Article Series · Labor Optimization
The first two articles in this series covered salary arbitrage and Global Capability Centers as strategic labor levers, and the diagnostic tools for right-sizing an existing workforce. This final article addresses the dimension that is reshaping every assumption in labor optimization: automation and artificial intelligence.
Automation: Capacity Building and Monetization Challenges
I have started and managed two major RPA organizations over the past fifteen years. Both began with similar projections: startup costs in the low six to low seven figures in licensing, a team of staff and contractors, and a year-one projection of 60 to 95 FTEs of freed capacity at an average run rate of $60K per year — yielding $3.6 to $5.7 million in first-year savings, growing every year thereafter.
My first question to both teams was the same: give me a list of the names of the people in those positions who will be eliminated at the end of year one. Both times, I received a blank sheet of paper.
This is the fundamental tension in automation: you will save 60 FTEs of effort, but it happens to be 25% of each of 240 roles. You still need the other 75% of each of those roles to be done by a human. Automation typically frees capacity within a role; it does not eliminate the role.
When the freed capacity exceeds approximately 20 FTEs, it becomes possible — with some effort — to restructure the work program and eliminate a percentage of positions. In my experience, I was able to reclaim roughly one-third of the freed capacity through this kind of restructuring. The rest was absorbed by the organization's seemingly unlimited appetite for additional capacity.
The investment — $2 to $5 million per year to stand up and run an RPA team — is therefore not a cost savings investment in the traditional sense. It is a productivity and quality investment. That is a legitimate and valuable thing to invest in. Just be clear-eyed about what you are buying.
How to Calculate RPA Savings
The right way to calculate the FTE savings from an automation effort is to measure the actual time eliminated per transaction and multiply by volume. An example: I was overseeing an IT team supporting an underwriting function. The underwriting process required staff to log into six different systems, retrieve data, copy it into a Word document, and then print it before beginning the actual domain-knowledge work of evaluating the policy. The front-end data collection averaged 24 minutes per transaction. Our RPA team built a bot that completed the same task in 1.5 minutes.
We calculated the FTE savings by multiplying the 22.5 minutes saved by the number of transactions per year, then dividing by 1,900 hours per year. Critically, nobody considered reducing the number of underwriters. What the bot did was remove 24 minutes of non-core, mechanical work from each transaction, allowing underwriters to focus entirely on the judgment-intensive work they were hired to do. That is the right framing for RPA value.
AI and the Coming Shift in IT Labor Economics
The more consequential transformation is not in the automation sphere but in AI's direct impact on the roles that make up the majority of IT labor: software engineers, QA testers, business analysts, DevOps engineers, and program managers.
AI code generation tools — Claude Code, Cursor, Kiro, and others — are already demonstrating dramatic productivity improvements in software development. I have spoken with former colleagues who have been writing significant quantities of code for two or more decades. They are in agreement: they will not be writing significant amounts of code going forward. Their role is shifting to prompting AI systems effectively and reviewing and editing the resulting output.
The code creation phase of projects — historically the primary bottleneck — is on a path to becoming a minutes-or-hours activity rather than a weeks-or-months one. Most of these tools also generate QA test cases and documentation as associated artifacts of the code they produce.
My Hypothesis: Projected Reductions by Role Over 3 to 6 Years
The following reflects my assessment, framed explicitly as an educated projection rather than a certainty.
- QA Tester / Engineer: 80% reduction
- Software Engineer: 70% reduction
- ETL / BI Developer: 40% reduction (assuming report design remains a human activity)
- DevOps Engineer: 85% reduction
- Business Analyst: 70% reduction
- Program Manager: 50% reduction
- Architects: Little impact, assuming AI systems do not also usurp the creative and judgment-intensive dimension of architectural work
These reductions will occur progressively as tools mature and organizational adoption spreads. Most of the eliminated roles will be at the junior and middle levels. Senior roles will remain necessary to review and direct AI outputs. This creates an obvious longer-term pipeline problem: today we develop senior software engineers by starting with junior engineers and giving them 15 to 20 years of experience. If there are no junior engineers entering the pipeline, how do we replace retiring senior talent in a decade?
One caveat: these projections assume reductions against the existing body of work. Every productivity-enhancing technological advancement has expanded the possible body of work such that more new jobs are created than those that were made obsolete. I do not see fewer total job opportunities in the future — but the job market will be disrupted as required skill sets differ.
The Timeline: Why It Is 2 to 5 Years, Not 2 to 5 Months
The tools need roughly another 12 months to fully mature to the point where the average organization is willing to deploy them broadly. Add 1 to 3 years to overcome corporate inertia, risk analysis, change management, and the resistance that accompanies any technology transition of this magnitude, and the realistic horizon for full-scale impact is 2 to 5 years from today. Early adopters are already moving. Most organizations are still in the learning stage.
What This Means for Your Labor Strategy Right Now
The approaching transformation does not invalidate the labor optimization levers described in the first two articles of this series. It requires them to be applied within a five-year lens.
Salary arbitrage still makes sense for any role you will need for at least a year. The ROI of moving a resource offshore is achievable within that timeframe even if the role is eventually eliminated by AI adoption.
GCC construction should factor in AI-driven reductions over a 2 to 5-year horizon. Mitigate space risk by negotiating lease clauses that allow for reduction of occupied space at renewal or even mid-lease. Office space in India costs approximately $1,200 to $1,700 per FTE per year — a small figure relative to the labor savings a GCC generates. The ROI case remains compelling even accounting for future headcount reductions.
Plan for workforce transitions proactively. Determine whether natural attrition can absorb the projected reductions or whether deliberate reductions will be necessary. If the latter, use larger, less frequent actions — spaced at least 18 months apart — rather than a series of smaller, more disruptive events. Repeated restructurings demoralize the workforce and drive out the best talent first.
Do not let AI speculation freeze strategic decisions. Executives are increasingly using AI uncertainty as a reason to defer offshoring, insourcing, and GCC investments. The economics of those programs deliver strong ROI within 18 months. Even in a scenario where AI eliminates half the positions at the 18-month mark, the savings are substantial. The speculation is not a reason to wait.
The Bottom Line
Automation and AI are real, consequential forces that will reshape the IT labor landscape over the next several years. They do not change the fundamental logic of labor optimization — they accelerate and intensify it. The organizations that will capture the most value are those that continue to optimize their labor base today while building the planning discipline to manage the transition that is coming. The money is there now. The transformation is coming. Both require your attention.