Capability Add vs. Work Removed: The ROI Question That Separates Real Value From Activity

Most AI projects do not fail because the technology is weak.

They fail because the organization confuses new capability with business value.

A team builds an assistant.

Employees use it.

Everyone reports that it is helpful.

Then the CFO asks a simple question:

“What changed financially?”

The room gets quiet.

This is the gap between AI activity and AI ROI in HR.

The Question Most AI Programs Avoid

The usual question is:

“What can AI help us do?”

That is a useful starting point.

It is not a business case.

The better question is:

“What work will no longer need to happen in the same way?”

That question creates pressure.

It forces leaders to examine the process, not the product.

It forces consultants to stop presenting adoption dashboards as transformation.

It forces practitioners to define the work before they automate it.

There are two different forms of value.

Capability Add

AI enables an additional activity that was previously difficult, slow, or impossible.

Examples include:

• Creating a new workforce scenario model.
• Giving managers a daily coaching assistant.
• Generating richer people analytics narratives.
• Producing personalized learning recommendations.
• Creating a new employee listening signal.
• Allowing a small HR team to conduct more analysis.

These capabilities can be valuable.

They can improve decisions.

They can create better employee experiences.

They can even produce new revenue or reduce risk.

But they also create a burden.

Someone must use the capability.

Someone must maintain it.

Someone must govern it.

Someone must review its output.

Someone must explain why it matters.

A capability add is not free value.

Work Removed

AI replaces or materially compresses work that the organization already performs.

Examples include:

• Fewer manual HR service tickets.
• Fewer handoffs between teams.
• Less duplicate data entry.
• Less time spent searching policy documents.
• Fewer recruiting coordination tasks.
• Less manual report preparation.
• Fewer routine questions sent to HR specialists.
• Less time spent checking, routing, and summarizing.

This is often easier to connect to ROI.

The organization already knows the work exists.

The baseline is visible.

The people doing the work are known.

The process cost can be estimated.

The before-and-after comparison is clearer.

In mature organizations, work removed often produces the fastest path to credible ROI.

Not because capability adds do not matter.

Because existing work already has a cost.

A Real Operator Moment

In one enterprise AI implementation conversation, the team was excited about a new HR capability.

The proposed solution could summarize employee questions, identify themes, and provide managers with richer insight.

On paper, it looked impressive.

The demo was smooth.

The language was polished.

The stakeholders agreed that it could help leaders make better decisions.

Then we mapped the workflow.

The insight would be generated after several existing steps.

A specialist would still collect the information.

Another person would still validate it.

A manager would still receive the summary through a separate channel.

The original report would still be produced because it was part of the operating rhythm.

The new AI output did not remove the old work.

It added another layer.

That changed the conversation.

The question was no longer, “Can we build this?”

The question became:

“Which existing decision or report disappears if we build this?”

Nobody had a clear answer.

So the team did not treat the prototype as transformation.

They treated it as an experiment.

That distinction matters.

A prototype can prove that AI is capable of something.

It cannot prove that the organization needs that capability.

The Hidden Cost of Capability Adds

A new AI capability usually creates at least five new demands:

1. Attention
2. Training
3. Governance
4. Maintenance
5. Integration

This is why “hours saved” is such a weak measure by itself.

If an AI tool saves a manager thirty minutes, but the manager now spends twenty minutes reviewing the output, the net gain is smaller.

If an AI tool creates a new report, but nobody stops producing the old report, there may be no work removed at all.

If employees use an AI assistant but still open the same ticket for final confirmation, the organization has added a front door without closing the old one.

The technology may be working.

The system is not.

Research from McKinsey shows that workflow redesign has one of the strongest relationships with measurable generative AI impact.

Yet only a minority of organizations report that they have fundamentally redesigned workflows around AI.

That is not a software problem.

That is an operating model problem.

A Simple Test for AI Transformation ROI

Before approving an AI use case, classify the value.

Ask four questions.

1. What work exists today?

Name the task.

Do not use broad language such as “improve employee experience.”

Say what happens.

For example:

• An employee searches three systems for a policy answer.
• The employee emails HR.
• HR routes the question to a specialist.
• The specialist checks the policy.
• The answer returns to the employee.
• The case is closed manually.

That is work.

It has steps.

It has owners.

It has delays.

It has cost.

2. What changes after AI is introduced?

Be specific.

Does AI:

• Remove a step?
• Reduce a handoff?
• Shorten the cycle time?
• Reduce the number of cases?
• Increase first-contact resolution?
• Allow one person to manage more volume?
• Improve the outcome enough to create measurable value?

If the answer is only “AI provides better insight,” you may have a capability add.

That is not a problem.

It simply needs a different business case.

3. What stops happening?

This is the most important question.

What report will no longer be created?

What queue will shrink?

What manual review will no longer be needed?

What meeting will become shorter?

What work will move from a specialist to a system?

What work will move from a person to a higher-value activity?

If nothing stops, the ROI case remains incomplete.

4. Where does the value appear?

Value can show up in several places:

• Cost avoided.
• Capacity released.
• Faster service.
• Lower vacancy cost.
• Fewer errors.
• Lower compliance exposure.
• Better retention.
• Higher quality decisions.
• More strategic work completed.

Do not pretend every benefit is a hard-dollar saving.

But do not hide behind soft language either.

State the mechanism.

Then decide how much confidence you have in the estimate.

The Work Removed Ledger

I use a simple way to make this visible.

Create two columns.

Capability added	Work removed
AI generates a workforce planning scenario	Analysts spend less time building the first draft
AI summarizes employee feedback	Fewer hours spent coding and grouping comments
AI answers policy questions	Fewer routine tickets reach HR specialists
AI drafts interview questions	Recruiters spend less time preparing for interviews
AI recommends learning content	Employees spend less time searching for relevant material

Then add three more fields:

• Current volume
• Current effort
• Owner after AI

The last field is often where the truth appears.

If the work still has the same owner, the same deadline, and the same required output, AI may have improved the process without changing the economics.

That can still be worthwhile.

But call it what it is.

Not cost removal.

Not transformation.

Productivity support.

What IBM’s AskHR Example Teaches

IBM’s AskHR case study is useful because it connects AI to work already happening.

IBM reports that AskHR handles more than 11.5 million employee interactions per year.

It also reports a 94% containment rate for common HR questions and a 75% reduction in HR support tickets compared with 2016 levels.

The important point is not the assistant itself.

The important point is the workflow change.

Employees ask questions.

The system resolves many routine requests.

Fewer cases move to human support.

The human team can focus on more complex work.

That is a clear work-removed story.

The AI capability matters because it changes the operating system around it.

Many vendors will show you a chatbot.

The better question is whether the chatbot removes a queue.

Capability Adds Still Matter

This is not an argument for efficiency-only thinking.

Organizations need new capabilities.

AI can help HR and business leaders do work they could not previously do at the same speed or scale.

A capability add may support:

• Better workforce decisions.
• Earlier detection of retention risk.
• More personalized manager support.
• Faster scenario planning.
• New forms of employee service.
• Stronger organizational learning.
• More informed transformation choices.

These outcomes can produce substantial value.

But the value often appears later and depends on behavior change.

A better workforce scenario has no value if leaders do not use it.

A manager assistant has no value if managers ignore it.

A new employee insight has no value if nobody changes a decision.

The capability must connect to a decision, action, or result.

Not more intelligence. More consequential action.

That is the standard.

The ROI Equation Changes

For work removed, the basic calculation is relatively direct:

Hours removed or compressed × loaded cost of work = potential value

Then subtract:

• Technology cost.
• Implementation cost.
• Training.
• Governance.
• Ongoing maintenance.
• Change management.
• Risk controls.

For capability adds, the calculation is harder.

You need to estimate the value of the improved outcome.

For example:

Improved decision quality × financial consequence of that decision = potential value

That may involve:

• Reduced regrettable turnover.
• Faster hiring into critical roles.
• Lower compliance exposure.
• Better productivity.
• Faster integration after an acquisition.
• Higher manager effectiveness.

The further the value sits from the AI action, the more assumptions you need.

That does not make the use case invalid.

It means you must label it correctly.

Fact is not the same as interpretation.

Interpretation is not the same as proof.

A serious AI transformation business case makes those differences visible.

The Builder’s Rule: Ship the Smallest Proof

Do not start with a grand platform.

Start with one workflow.

Map the current process.

Count the handoffs.

Measure the volume.

Identify the highest-friction step.

Prototype the smallest intervention.

Then ask what disappears.

This is where practitioners and consultants can create real value.

Not by producing another AI maturity score.

Not by creating a beautiful roadmap that nobody can operate.

Not by adding a vendor to an already crowded stack.

The valuable work is helping leaders decide:

• What should be automated?
• What should be redesigned?
• What should be removed?
• What should remain human?
• What should be built internally?
• What should be bought?
• What should never be introduced?

That is judgment.

That is where clients should pay more.

What This Means for CHROs and Consultants

For CHROs and CPOs, the opportunity is to move the AI conversation from enthusiasm to operating discipline.

Ask your team to bring you fewer demos.

Ask for more workflow maps.

Ask them to show:

• The current process.
• The proposed AI intervention.
• The work removed.
• The work added.
• The owner of the new work.
• The metric that will change.
• The decision that will change.
• The cost of being wrong.

For consultants, this is an opportunity to stop selling generic AI readiness.

Help clients identify where work is trapped in queues, handoffs, duplicated reports, manual reviews, and unclear ownership.

Then help them redesign the process.

The deliverable is not a slide deck.

The deliverable is a measurable change in how work gets done.

The Takeaway

AI capability does not automatically equal business value.

A new capability can create better decisions, new services, and stronger organizational performance.

But it also creates workload, governance, maintenance, and adoption demands.

Work removed gives leaders a clearer path to ROI because the organization can see the baseline.

The practical test is simple:

What work exists today?

What changes after AI?

What stops happening?

Where does the value appear?

Not every successful AI project will remove headcount.

Not every valuable outcome will appear as a cost reduction.

But every serious AI transformation should explain how the work changes.

The future will not belong to organizations that merely add AI tools.

It will belong to organizations that know what work to remove, what capability to build, and what human judgment to protect.

That is the real question behind AI transformation ROI.

Not, “What can AI do?”

But:

“What should no longer be done the old way?”

For more practical thinking on AI, work, and organizations, explore the work of the HR AI Institute or learn more about building with AI through the AI Agents course.