by

Users giving thumbs down to Copilot

Microsoft’s Copilot is meant to be the centrepiece of its AI story, but it is behaving like a half-finished beta demo.

The software giant has promoted Copilot as a ChatGPT alternative just as its once cosy relationship with OpenAI appears less exclusive, yet users continue to encounter basic issues. Current and former staff who worked on Microsoft’s AI kit say the branding is confusing and the various versions do not play nicely together.

It has a tendency to break down and shut down completely. One translation company reported that it was working fine until they started using the paid-for model, and it then entered an “I will think about it” mode indefinitely.

Only a small proportion of subscribers to Microsoft’s enterprise suite use Copilot, and the share who pick it ahead of Google’s Gemini or other tools has slipped in recent months, according to data reviewed by the Wall Street Journal.

The pressure is on because Microsoft chief executive Satya Nadella has tied Copilot to his plan to turn the company into an AI-first outfit, echoing the cloud-first pivot he pulled off about a decade ago. Current and former executives said Copilot sits near the top of Nadella’s to-do list.

Investors have been jumpy, too. Microsoft shares tanked after its earnings report last week stoked worries that growth in Azure, its key cloud-computing unit, is slowing and that its AI revenue still leans on OpenAI while Copilot remains a question mark. The shares fell nearly three per cent on Tuesday as software stocks slid on fresh nerves that AI tools could make pricey enterprise subscriptions feel less necessary.

Nadella wrote: “We have moved past the initial phase of discovery… the industry is entering a phase where we are beginning to distinguish between ‘spectacle’ and ‘substance.’”

Even if Copilot is lagging in the chatbot scrap, Microsoft is still pulling in many billions from AI-driven cloud demand, keeping it among the world’s most valuable companies. Analysts also figure it can close the gap because Microsoft productivity software sits in front of hundreds of millions of corporate users, a captive crowd for upsells.

Vole has built several Copilots into its products, including 365 apps such as PowerPoint and the GitHub developer platform. It also sells a consumer-facing version through Edge and via an app.

Internally, the Copilot universe is split into three buckets. There are enterprise tools sold to companies and professionals, Copilots for developers and IT staff, and the consumer chatbot overseen by Microsoft AI chief executive Mustafa Suleyman.

Microsoft hired Suleyman in 2024 to run consumer AI and to build models that can compete with OpenAI and others. So far, Microsoft has leaned on OpenAI and its rival Anthropic to power different Copilots, while insisting it will use the best models available.

Last week, Microsoft reported that it had sold 15 million corporate Copilot “seats” within its 365 productivity business, which has a base of more than 450 million paid seats. Late last year, it also claimed more than 150 million monthly active Copilot users across its first-party platforms, while Google’s Gemini has more than 650 million monthly users, and ChatGPT has about 900 million weekly active users.

Previously unreported numbers suggested Copilot is losing mindshare among the people already paying for it. From July through late January, the percentage of Copilot subscribers who use it as their primary option fell from 18.8 per cent to 11.5 per cent, based on a Recon Analytics survey of more than 150,000 respondents in the US. Over the same period, the share of paid users picking Google’s Gemini first rose from 12.8 per cent to 15.7 per cent.

Respondents who ditched Copilot said they were getting better quality elsewhere, with some blaming a rough user experience and restrictive usage limits. Recon’s data shows that workers with access to Copilot, ChatGPT and Gemini pick ChatGPT and Gemini at a higher rate than Copilot.

For some firms, the issue is not buying seats, it is using them. Citi Research analysts recently wrote that some companies are using only about 10 per cent of the Copilot subscription seats they are paying for, pointing to “Disorganized data silos” as a problem.

Microsoft’s chief marketing officer of AI at work, Jared Spataro, said daily active usage of 365 Copilot has grown 10-fold year-over-year and is outpacing the growth of other 365 enterprise offerings. He challenged Citi’s note, saying: “The pace of growth that we’re seeing is unlike anything we’ve seen before,” but he declined to share specific figures.

Customer surveys commissioned by Microsoft have found that users are baffled by the multiple Copilot flavours, according to people familiar with the matter. Some users have moaned for ages about Copilot being forced on them, popping up everywhere from documents to the browser.

Staff point to a lack of a cohesive experience across the different Copilots, a pain point Nadella has flagged before. About a year ago, Nadella reportedly sent a frustrated email to Microsoft executive vice president of experiences and devices Rajesh Jha after Nadella asked the enterprise version of Copilot in Edge to help with a public webpage, only for it to fail to respond to the prompt.

The specific problem Nadella raised was resolved, but similar interoperability issues keep recurring for users. In recent weeks, a new Anthropic AI product, Claude Cowork, has been praised for operating within and across 365 applications in ways that Copilot users report they struggle to achieve. The rollout of new Cowork features was cited as a factor behind Tuesday’s slide in software stocks, and Microsoft product leaders have discussed Cowork while working internally on a similar idea.

Spataro said the split between 365 Copilot and the consumer version is deliberate, so users can keep work and personal information separate. Some employees reckon organisational silos between Suleyman’s consumer-focused group and enterprise teams have made a unified vision harder to land.

Microsoft’s efforts to train proprietary models have also been constrained by limited computing capacity, with the company rationing server time to keep Azure capacity available for OpenAI and other customers. On several benchmark tests, Microsoft’s flagship in-house model ranks well below rivals.

Microsoft executive vice president of the cloud and AI group, Scott Guthrie, noted that Suleyman’s team only came into existence in 2024. He said the long lead time for building cloud infrastructure helps explain why the team only recently started to get the computing capacity it needed.

In its latest earnings report, Microsoft said it is allocating more computing power to improve Copilot because it has gained confidence it can monetise the products, but UBS said shareholders are not convinced by the trade-off.

 

 

Latest articles

Share

Featured articles

Hot topics

No results found.

Latest reviews