Cortana

Cortana was Microsoft's voice-and-text assistant, introduced in 2014 and retired in stages through 2023. It was named after the artificial intelligence in the Halo game series, and it was designed to span Windows Phone, Windows, and later iOS, Android, and the Xbox console 1.

Cortana handled reminders, calendar lookups, web answers, and natural-language device search. It never matched the usage of its competitors, and Microsoft progressively withdrew it as the company reorganised its assistant work around Copilot. The standalone Windows app was retired during 2023, and the remaining Microsoft 365 integrations followed later that year 2.

This entry covers the product's development, its platforms and capabilities, the economics of Microsoft's phone business, and the competing explanations for its retirement.

Cortana was announced at Microsoft's Build conference in April 2014 and shipped with Windows Phone 8.1. The name came from the artificial intelligence character in the Halo series, and Microsoft presented the product as a persona rather than a search box. The character's voice actor, Jen Taylor, recorded the assistant's lines, a decision that gave the software a continuity with the games that few corporate products attempt 1.

The framing was deliberate. Cortana was introduced as a personal assistant that learned a user's interests, stored them in a notebook, and used them to personalise answers. The wake phrase, Hey Cortana, made the assistant reachable whenever supported hardware was listening. Microsoft launched it first in the United States in beta, then in the United States, United Kingdom, and China, with other markets to follow 1.

That launch plan set the pattern for the product's later problems. The assistant's language coverage and feature set were uneven outside the first markets, and the persona raised expectations of conversational competence that the underlying intent-based system could not always meet. The name gave the product instant recognition among Halo players, but it also tied a general-purpose tool to a fictional character whose abilities it did not share.

The product spread across Microsoft's estate: Windows 10 in 2015, iOS and Android the same year, Xbox One in 2016, and a smart speaker built with Harman Kardon in 2017. Each platform was an attempt to establish the assistant where users already were, rather than waiting for them to come to Windows.

The strategy depended on integration with services Cortana could reach. Calendar, mail, contacts, and the Windows shell were inside Microsoft's control; music, mapping, and ride-hailing were not, and partnerships for those integrations were uneven. The Cortana Skills Kit, announced in December 2016, was meant to close the gap by letting third-party developers add commands through the Microsoft Bot Framework 3.

Microsoft said the kit would reach 145 million Cortana users across Windows, Android, iOS, Xbox, and partner devices. That figure was a company claim, not an independently audited count, and it did not translate into a developer ecosystem comparable to those around Alexa and Google Assistant. Windows Phone's small market share was the structural problem: an assistant built first for a minor mobile platform could not reach the audience its rivals were competing for 4.

Cortana could set reminders attached to people, places, or times, read and send messages, check a calendar, report weather and news, and search the web through Bing. On Windows it could open applications, find files, and answer questions about the device. The notebook stored preferences and interests so that answers could be personalised across sessions 1.

The skills kit let developers add commands, following the model Amazon had established with Alexa. Skills were built as bots and registered with Microsoft's developer service, which handled invocation and account linking. Take-up was limited, partly because developers had to support a platform whose user base was smaller than its rivals' and whose strategy shifted repeatedly 3.

Integration with Microsoft 365 proved the most durable element. In Teams, Outlook, and Microsoft 365 mobile apps, Cortana could join meetings, read email aloud, and answer work queries using calendar and directory data. Those features were delivered through enterprise-grade services with compliance commitments, which made them more defensible than the consumer assistant 5.

Cortana was built from several Microsoft systems rather than a single model. Speech recognition converted audio to text, a language-understanding service mapped the text to an intent and its parameters, and Bing supplied web answers, entity information, and local results. A dialogue manager decided whether to answer, ask a follow-up, or hand the request to an app 1.

This architecture was standard for assistants of its generation. It was reliable within supported domains, because each skill declared the intents it could handle, and brittle outside them, because an unrecognised phrasing fell back to web search. The system could not reason across domains or write an answer from scratch; it selected and filled a template 3.

Personalisation ran through the notebook, which stored explicit interests and inferred preferences. The design gave users some control over what the assistant remembered, and it allowed Microsoft to keep personal data within an account rather than in a per-request context. The trade-off was that the assistant's knowledge was only as good as the categories the product team had defined, and updating it required engineering work rather than a new model 5.

Cortana's fortunes were tied to Microsoft's phone business. Microsoft acquired Nokia's handset unit in 2014 for about 7.2 billion US dollars, then wrote down about 7.6 billion dollars and cut 7,800 jobs in 2015 as the business failed to gain share. Its smartphone share was about 3 per cent, against a mobile market dominated by Android and iOS 6.

That position was fatal for a consumer assistant. Assistants improve with usage, and usage follows the device people carry. Microsoft had a large Windows PC installed base, but PC interactions were shorter and more task-specific than phone interactions, and the assistant's most valuable context, location and messaging, lived on phones. The Harman Kardon Invoke speaker was a partner-built device sold at a premium, not a subsidised platform play of the kind Amazon used to put Alexa in homes 4.

The business model was never explicit. Cortana had no subscription of its own; its value lay in defending Windows and Microsoft 365 by making them easier to use, and in feeding queries to Bing. Microsoft did not publish Cortana revenue or usage figures, and the 145 million user claim was a reach estimate rather than a measure of engagement. When the phone business collapsed, the consumer case collapsed with it 3.

Usage never approached that of Siri, Alexa, or Google Assistant, and Microsoft's leadership came to see the standalone assistant as a strategic error. In January 2019, Satya Nadella said the company had decided to make Cortana a skill for Microsoft 365 subscribers rather than compete with Alexa and Google Assistant. Canalys data at the time put Alexa in nearly 75 per cent of the roughly 20 million smart speakers sold in the third quarter of 2018, with Google Assistant accounting for most of the rest 4.

The retreat was staged over several years. The iOS and Android applications were withdrawn in 2021. In June 2023, Microsoft confirmed that it would stop supporting Cortana in Windows as a standalone app, and the app stopped working on Windows 11 in August 2023 7 8. Cortana in Teams, Teams displays, Teams Rooms, Outlook mobile, and Microsoft 365 mobile was retired in autumn 2023; voice search and the Play My Emails feature in Outlook mobile followed in June 2024 2.

The functions that survived were absorbed into Copilot, the product line Microsoft now positions as its assistant. Windows Copilot replaced the taskbar entry, and Microsoft 365 Copilot took over the work-oriented summarisation and meeting tasks that Cortana had pioneered. The transition was presented as an upgrade rather than a withdrawal, and in important respects it was: a generative model could handle requests that the intent-based system had to route to a skill.

The most common criticism was inconsistency. Cortana worked well for calendar and reminder tasks inside Microsoft's services and poorly for general questions, local search outside the United States, and anything requiring a third-party integration. Users encountered an assistant that could be impressive in a demonstration and unhelpful in ordinary use 7.

Regional coverage was a recurring complaint. Features arrived later outside the United States, some languages were supported only for a subset of commands, and the skills catalogue varied by market. For a product whose value depended on being available everywhere, that fragmentation undermined the premise 1.

The strategy attracted criticism from developers as well. Microsoft asked them to build for a platform that had changed direction several times, from phone assistant to Windows feature to enterprise tool to skill inside rival assistants. Each change imposed new work and reduced the incentive to invest. Privacy advocates also questioned an always-listening microphone, though Microsoft emphasised on-device wake-word detection and account controls in response 5.

The persona itself became a liability. Microsoft marketed Cortana as a character, with jokes and banter, and reviewers judged the product against the expectation of conversation that the marketing created. The gap between the character and the command interpreter was wider than the gap between competitors' branding and their capabilities.

Cortana is instructive for three reasons. It showed that a capable assistant is not enough without an ecosystem: Microsoft had the operating system but not the phone, and the assistant's reach depended on devices it did not control. The Nokia write-down is the financial measure of that dependence 6.

It showed the cost of a fragmented strategy. The product changed purpose several times, and each change asked users and developers to relearn it. The skills kit arrived after Alexa had established the developer model, and it was withdrawn before it could build a catalogue 3.

And it showed that assistants are judged by reliability rather than by personality. The voice and character were distinctive; the misses, the regional restrictions, and the uneven application support were what users remembered. Cortana is now a case study in how platform position, not model quality, decides which assistants survive 4.

The name persists in Halo, and the enterprise features persist in Copilot. What did not persist was the idea of a general-purpose consumer assistant as a standalone product, which Microsoft now treats as a feature of the applications people already use. Modern assistants make the same bet on a different technology, and the durable lesson is that distribution and reliability outrank persona.

Three explanations for the failure compete. The first is structural: Windows Phone had about 3 per cent of the smartphone market, and no assistant tied to it could reach the scale needed to improve and attract developers. On this account, Cortana was doomed by a hardware decision made before it launched 6.

The second is strategic. Microsoft treated Cortana as a platform competitor to Alexa and Google Assistant when it should have been a capability inside Microsoft 365. Nadella's own retrospective, that Cortana needed to be a skill for Microsoft 365 subscribers, supports this reading, though it is also a justification for the strategy he subsequently pursued 4.

The third is product-level: the assistant was not reliable enough, and its personality-first marketing exposed the gap. This account is harder to test, because reliability and distribution reinforced each other. A better assistant on a failed phone platform would still have struggled, and a worse assistant on a dominant platform might have survived.

There is also disagreement about what the retirement means. Microsoft describes an evolution into Copilot, with the same intent and a better technology. Critics describe a decade-long retreat from a market Microsoft entered early and left late. Both descriptions fit the record: the consumer product failed, and several of its enterprise functions were carried forward under a new name 2.