# AI Is Saving Time. Why Aren’t Costs Falling?

Employees feel faster. The company’s costs barely move. What happens to all the time AI saves?

By Paras Madan | 2026-09-16


AI is already saving a lot of time inside companies and we all can see it.

A meeting that needed another 30 minutes of notes and follow-ups can now be wrapped up almost immediately. Research that took hours can take minutes. Developers can write and review code faster. Support teams can handle more conversations without adding the same number of people.

Yet there is a strange gap between what employees experience and what leadership teams see. People feel faster yet the company does not always become meaningfully cheaper and that gap is becoming one of the more important questions for companies spending heavily on AI.

## What saved time actually gives a company

Morgan Stanley is a useful example.

Its AI which they call [Morgan Stanley Debrief](https://www.morganstanley.com/press-releases/ai-at-morgan-stanley-debrief-launch) can sit inside an advisor meeting, take notes, create a summary, identify follow-ups and draft an email.

Now, that is a meaningful productivity improvement.

But Morgan Stanley does not describe that saved half hour as an opportunity to remove half an hour of labour cost. It talks about advisors spending more time talking to clients, making decisions and growing their practices.

So if we see from a distance: The employee became more productive. The employee did not become 10% cheaper.

That distinction sounds obvious, but a lot of enterprise AI ROI is still discussed as though the two are interchangeable.

Imagine a company has 100 people earning the same salary and AI makes each of them 20% faster.

On a productivity spreadsheet, it is tempting to treat that as something close to 20 people's worth of savings. In reality, all 100 people are still employed and earning the same salary.

The company has gained capacity.

What happens next determines whether that capacity has any financial value. The team could serve more customers. It could ship more products. It could avoid the next 20 hires. It could bring work in-house that was previously outsourced. Or it could simply fill the saved hours with more work. All of these are different economic outcomes!

This is why converting every hour saved into a dollar figure can create a very misleading picture of AI ROI.

Most companies do not have employees sitting around waiting for work.

When one task becomes faster, another task usually takes its place. The salesperson who saves an hour on research might make more calls. The engineer who finishes one feature earlier starts the next one. The manager who no longer spends an hour summarising meetings gets through a longer backlog.

This is still valuable. In a growing company, additional capacity can be worth far more than cutting costs.

But it also explains why an organisation can have thousands of employees using AI every week without seeing an equivalent decline in operating expenses.

The saved time has been consumed by the company. It has not disappeared from the cost base.

## Morgan Stanley and Klarna made different choices

- Morgan Stanley represents one approach: keep the people and increase what they can do.

- Klarna pushed much harder in the other direction.

In early 2024, Klarna said its AI customer-service assistant had handled 2.3 million conversations in its first month, equivalent to the work of roughly 700 full-time agents. It also estimated that the system could contribute around $40 million in profit improvement during 2024. [Klarna’s original AI customer-service results](https://www.prnewswire.com/news-releases/klarna-ai-assistant-handles-two-thirds-of-customer-service-chats-in-its-first-month-302072744.html)

This is much closer to what companies usually imagine when they talk about AI cost savings. Productivity is converted into fewer hires and lower labour requirements.

But Klarna also showed why this is harder than it looks.

By May 2025, CEO Sebastian Siemiatkowski said the company's push for AI-driven cost cutting in customer service had gone too far. Klarna began making sure customers could reach human agents again. [Bloomberg on Klarna bringing humans back into customer service](https://www.bloomberg.com/news/articles/2025-05-08/klarna-turns-from-ai-to-real-person-customer-service)

The two companies are solving different problems.

- Morgan Stanley is asking: How much more valuable can the same advisor become?

- Klarna tested a harder question: How much of the existing work can the company stop paying humans to do?

The second approach produces much more visible savings. It also forces companies to find the point where efficiency starts affecting the customer experience.

## A faster task is not a faster workflow

There is another reason AI productivity does not translate cleanly into company-wide savings as most jobs are made up of workflows, not isolated tasks.

Suppose AI reduces the time needed to prepare a contract from three hours to 30 minutes. The contract may still need legal approval, customer feedback, security review, negotiation and a signature. Now, one step became dramatically faster but the total process did not become six times faster.

The same thing happens in engineering. Code can be generated faster, but it still needs to fit the system, pass tests, survive review and eventually reach production. Sales research can take minutes, while the buying process still takes six months.

AI often removes a bottleneck and exposes the next one.

For leadership teams, this creates a different problem. Buying a better model will not fix a workflow whose remaining constraints sit somewhere else in the organisation.

## The cost of making AI usable

There is also a tendency to calculate the value of AI while ignoring what it costs to make AI usable inside a large company.

- The model itself is only one part.

- There is integration with existing systems, security review, access controls, data preparation, monitoring, evaluations, governance and training. In higher-risk workflows, companies often keep a human review layer as well.

Some of these costs decline over time. Others grow as AI spreads across the organisation. Now, this creates an unusual situation where AI can make each employee more productive while the company's technology bill rises.

## What changed after the time was saved?

This is where I think leadership teams need a different way of measuring AI.

Time saved is an operational metric. It tells you whether the tool is doing something. It does not tell you whether the company became better economically. However, the more useful questions come after the time has been saved:

Did we avoid hiring another ten people? Did the same team handle significantly more customers? Did salespeople create more pipeline? Did engineers ship more valuable work? Did we reduce external agency or contractor spend? Did errors, refunds or support escalations fall? Did the customer experience improve enough to affect retention or revenue?

If none of these things change, AI may still be making employees' jobs easier. That has value. But it is a different claim from saying AI has produced financial returns.

## Better tools are only the first step

The largest AI gains will probably come when companies stop adding AI to the existing organisation and start changing the organisation around what AI can now do.

That might mean one person owning a process that previously required three handoffs. A support system might resolve routine problems automatically while humans handle the difficult cases. A team might increase output without increasing headcount for the next two years.

Now, those are structural changes but they are also much harder than deploying a copilot. Buying software can happen in weeks but changing responsibilities, approval processes, hiring plans and customer workflows takes much longer.

That may be why the productivity gains from AI are appearing faster than the financial gains.

The first phase of enterprise AI has mostly been about giving people better tools.

That phase is already producing visible results. As we saw, Morgan Stanley's advisors can spend less time on administrative work. Klarna's AI can resolve a large share of routine support work.

But, the harder phase comes after that:

- Leadership teams have to decide what to do with the capacity they have created.

- Sometimes the right answer will be fewer people. Sometimes it will be more customers, faster product development or higher-quality service. Often it will be a combination.

But the financial value does not come from the hour that AI saved.

It comes from what the company changes because that hour no longer needs to be spent.