Generative AI: How ChatGPT signals new productivity potential in the digital workplace

Omdia view

Summary

Chat Generative Pre-Trained Transformer (ChatGPT), a natural language processing (NLP) model developed by OpenAI, is an AI-powered virtual assistant that has taken the internet by storm. Generative AI (a system that can create new information from the data it has learned) appears human-like in its ability to answer questions in the same way as a regular conversation.

By feeding enormous amounts of information from Common Crawl (a non-profit organization that crawls the web and makes its archives and datasets freely available), ChatGPT applies its conversational-style machine learning to generate responses. By processing 45TB of data (around 300 years of text), ChatGPT’s IQ has been evaluated (using a verbal-linguistic IQ and Raven’s ability test) and scored 147—a “highly gifted” rating.

Although ChatGPT has some well-deserved praise for its quality of responses, its ability to handle a wide variety of subjects, and conversational style, it is merely an example of a good quality, generative AI model. The outputs these models produce may sound convincing by design; however, the responses they generate can be wrong or biased. That said, the financial investment OpenAI will receive and the early examples of ChatGPT’s potential indicate how AI technology will transform the digital workplace.

Microsoft’s investment in ChatGPT reveals the significance of generative AI

Speaking at Ignite 2022, Microsoft CEO, Satya Nadella, highlighted that delivering efficiency with automation and AI would be a critical digital imperative to help organizations do more with less. In 2019, Microsoft invested $1bn in OpenAI (with an agreement for Microsoft to incorporate GPT features into Bing), and is considering an additional multiyear, multibillion-dollar investment to accelerate AI breakthroughs (see Further reading).

Furthermore, in a Microsoft earning call (for the period ending December 31, 2022), Nadella said: We [Microsoft] will soon add support for ChatGPT, enabling customers to use it in their own applications for the first time. And yesterday, we announced the completion of the next phase of our agreement with OpenAI. We are pleased to be their exclusive cloud provider, and we’ll deploy their models across our consumer and enterprise products as we continue to push the state-of-the-art in AI.

Microsoft is moving quickly to integrate OpenAI into its services. For example, Microsoft Teams Premium (a new $10 per user, per month licensing SKU) leverages OpenAI to provide an “intelligent recap” capability. This capability automatically creates notes, recommended tasks, and highlights from meeting recordings to save time and increase productivity. Microsoft will also provide direct access to OpenAI models, helping developers to build AI applications via the Azure OpenAI service.

Microsoft has also announced its intention to integrate ChatGPT within its Customer Experience Platform, Viva Sales. By integrating generative AI with CRM data, Microsoft plans to generate suggested email content for replying to customer inquiries, create personalized sales proposals, and auto-suggest customizable content, and reduce the manual workload for sales teams.

Finally, in 2017, Microsoft introduced Bing for business—an intelligent search experience for Microsoft 365. Bing for business uses AI and the Microsoft Graph to deliver more relevant search results based on organizational context. Furthermore, Microsoft created a personalized productivity dashboard (Microsoft 365 feed) within Microsoft Edge. The Microsoft 365 feed highlights “@” mentions, recommended documents, to-dos, upcoming events, important emails, and company news. Microsoft’s initial investment in OpenAI included an agreement to integrate elements of GPT in Bing. ChatGPT will also provide more relevant and accurate search results within Microsoft’s Viva Suite modules, such as Viva Topics and Viva Connections.

While these sizable investments highlight AI’s strategic importance for Microsoft, conversational and generative AI is nothing new. In this report, we explore how this technology is currently used, and how AI will improve productivity, collaboration, and customer engagement in the future.

Generative AI for customer service: early stages, but worth your attention

OpenAI’s ChatGPT has been generating a lot of buzz lately. Within a week of its launch at the end of November 2022, the mega chatbot surpassed 1 million users and has subsequently gone viral. It’s understandable why consumers and enterprises are drawn to the mega chatbot’s ability to quickly find web content and communicate it to users in a conversational way. We were pleased with the response ChatGPT gave when we asked it to explain Keynesian economics in simple terms.

Interestingly, ChatGPT leverages generative AI to create new, unscripted content, which means its algorithm can automate new text, audio, code, images, and videos during an interaction. When these capabilities combine with ChatGPT’s speed, access to a vast pool of web content, highly customizable responses, and conversational tone, it’s clear that the chatbot can become an incredible force multiplier. This naturally leaves contact center professionals wondering if there is a place for ChatGPT in the customer service world.

The short answer is, potentially; however, ChatGPT would have to overcome some significant hurdles.

First, the chatbot currently doesn’t have access to real-time information. Its data and knowledge of world events after 2021 is limited, so, for example, it cannot help you make flight recommendations.

Also, ChatGPT has a short memory. While it can remember what a user previously stated in a conversation, its memory is limited to about 3,000 words within an existing conversation. Any information beyond this is not stored.

Current issues—Omdia’s brief experiment with ChatGPT

ChatGPT can generate incorrect information, alarmingly, in a plausible way. We asked ChatGPT, “What is the most valuable baseball card from the 1980s?” Its response was:

The most valuable baseball card from the 1980s is considered to be the 1986–87 Fleer Michael Jordan #57. This card features Michael Jordan, who was a rookie in the 1986–87 season, and was issued by Fleer. It is considered to be one of the most valuable baseball cards of all time due to Michael Jordan’s immense popularity as a basketball player and his brief stint in baseball.

What’s problematic about this response is that Michael Jordan did not have a baseball card in 1986 or 1987. ChatGPT refers to his basketball rookie card (which sold for $1m at auction in 2022). While a handful of Michael Jordan baseball cards exist, they weren’t created until the 1990s, with most printed after he pursued his short-lived baseball career.

Dissatisfied with this response, we again asked the same question by copying and pasting the original question into the response generator box. ChatGPT offered an entirely different answer:

The most valuable baseball card from the 1980s is considered to be the 1980 Topps Rickey Henderson #482 rookie card. This card features Hall of Famer Rickey Henderson, who is widely considered one of the greatest baseball players of all time.

This is the answer we were expecting. In February 2021, a gem mint version of Rickey Henderson’s rookie card sold at auction for $180,100.

From a customer service professional’s perspective, there are several problems here. First, ChatGPT generated an incorrect response involving perhaps the world’s best and most well-known basketball player. Second, it did so with a decisive and authoritative tone. The response sounds convincing when enough facts are combined with an error and presented in such a way. Subsequently, this could lead a customer or prospect down the wrong path, which could be irritating and potentially dangerous. Automated self-service solutions must be accurate and consistent with customer responses – especially when there’s only one correct answer. Clearly, ChatGPT is not there yet.

Fortunately, OpenAI’s CEO Sam Altman understands these and other issues about ChatGPT and provided a candid warning in early December, when he tweeted, “ChatGPT is incredibly limited, but good enough at some things to create a misleading impression of greatness. It’s a mistake to be relying on it for anything important right now. It’s a preview of progress; we have lots of work to do on robustness and truthfulness.”

ChatGPT must overcome some significant hurdles to become a reliable customer service tool; however, it would not be wise to ignore its potential, especially as companies are investing heavily in this technology. Perhaps partnerships, innovations, and integrations could resolve many of its existing challenges. If so, ChatGPT could help lower the barrier to entry for contact centers looking to deploy chatbots that can handle very large datasets. And maybe we’ll see ChatGPT, or another generative AI tool, automate good first drafts to assist agents, and respond to emails, business chats, and social media.

These are only a few potential customer service use cases; there could be plenty of others. So customer service executives should undoubtedly pay attention to ChatGPT; however, like any responsible leader, they should proceed cautiously.

How could generative AI, like ChatGPT, be leveraged in communication and collaboration?

While ChatGPT has received recent media attention, AI is already heavily leveraged in unified communications and collaboration (UC&C) platforms. AI is the technology behind video blur or video backgrounds, intelligent video framing, modern noise cancellation, closed captioning, and meeting transcripts. ChatGPT will be an additional AI-based capability that vendors may choose to add to their UC&C platforms.

ChatGPT seems very good at providing conversational responses to textual queries and summarizing text-based information. Consequently, ChatGPT will depend upon a given UC&C platform’s speech-to-text engine. Furthermore, ChatGPT hasn’t been trained with domain-specific information or a domain-specific lexicon. These two limitations need addressing before ChatGPT can meaningfully contribute helpful responses in collaborative communications settings.

If we posit forward to the day that ChatGPT can ingest company-specific and domain-specific information and lexicons, then ChatGPT may prove to be extremely useful in the following tasks:

• Meeting summaries: Meeting transcription is a standard feature in many UC&C platforms. As a meeting transcription is fed into ChatGPT, it can summarize a meeting, shortening the time non-attendees need to invest in reviewing and understanding the meeting content. ChatGPT could also be used to identify actions for individual attendees or the group. For example, if combined with workflow automation integrations, ChatGPT could create action items in an organization’s task manager and automatically schedule follow-up meetings when meeting participants are available (as is the intent of Microsoft’s Teams Premium license.)
• Surfacing information in real time: By “listening in” to a conversation, ChatGPT could be used to surface information relevant to the discussion. ChatGPT has an excellent natural language understanding engine. However, this capability will require integration with an organization’s data store so that ChatGPT can ingest all the information in the data store, analyze it, and index it. This will function much like “Agent Assist,” which is used today to surface information to a contact center agent while conversing with a customer. In a collaborative meeting setting, ChatGPT would surface documents, messages, and other content relative to the subject and flow of the conversation.

Both use cases will improve employee and organizational efficiency, which is where the payoff will be for using ChatGPT capabilities in a communications and collaboration context.

Identifying where ChatGPT will not be useful in UC&C may also be helpful. For example, ChatGPT cannot be used to examine video: it has no linkage or training model on video images or scenes. Consequently, it cannot help frame video, provide video sentiment analysis, or read body language. It may be able to interpret limited sentiment analysis based on a stream of text, but it is not equipped to analyze a speaker’s amplitude or tonality, which modern speech sentiment analysis tools can do.

However, AI tools need to “self-train” from an organization’s unique dataset and integrate with other business applications such as email, calendaring, task management, and file storage to improve productivity. The upside to this may be eliminating repetitive tasks such as meeting scheduling, follow-up emails, and content summaries. With Microsoft Viva’s intent to support a more inclusive work culture with spaces for conversations, company news, and shared interests ChatGPT could be effectively applied to automate and simplify this objective.

Omdia believes the starting role of generative AI will be to improve employee efficiency. Still, we will only truly realize the real potential of these tools when they start assisting employees in becoming creative in meetings and their day-to-day work. Employees’ creativity lies in their ability to innovate, find novel solutions to problems or challenges, and make connections between disconnected ideas. UC&C technologies can use generative AI to create virtual water cooler moments—by intelligently bringing two or more people with connected ideas into a shared space to allow them to think freely and combine their thoughts into a cohesive vision.

Leveling up employee support with generative AI

Digitally augmenting traditional support approaches

The most significant success of ChatGPT is how it has brought AI capabilities further into the general population’s consciousness. No longer is AI viewed as a complex technology that exists only to replace jobs and human effort. Via their own experiences with ChatGPT, many people now understand how AI can be utilized in professional and personal settings. This further elevates interest in how AI solutions can help improve business operations and how people work. IT and business leaders have a keen interest in AI capabilities to help enhance the way people work. In Omdia’s 2022 Future of Work survey, IT and HR executives identified AI capabilities as 24-month investment priority, second only to zero-trust security capabilities.

Generative AI will augment human efforts in improving employee support

Through better contextualizing and personalizing employee support approaches, AI has exciting potential to help improve end-user and support admin experiences. Traditionally, an agent fields an inquiry, manually searches internal knowledge repositories for the appropriate solution, and replies. Instead, generative AI (trained on business domain content, employee technology profiles, and incident history) could offer contextualized responses to questions, automate service requests, and enhance technical support. In complex cases, where first-line resolution is impossible, the handling agent could be guided automatically through the relevant escalation process.

From a technical support perspective, the configuration management database (CMDB), which tracks an enterprise’s hardware and software assets, will be an essential data source for AI solutions. The business value of generative AI comes from how business-related text and data (from conversations, incident records, etc.) tunes and trains the AI model. However, if this training is not practical or easy, the mainstream adoption of generative AI will be slow or, indeed, fail.

Automated intelligent workflow

Another area that conversational AI will disrupt is citizen development. Low/no-code solutions currently democratize development activities via drag-and-drop interfaces; however, generative AI may provide an easier way to automate workflows and develop micro apps. In much the same way ChatGPT has been used to write lines of code, it could also be applied to simplify business processes and improve productivity.

At Microsoft Ignite 2022 (see Further reading), Microsoft announced plans to implement AI features within Power Automate for users to create no-code apps. Users can naturally describe what they want to automate (via OpenAI Codex) and an AI-based “copilot” automatically builds the workflow. When used with Microsoft’s Process Advisor (which indicates to users what they should automate to improve productivity), Power Automate could eventually optimize workflow automatically.

The potential for conversational AI to simplify how employees interface with different business systems and processes is also exciting. For example, when onboarding a new hire, an administrator could use conversational AI to state that a new employee has started. Natural language and branching conversations could then automate and guide the administrator through setting up the new hire with all the systems, software, and hardware they need. Traditionally, these manual processes require an administrator to interact and seek approval from multiple departments like IT, HR, and facilities management.

Enhancing support processes with AI is something that ServiceNow has been doing for a few years now with its virtual agent capabilities, and the company is excited by the potential advanced conversational AI models can offer. We asked the service management and workflow automation vendor to explain how conversation AI could be utilized within their platform. A ServiceNow spokesperson said they are currently working with customers to identify key use cases and integration points.

ServiceNow is excited about the possibilities large language models like ChatGPT bring to the ServiceNow Platform. ServiceNow continues to invest in AI to take the load and burden off customers, employees, and partners and drive measurable business outcomes.

The support benefits also extend to employee self-service workflows. Such processes can become more accurate and personalized across more complex use cases. As businesses increasingly look to integrate cross-functional teams and functions, conversational AI chatbots will be important in surfacing knowledge and enabling employees to interact with digital ecosystems.

Enhancing employee productivity

Generative AI will be an important technology for enterprises in 2023 and beyond, especially as other prominent technology vendors (notably Google) will start to introduce their own solutions. When Google plays its hand here, the vendor must clearly show how it can add business value rather than relying on its AI engineering and technical prowess.

Unfortunately, although Google officially announced Bard (Google’s alternative to ChatGPT) on February 6, and held a live stream event on February 8, it felt like more of a defensive response to Microsoft. Google also missed the opportunity of explaining how Bard would be used in an enterprise setting.

This was a wasted opportunity, given Google’s healthy market share in Google Workspace. Bard could be used as a digital assistant, undertaking some of the more mundane and repeatable tasks that employees handle. It could also be embedded within Gmail, Calendar, Meet, Chat, Drive, Docs, AppSheet, Cloud Search, and other services to automate and simplify work. Text recommendations, grammatical corrections, natural language-based approaches to querying databases, slide assistance, real-time support suggestions, and endpoint management are all examples of how AI could be used to enhance Google Workspace.

AI-powered content (Content AI)

Generative AI could also be used to parse information from content across an enterprise, automatically extracting metadata to make it easier to find and categorize information. An emerging technology category, Content AI aims to transform how content is created, processed, made discoverable, and automated within workflows to improve productivity.

At Ignite 2022, Microsoft introduced Microsoft Syntex, a new service that uses AI to process different types of content – images, forms, and unstructured documents. Besides digitally generated content, Syntex can automatically create tags by recognizing handwritten text, reading fields in forms, and parsing dates, numbers, names, and addresses.

Consequently, enterprises could automate content-based workflows (e.g., contract processing, approvals, etc.), integrate with a user’s taskbar to make it easier to search for answers to natural language requests, and use AI to generate summaries of key content.

Conclusion

ChatGPT has been trained on vast datasets and information from the internet (sources like Reddit discussions, archived books, and Wikipedia articles) to help it learn dialog and attain a human style of responding. Vendors should point out that ChatGPT is merely an example of generative AI's capabilities. Vendors should explain that to get the most value from generative AI, models need to be trained from domain-specific organizational data, not public internet sources.

Chatbots have historically underwhelmed customers; however, ChatGPT illustrates that AI can now provide a smarter, more human-like response. Today’s contact center agents have many tasks to fulfill, many of which can be automated. Rather than logging service tickets, conversational AI can create internal knowledge base articles, scripts, and web posts to better inform customers and agents about a problem and the potential solutions. More advanced customer engagement platforms offer sentiment analysis, but conversational AI is likely to be quicker, more affordable, and customizable to suit the needs of an enterprise. Such platforms will be able to monitor conversations and threads, interpret emails, and alert customer engagement teams of any follow-up actions to be made to ensure customer satisfaction.

Outside the contact center, vendors can capitalize on generative AI through workflow automation, improvements to collaboration and communication, and simplified employee support. Such technology will help to unlock the productivity puzzle plaguing organizations today, brought about by siloed data, manual processes, and a lack of organizational awareness.

Appendix

Further reading

ICT Enterprise Insights: Contact Center – 2023 (December 2022)

“AI in the contact center will help to weather economic uncertainty” (January 2023)

“Microsoft Invests In and Partners with OpenAI to Support Us Building Beneficial AGI” (July 2019)

“Microsoft Ignite 2022, Satya Nadella Keynote” (October 2022)

“Microsoft and OpenAI extend partnership” (January 23, 2023)

“Microsoft Fiscal Year 2023 Second Quarter Earnings Conference Call” (January 24, 2023)

“Microsoft Teams Premium: Cut costs and add AI-powered productivity” (February 1, 2023)

“Microsoft boosts Viva Sales with new GPT seller experience” (February 2, 2023)

Authors

Tim Banting, Practice Leader, Digital Workplace

Adam Holtby, Principal Analyst, Workplace Transformation

Brent Kelly, Principal Analyst, Unified Communication and Collaboration

Prachi Nehma, Principal Analyst, Unified Communication and Collaboration

Mila D’Antonio, Principal Analyst, Customer Engagement

David Myron, Principal Analyst, Customer Engagement

askananalyst@omdia.com