Generate Value From Genai With ‘small T’ Transformations

Neil Webb

Less than two years ago, generative AI made headlines with its amazing new capabilities: It could engage in conversations; interpret massive amounts of text, audio, or imagery; and even create new documents and artwork. After the fastest technology adoption in history — with over 100 million users in the first two months — businesses in every industry began experimenting with it. Yet, despite two years of broad managerial attention and extensive experimentation, we are not seeing the large-scale GenAI-powered business transformations that many people initially envisioned.

What happened? Has the technology failed to live up to its promise? Were experts wrong in calling for giant transformations? Have companies been too cautious? The answer to each of those questions is both yes and no. Generative AI is already being used in transformative ways in many companies, just not yet as the driver of a wholesale redesign of major business functions. Business leaders are finding ways to derive real value from large language models (LLMs) without complete replacements of existing business processes. They’re pursuing “small t” transformation, even as they build the foundation for larger transformations to come. In this article, we’ll show how they’re doing this today and what you can do to generate value with generative AI.

How Businesses Are Transforming With GenAI

Our project team interviewed the senior managers of various functions, including artificial intelligence, data science, innovation, operations, and sales, at 21 large companies. We focused on understanding what organizations with relatively early and broad GenAI adoption are doing and why. We also reviewed public information about companies similar to those we studied.

To start, we needed a definition of what digital transformation means. An early definition is “the use of technology to radically improve the performance or reach of an organization.”1 More recently, OpenAI’s ChatGPT synthesized this definition: “a comprehensive integration of digital technologies that fundamentally reimagines business models and processes, contrasting with incremental change, which focuses on gradual improvements.” Digital transformations, in general, consist of numerous technology-enabled improvements, often assembled over time, to create broader change in how a company operates. They are driven not by a single technology but rather by using the right technologies for the right tasks to deliver a new way of doing business.

Our research shows that most companies are following a more targeted approach to transforming with generative AI. While GenAI can potentially increase the speed and quality of many tasks, it also comes with a variety of risks around accuracy, security, and intellectual property management. The leaders we interviewed tend to apply the logic of a risk slope when making their decisions, attaching a higher risk to customer-facing processes than to internal ones. Leaders in industries like medicine and financial services also see these risks through the lens of regulatory compliance.

Some leaders are thinking beyond these highly publicized GenAI risks to also consider the costs and risks of preparing the organization for large-scale implementations. They seek to reduce investment in software licenses and hiring skilled people until the returns are clearer. They also see risks in scaling AI transformation pilots to the enterprise level without first cleaning up the data and back-end systems that feed into them.

At the lower end of the risk slope are discrete uses that can deliver immediate value at relatively low risk. More extensive transformations may provide significant value, but they also have higher risk. As a result, many IT and digital leaders are investing first in use cases at the bottom of the risk slope, aiming to deliver early value while also developing capabilities that can partially de-risk implementations that are higher in cost, risk, and potential return.

Three categories of transformation represent different areas of the risk slope, starting with low-risk individual uses, then moving to role- and team-specific tasks, and finally to products and customer-facing experiences.

1. GenAI in Tasks That Are Common to Individuals in Many Roles

At the lower end of the risk slope, employees are using LLMs in ways that are useful to many roles, such as writing, synthesizing information, generating imagery, and documenting meetings. The near-ubiquitous nature of GenAI can have real impact within the organization. As early adopters considered what tools to provide to which employees, those employees began using public tools such as ChatGPT or Gemini without asking permission. Wary of privacy and accuracy concerns but mindful of the added cost, many companies have begun to offer de-risked GenAI tools to at least some of their workers. Some are now starting to buy or create integrated tool sets that link generative AI to other functions that employees typically perform. Benefits vary by use and user, with individual initiative-taking and prompting skills influencing the value they derive. Consider the following approaches.

Company-specific LLM instances: One way to address privacy and security concerns is to license private instances of major LLMs, such as ChatGPT or Anthropic’s Claude, which can be accessed through secure cloud platforms. Employees can use these stand-alone GenAI tools for synthesis, writing, and other content-generation tasks without leaking confidential information. Beyond generation, some are using the tools for learning. For example, a technical product manager we spoke with at a large U.S. technology company uses an internal tool that simulates feedback from an executive on draft presentations and reports.

Built-in integration with common office productivity tools: Use cases and productivity gains expand when an organization can integrate an LLM with company information and desktop tools, such as Copilot and the Microsoft 365 suite or Gemini and Google Workspace. They create what one manager called “a super search engine,” able to identify useful documents both within a team and across a global company. Integration allows employees to pull content from different sources, such as emails, meeting transcripts, and internal documents. It also makes it possible for a user to request specific information from their inbox, such as a list of open items from the past week, and to prepare spreadsheets or presentations without having to start from scratch, given the LLM’s access to company information.

Custom integration: Some companies are going beyond basic desktop integration to add company-specific intelligence, by training models on terminology and information that are proprietary to the company. Global consulting firm McKinsey built Lilli, a platform that links generative AI to its intellectual property from over 40 internal sources. The effort involved significant technical hurdles — for example, the tool needed to be modified to read PowerPoint slides, which are one of the company’s main ways of communicating project information — but the platform is providing value to the company. If a consultant has a question about green energy business models in less-developed economies, for instance, Lilli can quickly find and synthesize information from projects that have already studied the problem somewhere in the world. The platform’s capabilities, combined with robust employee education, led to about 75% of employees actively using Lilli in less than a year, time savings of up to 30%, and substantially improved quality, McKinsey has reported.

McKinsey is not alone in developing these specialized models for the general workforce. Another company we studied has extended its specialized LLM to not only find information but also autogenerate initial drafts of proposal text and slide decks.

External tools: Company-specific tools can be useful and secure, but some people will look elsewhere for functionality that they cannot find internally. One program manager in a large tech company told us that he uses ChatGPT for tasks involving nonconfidential information, such as writing specifications or structuring documents, while using DALL-E to create simple visuals easily instead of using more cumbersome tools he has access to. A coworker in product management turns to Superwhisper — a voice-to-text/text-to-voice tool that keeps all information on his device — to dictate, summarize, and clean up his thoughts during performance reviews. He also uses tools like Perplexity to fact-check and cross-reference information in his writing tasks.

2. Specialized GenAI in Specific Roles and Tasks

Companies working their way up the risk slope are developing generative AI capabilities that will improve productivity and quality in specific job roles or business processes. Here, there is less tolerance for unacceptable output, though not yet to the same degree as with customer-facing applications. These solutions typically maintain a human in the loop, where employees interact with the tools and review the outputs rather than allowing the GenAI tools to automatically make decisions or produce outputs. Consider the following use cases.

Coding and data science: This is one of the earliest and most common GenAI-assisted tasks in every industry. Rigorous studies have shown major productivity gains for software engineers who use coding copilot tools to speed up tasks like writing code, finding useful libraries, or conducting code reviews.2 Yet another productivity gain comes from generating sample code or data for training purposes. Data scientists, meanwhile, are using generative AI tools to conduct data analysis or produce scripts for analytic tools. By creating readable documentation, GenAI offers significant time savings to coders, data scientists, and their managers in performing what many consider a tedious task. Copilot tools can not only help experts work more efficiently but also help novices improve their skills. However, these benefits are not automatic: People often need training to get the most benefit from the many features available.

Support of customer-facing individuals (with a human in the loop): One of the first uses of role-specific GenAI for customer-facing applications is in customer service. The tools can help agents find information quickly and also suggest actions in real time. Some can even coach the agent later, synthesizing numerous calls to identify patterns and opportunities for improvement.

In one study of a tool that provides GenAI-based coaching for call center representatives, MIT researchers found that access to the tool increased productivity, as measured by the number of issues resolved per hour, by 14% on average, with a 34% improvement for novice and low-skilled workers.3

Current examples of how generative AI can assist with customer service include:

• Amazon Pharmacy’s internal chatbot to support customer care representatives can retrieve answers from the help center knowledge base and summarize the information for the representatives, allowing the reps to answer customer questions in less time, according to the company.
• Morgan Stanley found that its knowledge assistant tool, trained on more than a million pages of internal documents, speeds financial advisers’ process of finding information, allowing them to spend more time focusing on customer needs. A new tool summarizes customer video meetings and drafts tailored follow-up emails. This kind of interaction is not limited to finance.
• Sysco, the world’s largest wholesale food distributor, is using GenAI in tasks ranging from making menu recommendations for online customers to generating personalized scripts for sales calls, based on customer-specific data.

Online content generation: CarMax, the largest omnichannel used-car retailer in the U.S., was an early adopter of OpenAI’s generative tools. CarMax uses AI to produce text for its car research pages, which help customers make a purchase decision, and to embed keywords and organize content in order to boost a web page’s search ranking. It summarizes customer reviews of a car’s model into a few sentences, saving customers from wading through hundreds of individual reviews. Summarizing over 5,000 car pages manually would take multiple humans 11 years, according to the company. With generative AI, it runs this process regularly, and it takes only a few hours. The quality has been even better than anticipated since it did a little fine-tuning, with an 80% editorial review approval rate. CarMax has since expanded its use of GenAI to include marketing design, chatbots for customers, and tools for internal associates.

Creative processes: Dentsu, one of the world’s largest creative agencies, uses generative AI in all stages of the creative process, from proposal to project planning to creative ideation. Employees can use it to turn a few lines of copy into a proposal, manage complex budgeting spreadsheets, or make sense of notes from numerous planning meetings. In creative sessions with a client, instead of gathering ideas and making the client wait days or weeks for visual concepts, the team can iterate with the client in real time. “They’re getting less time on mundane, tedious tasks,” said Kate Slade, Dentsu’s emerging technology enablement director. “They can be creative and create higher-quality content with less effort.” Dentsu and other companies can use foundational LLMs to generate product image mock-ups in the conceptual phase or use specialty tools like Flair.ai to create polished product photos, including for clothing and accessories, which are displayed on AI-generated models.

Finance and regulatory: Multiple surveys have shown that finance teams are relatively late adopters of new technologies, with CFOs citing technology gaps, data concerns, and competing priorities as reasons for that lag.4 However, some companies are innovating within this business function. One international energy company we studied created a tool using a mix of GenAI, traditional AI, and other algorithms that can suggest mitigations or help rewrite an audit report. Other companies are using generative AI to assist in drafting reports for audits or regulatory compliance. At Amazon, the finance function uses a mix of rules-based AI, machine learning, and LLMs to address tasks in fraud detection, contract review, financial forecasting, personal productivity, interpretation of rules and regulations, and tax-related work. Managers have reported improved performance on those tasks and that employees have been able to shift their efforts away from repetitive tasks to instead focus more on work that involves critical thinking.

While productivity gains are the expected and common benefits of applying GenAI to specialized roles and tasks, the technology’s true impact extends further: Generative AI is fundamentally transforming what professionals can achieve across industries. By not only enhancing efficiency but also expanding the realm of possibilities within various functions, GenAI is enabling innovations and reshaping traditional processes.

3. GenAI in Products and Customer-Facing Interactions

When people ask about GenAI-enabled business transformation, they often mean changes in products and other customer experiences. However, these changes are often higher on a company’s risk slope, and judicious action is warranted even as companies strategically consider more extensive future applications. Traditional companies are starting to implement GenAI-enabled customer service to answer simple queries and to GenAI-enable the sales process. Meanwhile, major software companies are already incorporating GenAI-powered functionality into their products. Consider the following use cases:

Direct customer service interactions: GenAI is taking the traditional phone menu or robotic process automation-enabled chatbot to a new level of sophistication. It offers natural language interactions and flexibility not possible with rules-based AI, plus it adds multilingual capabilities. For example, life insurance company John Hancock built chatbot assistants to handle common queries, freeing up human agents to handle more complex issues. This reportedly reduces company costs, customer wait times, and employee time spent on simple tasks. Now GenAI is expanding to voice interactions in organizations like Starbucks, Domino’s, CVS, and banks. It’s only a matter of time before these tools expand to include video.

Personalized shopping experiences: Customers are accustomed to getting e-commerce product suggestions based on what they (or others) have bought, or banner ads based on what they have viewed. Companies now are using GenAI to tailor the shopping experience throughout the customer journey.

For example, Tapestry, the parent company of brands such as Coach and Kate Spade, uses real-time language modifications to personalize the online experience to individual shoppers as they are moving through a retail site. This includes injecting a conversational tone that mimics the experience of engaging with a store associate. Tapestry claims to have seen an e-commerce revenue increase of at least 3% due to these personalization changes.

Amazon is also personalizing the customer journey, by offering product recommendations and descriptions that fit a customer’s holiday, sport, or diet preferences, or their household size. In addition, it is executing a phased rollout of tools to help vendors improve the shopping experience that they provide.

GenAI video generation and a chatbot for insights and recommendations on their business performance are currently in beta mode with some U.S. retailers.

Enhancement of existing software products: Even as traditional companies experiment with GenAI in customer interactions, leading software companies have already begun integrating generative AI capabilities into their products, whether to improve existing features or add new ones. Generative AI’s natural language processing capability smooths the path for a user of Lucidchart, for example, to create a flowchart just by writing what they want rather than having to go through the steps and menus manually, resulting in a draft that is editable and sharable.

Canva, another visual communication tool, uses ChatGPT to ease the process of creating and modifying slides, images, videos, presentations, and social media posts. This increases productivity for any user, with the added benefit of decreasing frustration for inexperienced users.

Similarly, Adobe has embedded GenAI features throughout its product suite so users can easily create and modify images, adjust them for brand style, or “chat with a PDF.” In marketing campaigns, Adobe’s GenAI features help to track consumer behavior, personalize content, and improve performance measurement.

As these and other companies integrate generative AI into their product suites, their corporate customers may choose to wait for features rather than build the functionality themselves.

Such integrations into products and customer-facing interactions are enhancing customer experiences without a need for employees in the loop. This strategy serves a dual purpose: It personalizes interactions to increase customer engagement and sales, and it empowers users to achieve new levels of productivity and capabilities on their own.

Generating Transformation

What is the right kind of transformation with generative AI? The answer is, anything that a company can do to change its performance or reach using the technology, in concert with other technologies and, often, human action. Our research uncovered examples in industries from fashion to finance and roles from auditing to marketing. The companies we studied are being careful as they work their way up the risk ramp. They are pursuing small-t transformation, often with a human in the loop, as they build capabilities that can enable the development of applications that have higher value and risk.

Our research suggests a number of actions leaders can take to generate transformation with generative AI.

Identify key pioneers in your organization, from decision makers and stakeholders to power users, and develop your message for them. With generative AI, innovation often comes from “cyborgs” — early adopters who integrate the technology into their work and are motivated to use it to solve a problem for themselves or for customers.5 However, these workers, as well as the later adopters, may be concerned about a negative reaction from their employer or about GenAI replacing their jobs, hence the need to communicate your innovation vision. Bring your IT and data teams to the table, as well as employees from domains already adopting GenAI or who are the most interested.

Assess where your company is now on the risk slope relative to the companies we’ve described. What are you already doing, and what would be the next level of complexity and reward? Look at the opportunities in the areas of individual productivity, role-specific enhancements, and innovations in product or customer engagement. Keep in mind that while companies can develop in all three simultaneously, the maturity levels likely will vary.

The companies we studied consider security issues, integration with other systems, and output sensitivity in deciding where to begin and where to go next. CarMax’s GenAI solutions currently have some human-in-the-loop elements, according to Shamim Mohammad, the company’s executive vice president and chief information and technology officer. “As GenAI matures and becomes ever more sophisticated, transformational or game-changing use cases will emerge,” he said. “But in my view, it will take some time before organizations can deploy GenAI solutions without human intervention and supervision comfortably.” As leaders consider moving beyond the low-hanging fruit that many have pursued so far, they face challenges that must be cleared away, including scalability, management buy-in, and the need for foundational capabilities.

Consider scalability. Numerous research participants described how the process of going from pilot to scale is nontrivial. “It’s easy to do the proofs of concept, but bringing it to the right level of trust among a large group of users is much more difficult,” according to Pentti Tofte, staff senior vice president of data analytics at FM, a large commercial insurer. Data outside of the pilot environment is messier and less connected than it is inside of it. Low-probability events appear more often when working with large volumes of data or customers. According to the head of AI at a large bank we spoke with, “The more stuff you do, the more stuff you find to do.”

Beyond the technical questions of scaling specific pilots, leaders are asking strategic questions about whether it is better to move quickly — even if it means training and integrating custom models and reworking them as technologies change — or to wait until software vendors incorporate new features into the systems they already use.

Secure management buy-in. Management buy-in is essential for larger projects, since managers have heard about the risks of generative AI and may have learned to be skeptical about the promise of new technologies. An executive at a medium-sized tech company in New England reported that their GenAI innovation stagnated until the CEO saw its potential, allocated resources, and communicated how GenAI would be expansive for employees as well as the company. Small-t innovations can help to make the value story real and make the case for investments that can reduce the perceived risk of larger opportunities.

Investigate foundational investments that can improve the risk-return ratio higher up the risk ramp. Some of the boldest use cases will require extensive investment in data cleansing, model training, and integration before they can be ready for a real-world test. “The most important thing with GenAI is, you have to make sure your data is correct because good data will give you good results from AI — GenAI especially,” Mohammad said. “You also have to have the right AI governance in place to ensure that the AI is being deployed responsibly.”

Two notable examples where this is the case are know-your-customer applications in financial services, and regulatory compliance in financial services and health care. Large banks and insurers may have thousands of people doing these tasks, and much of the work is about integrating and interpreting large amounts of unstructured information.

But getting GenAI to produce accurate results in a tightly regulated environment is very difficult. “In this space, the fruit is not so low,” said an executive from the financial services industry. What seems like a perfect application for GenAI is slowed by the need to build the right internal foundation of data and process before proceeding in these areas, where inaccurate decisions have high costs. If generative AI is going to create an enduring transformation of the business, “it doesn’t go hand in hand with the gold-rush mentality that we have to mine it now,” said Prem Natarajan, chief scientist and head of enterprise AI at Capital One. Instead, companies should think this through, he advised: “Figure out how to do it thoughtfully and responsibly. Build the scaffolding to bring everybody along.”6

Maintain a long-term perspective. The gold-rush mentality is real, but costs and uncertainty are too. “The transformative cases take longer to build the business case, test the models, change behaviors, etc.,” said Chris Bedi, chief customer officer at software company ServiceNow. “The challenge is not only technical but also leaders taking time to reimagine their future with big ideas.” Leaders are trying to separate hype from reality while also understanding that they may need to invest in data and technical foundations before they can get the returns they seek. Applications such as customer service and personalized shopping can show real innovations with material returns, but more complex work is needed to make broader change happen.

Companies are already using GenAI to pursue small-t transformation nearer to the bottom of the risk slope. For larger transformations, GenAI will be one of many pieces in the puzzle. “Instead of one transformative thing, we’ll stitch together many technologies, including AI, to reinvent a whole process,” said FM’s Tofte. Although it may take time before your company feels ready to launch transformations higher on the risk slope, you need not wait to make progress. You can experiment on some tasks while making foundational investments in data and integration that will make larger transformations possible over time. Choose the areas where you want to invest, both in the short and long terms. Then invest in building awareness and cross-cutting capabilities that can make you faster and more efficient in the future.