AI Should Remove Bad Work Before It Removes People

One of the biggest problems I have with the AI conversation is that people seem to think the goal is replacing employees with AI.

It's not.

Cost Cutting Is Not The Same As Operating Improvement

There is a reason cost reduction is tempting.

It is visible. It is measurable. It can happen quickly. If a company cuts payroll, the savings show up in the model right away.

Revenue growth is harder.

It requires better execution, better customer experience, better sales follow-up, better delivery, better management, and better decisions over time. It is less controllable in the short term.

But that is exactly why the distinction matters.

The goal is not to choose between growth and cost discipline. The goal is to remove waste in a way that makes the operation stronger.

BCG’s transformation research is useful here because it does not treat cost reduction as irrelevant. Cost reduction can create breathing room, especially in the middle of a turnaround. But over time, the bigger value creation story depends on growth. A company cannot cut its way into becoming a stronger business forever.

Bain makes a related point from a different angle. The best companies don't treat growth and cost as opposites. They reduce waste, complexity, and unnecessary work in ways that make the business more capable, not more fragile.

That is the standard AI should be held to.

If AI only makes payroll smaller, that may improve the next quarter.

But if AI improves response time, follow-up, reporting, handoffs, onboarding, decision-making, and customer experience, it can make the business more profitable without weakening the system that creates revenue.

That is a much better goal.

The question is not simply, “Can AI reduce cost?”

Of course it can.

The better question is, “Can AI remove waste in a way that increases capacity, improves execution, and supports profitable growth?”

That is a deeper operating question.

A business can become more profitable because it eliminates people.

But a business can also become more productive because the same people can now handle more work, make better decisions, respond faster, follow up more consistently, and spend less time fighting bad systems.

Those are very different operating philosophies.

The first one is defensive.

Cut cost. Shrink payroll. Hope the work still gets done.

The second one is growth-oriented.

Keep the people who understand the business. Remove the low-value work around them. Give them better tools. Improve the workflow. Increase output.

That is the version of AI I am much more interested in.

Staff Are Not Just A Cost Line

Payroll is usually one of the biggest expenses in a business, so it is natural for owners to look there when margins tighten.

I understand that. Sometimes headcount is wrong. Sometimes a business has hired ahead of itself. Sometimes a role no longer makes sense. Sometimes cuts are necessary.

But that can’t be the default move.

In most founder-led businesses, employees are carrying a lot more than their job descriptions show.

They carry:

• Customer context
• Vendor context
• Internal history
• Exceptions
• Workarounds
• Judgment
• Relationships
• Local knowledge
• The memory of why something is done a certain way

The reality is that most of this is not written down anywhere. New tools are making it easier to capture institutional knowledge in real time, but most businesses are not starting from a clean system. They have years, sometimes decades, of context stored in the heads of their staff.

AI can automate tasks. It can help capture knowledge. It can make information easier to find. But it does not automatically replace judgment, trust, context, or the informal knowledge that keeps a business moving.

This is the part that gets missed when a leadership team looks at a spreadsheet and says, "If AI can do 40% of this role, we can cut 40% of the people."

That sounds clean in a model.

It is rarely clean in reality.

Most jobs are not just bundles of tasks. They are part of a system. When you remove the person, you don’t just remove the task. You remove the judgment around the task, the exception handling around the task, and the connective tissue between that task and the rest of the business.

Sometimes that is fine.

Often it creates a new bottleneck somewhere else.

The Klarna Lesson

Klarna is the example everyone points to because it is so public.

In February 2024, Klarna announced that its AI assistant had handled 2.3 million conversations in its first month, about two-thirds of its customer service chats. The company said the assistant was doing the equivalent work of 700 full-time agents, reduced repeat inquiries by 25%, and could drive $40 million in profit improvement for the year. You can still read the original announcement through PR Newswire.

On paper, that looks like exactly what every executive wants AI to do.

Faster resolution. Lower cost. Fewer repeat issues. More coverage. Better economics.

But a year later, the story became more complicated.

In May 2025, CX Dive reported that Klarna was turning back toward human customer service. The company said it wanted customers to always have the option to speak with a human. CEO Sebastian Siemiatkowski also acknowledged that cost had become too dominant as an evaluation factor and that the result was lower quality.

That does not mean Klarna's AI failed.

It means the operating model was incomplete.

The AI could handle a lot. It could probably handle more than skeptics expected. But customer service is not only about speed. It is also about trust, frustration, nuance, exceptions, and knowing when a person needs to feel heard instead of processed.

The lesson is not "AI does not work."

The lesson is that replacing humans entirely is harder than replacing visible tasks.

The Rehiring Pattern Is Becoming Obvious

Klarna is not just an isolated anecdote.

Gartner predicted in February 2026 that by 2027, 50% of companies that attributed customer-service headcount reductions to AI will rehire staff to perform similar functions, often under different job titles.

The same Gartner release included a useful detail: only 20% of customer service leaders surveyed had actually reduced agent staffing because of AI. Most kept headcount steady while supporting more customers.

That second point is the more interesting one.

The better companies may not be asking, "How many people can we cut?"

They may be asking, "How much more can this team now handle?"

That is a different question. It leads to a different implementation.

Fast Company also reported on what it called the "AI boomerang" effect. According to Robert Half research reviewed by Fast Company, 32% of hiring managers said their organizations eliminated a role or let someone go primarily because of AI or automation productivity gains, only to later rehire for that exact role.

The reasons were predictable:

• The role required institutional knowledge.
• The productivity gains were smaller than expected.
• The AI needed more human oversight than expected.
• Quality control became a bigger issue.
• The remaining team got overloaded.
• Business demand increased.

None of that should surprise an operator.

That is what happens when the system is understood too narrowly.

Layoffs Often Look Better Before They Happen

This issue is bigger than AI.

Companies have been overestimating the benefits of layoffs for a long time.

Harvard Business School's Working Knowledge published a piece on why layoffs can be bad business. The article summarizes Sandra Sucher's research and points to the hidden costs of layoffs: lower profitability, weaker innovation, lower productivity, and damage to trust.

That has been my experience with businesses too.

A layoff looks simple from the outside:

• Payroll goes down.
• The P&L improves.
• The company looks more efficient.

But inside the business, other things happen:

• Remaining employees slow down because trust is lower.
• Managers spend more time explaining decisions.
• Customers feel the service change.
• Institutional knowledge disappears.
• Exceptions start flowing back to the owner.
• Rework increases.
• Hiring costs come back later.

The spreadsheet captures the wage savings immediately.

It usually misses the drag.

And drag matters.

In a founder-led business, drag often shows up as the owner getting pulled back into work that the organization used to handle. The business may have reduced payroll, but it also reduced capacity, judgment, and resilience.

That may improve the P&L. But it is not the same thing as building a better business.

AI Should Remove Bad Work Before It Removes People

The first question should not be, "Who can AI replace?"

The first question should be, "What work should humans not be doing anymore?"

That includes work like:

• Rewriting the same email every day
• Searching across five systems for the same answer
• Manually summarizing meetings
• Copying data between tools
• Chasing missing information
• Cleaning up CRM fields
• Building the same report every week
• Answering repeated internal questions
• Drafting first-pass proposals
• Sorting inbound requests by hand

That is where AI and automation can be useful quickly.

Not because they make people irrelevant.

Because they make people less wasted.

There is a difference.

If a project manager spends two hours every day chasing updates and rewriting status emails, the highest-value move is probably not to eliminate the project manager. It is to remove the two hours of low-value work so the project manager can manage more work, catch problems earlier, communicate better with customers, and keep jobs moving.

If a salesperson spends half the week updating notes and preparing follow-ups, the answer is not automatically fewer salespeople. It may be better CRM hygiene, automated call summaries, suggested next steps, and faster proposal generation.

If a customer service team is drowning in repetitive questions, the first move should be a better knowledge base, AI-assisted answers, routing, summaries, and escalation rules. Then let humans handle the moments that require empathy, authority, or judgment.

This is where AI works best in real businesses.

It compresses the administrative layer around human work.

Productivity Is A Systems Problem

McKinsey has written about productivity through the lens of what it calls skill gaps, will gaps, and time gaps. In plain English, people are less productive when they don't have the right capability, are not engaged in the work, or spend too much time on low-value activity. The McKinsey piece is worth reading because it treats productivity as a system, not just an individual performance issue.

That framing is useful for AI.

If a team is underperforming, the answer is not always fewer people or different people.

It may be:

• The process is unclear.
• The tools don't talk to each other.
• The reporting is late.
• The team does not know what matters.
• Managers are stuck in meetings.
• The same decisions keep coming back to the owner.
• Everyone is busy, but too much of the work is low-value.

AI can help with some of that.

But only if the business starts with the workflow.

If you automate a bad process, you usually get a faster bad process.

If you automate around unclear ownership, you create faster confusion.

If you add AI without a source of truth, you create another place where people have to check the answer.

This is why I don't like tool-first AI implementation.

The question is not, "Where can we use AI?"

The question is, "Where is the business constrained?"

The Better Operating Question

Here is the question I would ask before any AI-related headcount decision:

What would happen if we kept the team and made each person 20% to 40% more productive?

You don’t need them to work harder. You need to remove friction that’s holding them back.

What if:

• Customer response time improved?
• Proposals went out faster?
• Managers had better weekly reporting?
• The owner answered fewer repeated questions?
• Sales follow-up became more consistent?
• Fewer handoffs got missed?
• Onboarding took days instead of weeks?
• The same team could serve more customers without burning out?

That is a growth conversation.

The layoff conversation is usually a cost conversation.

Both matter. But they produce different behavior.

If the goal is only cost reduction, AI becomes a justification for cutting.

If the goal is productivity, AI becomes a way to increase capacity.

One mindset shrinks the business.

The other can make the business stronger.

Cutting Staff To Fund AI Is Usually Backwards

There is another version of this mistake happening now.

Companies cut staff to "fund AI."

That sounds decisive. It sounds modern. It sounds like a board-level strategy.

But it often confuses budget with return.

Gartner said in May 2026 that among organizations piloting or deploying autonomous business capabilities, about 80% reported workforce reductions, but those reductions did not appear to translate into better ROI. Gartner's phrase is the one every owner should remember: workforce reductions may create budget room, but they don't create return.

That is exactly right.

You can cut people and free up cash.

That does not mean you created a better business.

A better business has:

• Better workflows
• Better information
• Better decision-making
• Better customer experience
• Better management cadence
• Better capacity
• Better margins
• Less owner dependency

AI can help with those things.

But only if it is implemented as part of the operating system.

What I Would Do First

Before downsizing because of AI, I would run a productivity audit.

Not a generic AI audit.

A practical operating audit.

I would ask:

1. What work is repeated every week?
2. What work requires too much owner involvement?
3. What work is slow because information is scattered?
4. What customer communication could be faster or more consistent?
5. What internal questions get asked over and over?
6. What reporting is still manually assembled?
7. What decisions are delayed because the right information is not available?
8. What tasks do good employees hate because the system around them is bad?

Then I would sort the work into four categories:

• Eliminate it.
• Clarify it.
• Delegate it.
• Automate it.

AI belongs mostly in the fourth category, but many businesses need the first three before the fourth works.

That is the part most companies skip.

They try to automate before they understand.

The Goal Is More Output Per Person

The goal should not be to keep every job exactly the same forever.

That is not realistic.

AI will change roles. Some work will disappear. Some teams will need fewer people in one area and more in another. Some employees will adapt quickly. Some will not.

But for most founder-led businesses, the smarter first move is not mass downsizing.

The smarter move is to increase output per person.

Make your existing team more capable.

Give them better systems.

Remove low-value work.

Build clearer workflows.

Use AI to help people make better decisions and move faster.

Then measure what happens.

If productivity improves and the business grows, you may not need fewer people. You may need those same people doing higher-value work.

That is the opportunity.

AI is not just a cost-cutting tool. It is a capacity tool. And in a good business, capacity should create growth.

Source Notes

• BCG transformation research: BCG Henderson Institute, June 2020
• Bain on growth versus cost: Bain, January 2023
• Klarna's original AI assistant announcement: PR Newswire / Klarna, February 2024
• Klarna's return to more human customer service: CX Dive, May 2025
• Gartner prediction on rehiring customer service staff: Gartner, February 2026
• Robert Half / Fast Company reporting on AI boomerang rehiring: Fast Company, June 2026
• Harvard Business School on hidden costs of layoffs: HBS Working Knowledge, October 2023
• McKinsey on productivity gaps: McKinsey, April 2024
• Gartner on AI layoffs and ROI: Gartner, May 2026