If you are wondering about the AI ROI gap – you know where companies are supposed to see all this savings from AI, you are not alone. Companies are firing thousands of staff because of the savings that AI is going to bring them. Oh, and it’s going to be delivered by little leprechauns driving tiny carriages. I guess that was silly, but the first part that seems to make sense to a lot of CEOs is silly too.
The companies that haven’t been able to figure out remote work in the past 6 years are suddenly making major structural and staffing decisions about AI that they haven’t even used yet. If you haven’t documented your processes, you shouldn’t be making staffing changes.
The False Promise of Headcount-first AI Adoption
The dominant narrative is pretty simple: AI = labor replacement = cost savings. It’s all over the internet, so it must be true. Oracle has laid of 21,000 staff in the last year either because of AI, or anticipation of AI savings. It’s hard to tell what the truth is is. Some of us have seen this play before.
I was a casualty of outsourcing a few decades ago. The argument was similar. We can have a small army of people doing programming, testing, etc., for a couple of dollars an hour. We’ll get rid of these high-priced employees and consultants and get things done faster at a lower cost. Most people remember the stories of the layoffs. There was much less fanfare as many of the outsourcing contracts were quietly cancelled and people were brought back. Usually this was done by a different group of people than those who made the cuts in the first place. You can guess what happened to them.
Here we are again, watching history repeat itself. Companies announce layoffs citing “AI efficiencies” before tools are deployed or proven. Really – how do you know? Leaders are just tapping into sound bites and assuming facts based on volume. Investors and boards are amplifying the problem by exerting pressure to show immediate cost reductions, which drives premature cuts. There’s an underlying assumption that existing work is efficient, just too labor-intensive. If only companies would test that assumption first, they would find out that it’s probably wrong.
Why AI ROI isn’t Materializing
Adding technology to a process rarely makes it more efficient. It just makes things happen faster. AI tools automate tasks, not dysfunctional processes, so garbage in, garbage out still applies. When you cut people first, institutional knowledge walks out the door with the people who understood the workarounds. The remaining staff inherits more broken processes with less support to manage them. That’s when people start to ask questions.
AI adoption often stalls without the domain experts needed to train, prompt, and validate outputs. The CEO may look like a hero to the board and stockholders for a quarter or so before reality starts to set in. Then short-term labor savings are offset by rehiring, contracting, and productivity losses
The Process Problem
Most enterprise processes weren’t designed. Rather, they accumulated over decades. Many times the process was altered to account for someone’s skill or lack of skill, or a position that needed to be looped in for political reasons. When I was in the Air Force, I showed how something that should take four hours expanded into a process that took 7-10 days. Not surprisingly, that revelation was not well received. Things like that make management look bad. But it wasn’t really their fault. It was a series of small changes over time where people made adjustments and just took the process as gospel.
Manual steps, approval chains, and redundant handoffs made sense once and were never revisited. Then they get documented into workflows and new computer systems. They might work faster, but they generally stopped being efficient a long time ago. Now enters – AI. Myths, hopes and eyes filled with dollar signs cloud the judgement for what comes next for many companies.
Whether you address staffing first or later. AI inserted into a broken workflow produces faster broken results. If you’ve already gotten rid of the people, you don’t always have the right people around to clean up the mess. Or to bring Jurassic Park back online after the main switch is pulled.
Accumulated Process Debt
Process debt is an accumulation of inefficiencies that are assumed to be normal and efficient over time. They actually slow productivity and are financially draining, but no one wants to see that. They are also essentially invisible until you try to automate processes.
I once had a senior executive tell me, “I’ve been here 30 years, and I didn’t know we were doing that.” Initially, when I showed him the process map we had created, he told me we were wrong and wanted to know where we got the information. The supervisor who oversaw the work was basically summoned to the meeting to confirm what we had documented. He stepped through the process and confirmed that it worked as documented.
They were still puzzled so they called in a secretary who had been there almost 40 years, who explained when the process started and why. Then everyone agreed that it made no sense – and hadn’t for almost 30 years. These are probably representative of some of the processes we are trying to automate with AI. If you don’t have an accurate process map, there is no way to measure whether AI is actually helping.
What Process-First Transformation Looks Like
Process mapping is often an exercise in delayed gratification. It’s amazing how many people will take the “We know what we do” attitude and just start making changes. It is really important to map current-state workflows before any AI or staffing decision. And yes – there will be surprises.
You may not need to look at all processes initially. Identify high-friction, high-volume tasks that are also well-defined and auditable. Now the hard part – assume nothing is sacred. Not the boundaries of the process, the people or departments doing the work, the building it is being done in, or the software that you had custom built, etc. Once you say, “you can change everything but…” you are recreating a new batch of inefficiencies that have snuck into the process up to this point. AI may come up with a completely different way of doing things.
Redesign the process assuming AI assistance. Don’t just bolt AI onto legacy steps. So many process redesigns turn into exercises in automating paper. Things might get sped up a bit, but rarely will you see significant changes without taking things down to the studs and starting over. We all know that jobs could be at stake, but don’t just assume a change in headcount. AI could just make everyone more efficient – which is a best-case scenario. It is important to retain process experts through the transition; they become the AI supervisors and validators. From that, you can set measurable baselines so ROI can actually be tracked.
The Right Sequencing Model
- Step 1 — Audit: document current processes; identify waste, bottlenecks, and tribal knowledge dependencies
- Step 2 — Redesign: rebuild workflows around AI-assisted execution; eliminate steps AI makes obsolete
- Step 3 — Pilot: run AI tools alongside existing staff; measure against the baseline
- Step 4 — Optimize: refine prompts, models, and integrations based on real outcomes
- Step 5 — Right-size: make staffing decisions based on demonstrated capacity, not projections
Implications and Recommendations
Cutting staff because of a new technology usually is good for a stock spike and maybe a bonus for one quarter. Then reality tends to set in. For long-term results, leaders need to reframe the AI business case from cost-cutting to capacity expansion. If jobs ultimately need to be cut, HR needs to protect process knowledge during transitions and offer reskilling before severance. You might need Doug from accounting a year from now; having him in a different role benefits both sides.
Boards also need to be wary of shiny objects. It is so tempting to buy into the hype to capture market share. But done wrong, it can be equally disastrous. Boards should require evidence of process readiness before approving AI-driven headcount reductions. The companies that will win are those that use AI to grow output, not just shrink payroll.
