Blogger reviewing WordPress analytics beside a banking employee working with Copilot and Excel.

AI Helps Me Run Two Blogs. Why Are Its Economic Gains So Hard to Measure?

I use artificial intelligence to manage my WordPress blogs. In banking, I see colleagues using Microsoft Copilot for everyday office work. The technology is useful, but measuring its economic value is another matter.

I open my WordPress dashboard and examine how readers are responding to an article. Some posts continue attracting visitors long after publication. Others need attention because the information has changed or their traffic has declined.

Artificial intelligence helps me investigate these patterns. I manage two WordPress.com blogs. AI assists with analysing posts, identifying possible improvements and reviewing articles when new information becomes available. I still decide what to change and whether the evidence supports the recommendation.

At the bank, I see another side of the same development. Many employees use Microsoft Copilot to draft emails and manage Excel files. Yet economists continue to debate how much productivity AI is actually creating.

AI Has Changed How I Manage My Blogs

A published article does not stop needing attention simply because readers can find it online. New facts emerge. Search traffic changes. Sometimes a reader leaves a comment that exposes a weakness in the original argument.

AI now helps me bring parts of the editorial process together. I can ask it to examine traffic information I provide and suggest possible reasons for a change. It can help identify outdated claims and recommend areas that need further investigation.

But AI can offer a convincing explanation for declining traffic without knowing the actual cause. A post may lose visitors because search demand has fallen. Google may have changed how it displays results. Another website may have published a better article.

I still need to examine the evidence. In my earlier article about the cost of using ChatGPT Work, I described how quickly a research and publishing workflow consumed part of my available usage allowance.

A tool may reduce the time required for a task while introducing another cost. For an independent blogger, the subscription price matters. So does the time spent checking the machine’s work.

I cannot claim that AI has increased my blog traffic by a particular percentage. I have not established that through a controlled comparison. What I can say is that AI has changed how I approach editorial work.

I See the Same Development in Banking

My banking experience gives me a different perspective. I work with SWIFT and cross-border payments. The work requires attention to payment instructions, message formats and the rules governing international transactions.

A payment message is not an ordinary piece of correspondence. The information must be accurate. The relevant controls must be followed. A mistake can delay a transaction or create additional work for the institutions involved.

Alongside these specialised operations, banks also handle large amounts of routine office work. Employees prepare emails, review spreadsheets and organise information for colleagues or managers. Many of my colleagues use Microsoft Copilot to assist with such tasks.

A clear first draft may take less time to prepare. A spreadsheet formula may become easier to understand. But banking exposes the limits of that convenience.

Suppose an employee asks AI to explain a problem with a cross-border payment. The system might produce a polished explanation using the right financial terminology. That does not establish that the explanation is correct.

A payment may involve an incorrect account identifier, a message-format issue or a compliance-related restriction. The actual cause has to be established from authorised records and applicable procedures. AI-generated text cannot replace that verification.

Banking information cannot simply be copied into an external AI service because an employee wants a faster answer. Any use of AI must respect the institution’s approved systems and information-security rules.

Saving time on an email is useful. It does not necessarily mean the bank has processed more transactions or reduced operational risk. Those outcomes require separate evidence.

Researchers Have Found Genuine Productivity Gains

In 2025, researchers Erik Brynjolfsson, Danielle Li and Lindsey Raymond published a study in The Quarterly Journal of Economics. They examined a generative AI assistant among 5,172 customer-support agents.

Workers with access to AI assistance resolved an average of 15% more customer issues per hour. Less experienced employees benefited more than highly experienced colleagues. The published research provides evidence that AI can improve performance in a real workplace, but not that every employee becomes 15% more productive.

A second study examined 7,137 knowledge workers across 66 firms. During the later part of the experiment, regular AI users spent about two fewer hours per week on email. Yet researchers did not detect broader changes in the quantity or composition of employees’ tasks resulting from individual access to AI.

The NBER study reveals an important limitation. An employee can save time without the organisation immediately changing how it operates. Whether saved time becomes additional economic output depends on what happens next.

Robert Solow Asked a Similar Question About Computers

In 1987, economist Robert Solow observed that computers were spreading through the economy while their expected contribution remained difficult to identify in productivity statistics. His observation became known as the productivity paradox.

Artificial intelligence has revived the argument. A company may introduce AI into several departments. Employees begin using it to prepare documents and complete routine tasks. But does the company serve more customers or reduce the cost of delivering a service?

In March 2026, researchers at the Federal Reserve Banks of Atlanta and Richmond published findings from a survey of corporate executives. Companies reported that AI had increased productivity by 1.8% in 2025. But gains implied by AI-related revenue and employment effects were considerably smaller.

The researchers suggested that some benefits might take time to appear in revenue. Their Federal Reserve analysis identifies a gap between reported improvements and financial measurements. It neither proves that AI is failing nor guarantees that every reported benefit will become measurable.

A Faster Employee Does Not Automatically Mean a More Productive Bank

Consider a hypothetical banking employee who spends 30 minutes preparing a routine report. With AI assistance, the first draft takes 10 minutes. But reviewing figures and correcting errors takes time too.

Suppose the employee uses the remaining time to investigate a payment discrepancy. That work may prevent a costly mistake. The value of preventing an error does not always appear as additional revenue. Sometimes successful banking operations are defined by problems that never occur.

A support officer may spend hours investigating a technical problem affecting payment processing. Resolving it restores normal operations, yet the contribution may not be visible in a simple revenue-per-employee calculation.

AI might make technical documentation easier to understand or help prepare a clearer explanation for colleagues. The benefit still depends on accuracy and the employee’s ability to apply the information correctly.

The Cost of AI Must Also Be Counted

AI tools require subscriptions or usage allowances. A bank faces a larger version of the same calculation. Software licences cost money. Employees need training. Information-security controls and human review also require resources.

A tool that saves five minutes on a task may be worthwhile if employees perform that task hundreds of times. It may be less attractive when the task occurs occasionally and the output requires extensive checking.

This is where the debate about expensive AI data centres becomes relevant. Technology companies invest heavily because they expect demand for AI computing to grow. Their customers may find AI useful, but usefulness does not guarantee that infrastructure investors will earn adequate returns.

During the dot-com boom, many investments failed even as the internet continued to spread. AI may become an ordinary part of office work while some companies financing its infrastructure struggle to recover their investments. Those are not contradictory outcomes.

Pakistan Should Measure AI by the Work It Improves

Pakistan’s banking sector provides a useful setting for examining AI adoption. Banks already depend on digital systems for processing payments and managing operational information.

A bank should not judge an AI system merely by how quickly it produces an email. It needs to know whether the output is accurate and whether the employee saves time after the necessary checks.

For cross-border payments, that distinction matters. A wrong instruction can create additional investigation work. A confident but inaccurate explanation may delay identification of the real problem.

I would also want to know whether AI helps employees understand technical procedures more effectively. The customer-support research suggests such learning effects are possible. It does not establish that the same gains have already occurred in Pakistani banks.

I Want to Measure What AI Actually Changes

For my blogs, I could compare the time needed to research and update similar articles with and without AI assistance, including source checking and corrections. Traffic would require a separate test, because search rankings change for reasons unrelated to AI.

In banking, I would examine comparable tasks before and after the introduction of an approved AI tool. Time saved would matter, as would accuracy. For sensitive operations, the cost of an error could outweigh faster drafting.

These measurements would not settle the national productivity debate. They would provide a better foundation than assuming that every AI-generated document represents an economic gain.

The Missing Hour

When I review my WordPress analytics, AI can help me decide which article deserves attention. When a banking employee uses Copilot to prepare a routine email, the tool may reduce drafting time.

Research shows that AI can improve performance in particular jobs. It also shows that gains vary and company-level results may take longer to appear.

I want to know what happens to the work. If an employee saves an hour, does the organisation use it to serve more customers? Does it improve accuracy? Does the employee simply receive another assignment?

A bank can measure how many transactions it processes. I can measure how many readers visit my blogs. Neither figure, by itself, tells us how much value AI has created.

AI may save the hour. The economic result depends on what we do with it.

Leave a Reply