Opinion A year ago it looked as if the world could be Microsoft's oyster. The software giant dominated the enterprise, was catching up to cloudy rivals, and then managed to purchase forty-nine percent of the for-profit subsidiary of ChatGPT creator OpenAI.
Having secured a stake in the leading purveyor of generative AI, it started to build it into products that attracted enormous attention, like Bing. Microsoft dangled the enticing prospect that its long-suffering search services could improve to the point at which they would challenge arch-rival Google.
It hasn't happened. And now it looks like it won't – that Google will deliver a more useful AI, faster.
Microsoft's missteps started in May 2023 when it launched a company-wide strategy that could have been titled "Copilot all the things!" Windows 11 got a Copilot assistant and a Copilot key. Edge got a Copilot panel, while Microsoft 365 got the full treatment.
It has become increasingly clear over the last few months, though, that those AI projects have been – to say the least – underwhelming. Enterprises of every shape and description are making decisions on whether to adopt pricey Microsoft 365 Copilot AI integrations into their office suites, and CFOs grill CIOs about ROI: Is it really worth doubling the monthly Microsoft tax?
The answer has mostly been a shrug of the shoulders. Having Word or Outlook write for you, or PowerPoint create slideware on your behalf, is nice in theory. But few have adopted those features, or learned how to make them save the sort of time that would justify the investment.
Though Copilot is touted as an ideal assistant to help with calculations, Excel users quickly learned that a language model trained on billions of words of English has no effective grasp of the very different paradigm of a spreadsheet. Yes, Copilot can compose a quick formula, but it can't operate within the context of a worksheet with anything like Word's facility to parse the meaning of a text document.
To bridge this fundamental gap in capability, Microsoft Research recently published a paper on SpreadsheetLLM – a from-the-ground-up approach which looks as though it might yield an AI that can "think" in Excel. Or at least, maybe a few years from now. For now, the big improvement to Excel is Python, not AI.
There is one shining star in the Microsoft 365 Copilot firmament: Teams. The one app everyone loves to hate – created to undercut Slack – Teams became an indispensable hybrid working tool during the pandemic. Copilot for Teams acts as the perfect secretary – taking transcripts of meetings, generating a summary, complete with action items. It even mails the participants.
It's very helpful, but with a caveat: you might not want your every meeting recorded and transcribed. Anecdotally, a few of those transcripts have ended up as evidence in lawsuits.
Thanks for that, AI.
That gotcha points to the other reason Microsoft finds itself dead in the water in its efforts to drive Copilot into the enterprise: making it work well requires access to all of an enterprise's documents. That means effectively trusting your entire set of enterprise documents to OneDrive/SharePoint.
Microsoft should know that’s not going to happen. Storage and analytics vendors spent the early 2000s suggesting their customers would go on metadata creation sprees to tame unstructured data and make it useful. That dream – and a dozen others aimed at organizing corporate data – died.
To this day, very few enterprises have done the sort of deep data classification and sharing permissions necessary to create a safe-enough enterprise-wide environment for sharing sensitive documents. Nor have they trained their staff in how to work with data classification and data sharing with AI systems.
By throwing Copilot at all the things, Microsoft tried to force enterprises into a future for which they found themselves wholly unprepared.
- AI has colonized our world – so it's time to learn the language of our new overlords
- AI stole my job and my work, and the boss didn't know – or care
- Big Tech's eventual response to my LLM-crasher bug report was dire
- Microsoft's Recall should be celebrated as the savior of SMEs and scourge of CEOs
While most enterprises would admit they need to adopt better data classification and sharing practices, few seem enthusiastic about making that investment a priority just to get a few extra features in Microsoft 365.
They'll need a better justification – a killer app that clears away the logjam of legacy business practices and IT policies. Microsoft isn't offering that – nor, from the latest upgrades, does it look to be on the cards.
But Google may have a killer app worth the effort: the NotebookLM, which looks to be the first RAG tool that can really help enterprises find the needles in the haystacks of their documents.
NotebookLM is marvelously simple. Feed it documents and it will let you query them with prompts. Results are pleasing, and arrive swiftly. Ingestion tools could be better, but the results are strong enough to make it worth the effort.
With NotebookLM Google has an opportunity to invent "search for the age of AI" Not some AI-poisoned version of Google Search, but a new tool, designed to make sense of enterprise data – not just help you make more of it.
That's something that Microsoft could have offered but didn't, despite spending decades trying to make SharePoint the place you store everything.
Remember also that Microsoft had to catch up and build decent search into Windows after 2004's Google Desktop, which searched local documents, made the OS look clunky.
However, before we shed a tear for poor Microsoft, bear in mind that after its current sale of shares to investors, OpenAI carries a valuation of $157 billion. That makes Microsoft's stake worth nearly $80 billion. Even as Microsoft disappoints with AI, it has found a way to win. ®