Gen-AI Value Realization & Use Cases

We have seen various shifts in the technology era from cloud-first and mobile-first to now AI-first. Enterprises plan to pivot to an AI as a business capability mindset driven by an AI-first approach across all initiatives.

There are huge expectations in realizing the anticipated benefits of Gen-AI across enterprises. We all have talked about the super smart chatbots that can solve all our problems, but it's not that simple to realize that desired experience when you factor in the challenges of diverse data sets, data training, quality, skills, and, ultimately, overall ROI.

Hence, it's important to structure the approach that balances the current state of maturity of the enterprise with business and technological goals.?We can structure it in three phases:

1. Everyday GenAI
2. Transform Existing Functions using AI
3. New Products and Services Using AI

?Let's understand the details of each

Everyday GenAI or AI as Co-Assistant

• What - Deploy gen-AI capabilities enterprise-wide to enhance the productivity of employees.
• How - provision out-of-box capabilities providers, aka the Co-Pilots, or assess products that are adding GAI capabilities (e.g., Salesforce, Braze, Adobe). As these tools are not autopilots but co-pilots, existing data privacy, confidentiality, and intellectual property guidelines can be modified to allow for these new providers. Change management and training efforts can enable rollouts and address data confidentiality concerns.
• Value ($$) - set clear frameworks and metrics to assess value and benefits. Metrics around workforce productivity, spend, team velocity, and Dora metrics can be refined to track and assess gains.

• Risks - Costs can increase fast as usage for tools goes up across the enterprise. Vendor lockins may happen as the usage of out-of-box providers increases.
• Example: imagine an enterprise that has enabled Copilot for engineering, AI note taker for calls, and text-generating LLM for other team members. We have a 10-member team of product managers, engineers, architects, and quality folks who were following a 60(new feature)-20 (tech debt)-20 (bugs) cadence in their regular sprint. After enabling co-pilots, we can expect a faster turnaround of new features as new ideas, research, coding, and note-taking can be aided by AI tools (hence the 50-30-20 flip). Over 4-5 months we can anticipate better quality and faster turnaround time, hence overall timeline can be reduced. A five percent overall gain across ten sprints may translate to 400 man-hours gain (or more), this can easily offset licensing costs and also reduce spending. Similar gains can be estimated for marketing teams if they start using tools for research, strategy, content writing or AI image generation, all these can help increase productivity and reduce efforts in day in a life for marketers.
• Overall, this approach can showcase quick wins, and an increase in cost can be offset by incremental productivity and savings. However, large-scale pilots would be needed to assess the value realized and streamline the metrics framework. The time to value for enabling this area should be quick.

Alter Existing Functions using GAI/AI

• What - Evaluate and enable use cases that allow you to augment capabilities of internal or external facing applications using Gen-AI. These can be in areas of procurement, supply chain, finance, customer service, customer-facing digital apps, and services.
• How -? Leverage enterprise data to increase context with large language models. Enterprises can use emerging RAG or Fine-Tuning approaches, as covered in the previous article. There will be need to reassess ways of working, roles & responsibilities, security considerations and tech stack to enable the outcomes. We can expect hybrid approaches of combining GAI with traditional machine learning in a few use cases.
• Value ($$) - Early value will come in the form of service availability and faster processing time. Customer satisfaction will increase as the responses become better via regular training and improvements.

• Risks - Since use cases impact existing applications or functions, organizations have to deal with increased regulatory and privacy compliance needs. There will be additional costs tied to enablement of new technology as very few use cases are in production currently, hence moving from piloting to launch will bring its share of learnings, beyond just high compute costs. The correct response once doesn't mean it will always be the same; regular data checks and training will be needed. Forgoing data quality, labeling, and classification can lead to hallucinated results; it has to be addressed continuously and can't be prioritized after one time.

?Examples :

• HR functions can create an interaction mechanism for employees, allowing them to query questions from massive enterprise data and policies.
• Banks are creating advanced virtual assistants who can guide users through the loan process and give personalized insights based on the interactions (chat history, personal details)
• Retail sites provide summarized insights from reviews by analyzing massive amounts of customer reviews using LLM models.
• Procurement groups can create a smart assistant which can quickly review draft statement of work (SOWs) using enterprise guidelines, policies and existing vendor terms, this can increase speed and reduce lead times in creation of contracts.
• In summary, altering existing functions' use cases will have varying degrees of maturity curves, which will require continuous efforts to increase accuracy, quality of responses, and balance costs. Time to value will not be quick, and the cost to value will be medium to high.

New Business Models Using AI

• What - Evaluate AI first use cases that can create new digital/physical products and services for new or existing customer personas. These use cases will drive P&L changes in true AI as a business capability model.
• How - Establish digital native integrated product teams working together to prioritize new use cases based on customer and business needs. Adhere to the lean approach of starting small and failing fast. Establish the foundations for data platforms, quality, traditional ML capabilities, and Operations (ML-LLMops) and augment that with LLMs.
• Value($$) - Value will come late as the majority of use cases will need to prove product market fit and may also formulate the creation of new products, services, or categories. For existing products/services, an uptick in activation, retention, and monetization metrics should be assessed.

• Risks - There will be bigger risks of LLMs locking in; hence enterprises may want to POC using Close LLMs (OpenAI, Gemini) and switch to open LLMs (Mistral, HuggingFace); this will require the evolution of interoperability capabilities along with forward and backward integration. Longer-term funding will be needed as value realization will not be quick. Enterprises may be enticed to create their own LLMs for these use cases; a thorough analysis is critical.

By 2028, more than 50% of enterprises that have built large AI models from scratch will abandon their efforts due to costs, complexity and technical debt in their deployments. Source: Gartner

Examples

• Personalized AI guide in a medical mobile app that can use medical imaging using advanced ML algorithms, conduct sentiment analysis of user questions and facial expressions, and provide personalized responses and clinically guided pathways.
• AI personal shopping assistant (pay $X to use it) which analyzes social trends, posts, buying history, and fit options and not only recommends products but also helps interact like a stylist and, in the end, manages the ordering process
• In summary, use cases that create new business models and products using AI should be treated as strategic and competitive differentiators and, hence, should be given a long lead time in value realization.

It is a fine balance between value, effort, and risk across three approaches. Most enterprises will plan to initiate all three in parallel and leverage learnings from each approach.

As per my experience, altering existing functions using GAI/AI will be most complex because of already existing technology and data and new change management for processes and talent. Differentiating competitive advantage will come from creating new products and business models using AI.

Where are you on this exciting journey?

Previous Articles

AI & GAI Trends 2024

GenAI - RAG + Fine-Tuning for Enterprise AI

Good Reads

Gartner - GAI Value, McKinsey AI Value Chain, GenAI in Banking, Potential of GAI (BCG), GAI for Software Development

Note: The ideas, views, and opinions expressed in posts and profiles represent my views, not those of my current or previous employers or LinkedIn.