropped.
Demand was not the constraint. Delivery was.
We built a custom AI SEO platform that automated roughly 90% of the manual work. The system handled:
- Research across Reddit, keywords, competitor websites, competitor Google Business Profiles, competitor content schedules and BrightLocal data.
- Strategy creation based on that combined research.
- Task and deliverable generation based on the strategy.
- Recurring progress reviews to check whether the work remained aligned with the strategy.
- Content creation and posting.
- Review-response workflows.
The platform also tracked the work that still needed human implementation. One SEO operator and the web developer were accountable for completing Google Business Profile changes, website updates and other tasks the system assigned.
That distinction matters. The system did not automate disconnected SEO tasks and leave the department to coordinate the rest. It connected research, strategy, execution and accountability as one operation.
Seven people and roughly $20,000 in monthly payroll became one operator and a total operation costing approximately $5,000 a month, including the person and the system costs. That created an immediate saving of roughly $15,000 a month after launch.
More importantly, the 70-client constraint moved. The department has since passed 110 clients without finding another defined capacity ceiling, and the team continues to progress well.
If we had valued that platform only through payroll, we would have missed the larger return. It lowered the operating cost, increased the quality and consistency of delivery, strengthened retention and created capacity for at least 40 additional accounts.
Experience note: this platform was built during Harry’s previous in-house role as Head of AI, before Build, Ship & Scale. The client name and identifying details are withheld. Results are specific to that operation and are not a guarantee of future performance.
05 · Count the complete return
The financial case has four parts
Once the constraint and the new operating model are clear, the investment case becomes much easier to structure.
| Area | What to calculate |
|---|---|
| Additional capacity | The profit available from clients the operation can now support |
| Direct savings | Payroll and subscriptions removed or future hires avoided |
| Retained profit | The contribution protected when results and retention improve |
| System cost | The build, AI usage, hosting, supervision and maintenance |
The system costs should be complete. Depending on the build, that includes model and API usage, hosting, any human supervision the workflow requires and ongoing maintenance with the development team.
Some of the current operating cost is harder to put into a clean cell. What does it cost when work is forgotten and a client leaves? What is the cost of managers checking avoidable mistakes? How much growth is lost because the team cannot accept another ten accounts? What does inconsistent output from less productive team members do to the client experience?
Founders rarely have perfect figures for those questions. They still understand how expensive the answers can become.
The model does not need to manufacture certainty. Use the numbers the business can defend, state the assumptions and keep the harder-to-quantify value visible alongside them.
A practical 12-month model
Capacity profit + direct savings + retained profit − build and operating costs
Once those inputs are defined, calculate the percentage return and payback period:
AI system ROI formula
ROI % = (total financial benefit − total investment) ÷ total investment × 100
Payback period = initial investment ÷ monthly net benefit
A worked AI system ROI example
Take an illustrative system with a $50,000 initial build cost and a conservative $15,000 monthly net benefit after usage, hosting, supervision and maintenance. Over 12 months, the financial benefit is $180,000.
The first-year ROI is 260%: ($180,000 − $50,000) ÷ $50,000 × 100. The simple payback period is approximately 3.3 months: $50,000 ÷ $15,000.
This example deliberately excludes speculative growth. If the system also unlocks profitable client capacity or protects revenue through better retention, model those benefits separately and state the assumptions.
Monthly operating savings can begin in the first month after the completed system goes live. Capacity and retention returns build as the company uses the system to grow and serve clients more effectively.
06 · Choose the highest-leverage build
Do not optimise an expense while the constraint remains
A costly department is not automatically the best place to begin. If it is not stopping the business from reaching its target, replacing it may produce a saving while leaving the most important commercial problem untouched.
That is why the order matters:
- 01
Find the constraint.
Identify why the business has not reached the next operating target.
- 02
Design the new capacity.
Define what the operation should be able to handle after the system is live.
- 03
Model the profit unlocked.
Value additional capacity using realistic profit and retention assumptions.
- 04
Add the efficiency gain.
Include payroll removed, hires avoided and other direct savings.
- 05
Subtract the complete system cost.
Count the build and the ongoing cost of operating it properly.
The result is a more useful question than, “How many hours will this save?”
Ask instead: “If this system removes the reason we cannot grow, how much more profitable can the business become?”
Once the investment case is sound, compare the practical routes in our guide to building, buying or hiring for an AI system.