AI Is Everywhere. Its Business Impact Is Harder to Find.

An email drafted in thirty seconds. A meeting summarized automatically. A sales proposal put together in minutes.

For people who use AI regularly, the gains are easy to see. One question is much harder to answer: where do those gains show up in the company’s results?

The connection between an employee saving time and the business generating more revenue or a better margin is less straightforward than it looks.

That gap recalls a long-standing puzzle in economics: the Solow paradox. It also helps explain a very current challenge for sales leaders — turning a useful technology into measurable performance.

01The Solow paradox: why the gains take time

In 1987, economist Robert Solow captured the computer age’s productivity puzzle in a line that became famous: “You can see the computer age everywhere but in the productivity statistics.”

Companies were buying computers. Their use was spreading. Yet the expected acceleration in productivity remained hard to detect.

In its most common form, productivity measures output per hour worked. At company level, the question is much the same: how do you create more value from the resources you use?

Installing computers was not enough. Businesses also had to change working practices, train people, reorganize information flows, and develop new services.

Economists Erik Brynjolfsson, Daniel Rock, and Chad Syverson have since developed this argument: major technologies require complementary investment in skills, processes, and organization. Those investments can come well before the measurable returns, producing a productivity “J-curve.” [1]

AI raises a similar question today. We know it can speed up individual tasks. We understand far less about what it takes to translate that speed into value across a business.

The history of computers offers a possible explanation. It does not tell us when AI’s wider benefits will arrive, or how large they will be.

02Time saved is only the start of the equation

Consider a sales rep who spends an hour preparing for a customer meeting. With AI, the preparation takes twenty minutes.

The gain looks obvious: forty minutes saved. But assessing the economic benefit requires a few more questions.

Is the preparation just as useful? Is the information accurate? Does the rep use the extra time to develop an opportunity, qualify a prospect more thoroughly, or meet another customer?

Three different outcomes need to be distinguished:

• The task-level gain: preparation takes less time.
• The team-level gain: more worthwhile meetings are prepared and held.
• The business outcome: revenue or margin increases, or costs are genuinely avoided.

Moving from one level to the next takes organizational decisions. Time saved creates available capacity. Its value depends on what the business does with it.

03Why the gains get lost inside organizations

A company can find dozens of uses for AI while leaving the obstacles to performance untouched.

A proposal gets written faster but still waits days for internal approval. A meeting summary is generated automatically, but nobody follows up on the agreed actions. A prospecting campaign produces more messages, aimed at the same poorly chosen contacts.

In each case, AI speeds up one step without addressing the main constraint. Four issues deserve particular attention.

Processes that have not changed

Adding an assistant to a workflow does not remove unnecessary approvals, unclear ownership, or scattered information.

Look at the whole workflow. Where does work sit waiting? Where do errors occur? Which step actually limits the team’s capacity?

Data that is not good enough

A sales recommendation is only as useful as the information behind it.

If the CRM holds stale opportunities, incomplete account records, or gaps in customer history, people will have to check and supplement the tool’s answers. That work eats into the expected gain.

Costs that are easy to underestimate

The subscription is only part of the bill.

Integration, training, maintenance, review, and corrections all count. A task can become faster for the person doing it while creating extra work for their manager or colleagues.

Metrics that track usage, not results

Activated licenses, prompts submitted, and documents generated can tell you whether people are using AI. They tell you far less about performance.

A team can use AI heavily without improving its results. Another can create substantial value from a narrow use case aimed at one specific problem.

04There is real evidence — in specific settings

None of this means AI produces no tangible gains.

A field study of customer support found an average productivity increase of roughly 14%, measured as issues resolved per hour. The gains were larger for less experienced employees. [2]

The finding is useful because it is specific: a defined activity, a tool embedded in the work, and an observable measure.

It does not justify promising the same improvement to every team. It does show what useful evidence looks like.

Equally, the absence of an unmistakable acceleration in economy-wide statistics does not prove that every use case is ineffective. A task, a company, and an economy are different units of measurement.

05In sales, more activity is not better performance

For sales leaders, the danger is a familiar one: confusing how much work gets done with how effective that work is.

AI lets a team write more emails. If those emails do not generate relevant responses, the extra activity adds little value.

It lets reps prepare more proposals. If the opportunities are poorly qualified, it mainly speeds up the production of offers that will never close.

Choosing the right measures is therefore critical.

These outcomes also depend on market conditions, the prospects involved, and the team’s working practices. A meaningful comparison is still needed to assess AI’s contribution.

06Rethink the work before adding AI

To get more value from AI, start by examining the processes it is supposed to improve. Some need simplifying. Others need redesigning — or removing altogether.

A report nobody uses does not need to be generated faster. It can go. A proposal that requires five levels of approval calls for a review of the approval rules first. A rep entering the same information into three systems needs a simpler workflow.

Automating an unnecessary task mainly helps it survive.

Before starting an AI project, step back and ask three questions:

• Remove: does this task still create value for a customer or the business?
• Simplify: can we reduce the steps, approvals, or data entry involved?
• Redesign: how should the work be organized given the capabilities now available?

Only then decide where AI can make a useful contribution.

Take the follow-up after a sales meeting. The obvious use case is an automatically generated summary. A more ambitious approach looks at the entire workflow: what information needs to be retained? Who needs it? What actions should follow? Which manual entries can disappear?

The objective becomes more reliable follow-up, fewer manual steps, and clear ownership of next actions. The summary is a means to that end.

07Test the new workflow and measure the impact

Once the process has been redesigned, test AI’s contribution within a limited scope:

1. Define the intended outcome: a shorter turnaround, better qualification, or fewer errors.
2. Establish a baseline, including the time spent by everyone involved.
3. Test the new workflow against a meaningful comparison.
4. Calculate the net benefit, including training, review, and correction costs.
5. Scale, adjust, or stop based on what the results show.

Decide how the freed-up capacity will be used: more time with customers, better opportunity management, or lower costs. That decision is part of the project, not something to leave until afterwards.

08AI’s value depends on how the business works

The Solow paradox reminds us that adopting a technology and improving performance do not necessarily happen at the same pace.

For sales leaders, AI’s benefits also depend on their willingness to challenge established habits: dropping a report that no longer serves a purpose, shortening an approval chain, reassigning responsibilities, or eliminating duplicate data entry.

Those changes require management decisions. Buying a tool is not a substitute for making them.

Getting the most out of AI means being willing to change the work itself — including removing some tasks before trying to automate them.

Which process in your sales team should be simplified or removed before you add AI?

09Sources

[1] Erik Brynjolfsson, Daniel Rock, and Chad Syverson, The Productivity J-Curve: How Intangibles Complement General Purpose Technologies, NBER, 2018.
Read the NBER paper

[2] Erik Brynjolfsson, Danielle Li, and Lindsey R. Raymond, Generative AI at Work, NBER working paper, 2023. The 14% figure refers to this version.
Read the 2023 working paper (PDF)