Beyond the Hype: Real-World Applications of Generative AI
Sri Gari
Feb 13, 2024
If you haven’t read already, please read my previous blog on “Is Generative AI Worth The Hype” that goes over what is different about gen AI. Here in this blog we go deeper into potential use cases of gen AI.
Gen AI seems to be the universal fix, or so some enthusiasts would have us believe. In reality, generative AI is a transformative tool, best wielded with a clear understanding of its strengths and limitations.
The real magic of AI lies in its ability to augment our capabilities, inspiring us to tackle challenges with renewed creativity and insight. Let’s dive in to understand where gen AI can be applied effectively in the enterprise world.
What are the criteria for AI applications?
First, let’s understand the strengths and limits of gen AI, keeping in mind that this could change quickly as the technology evolves.
The core ability of gen AI is to generate content like text, images, audio, video, and code, based on user input or provided context. Based on this ability, below are a few strengths and limitations of gen AI.
Strengths of Gen AI
Creative content generation: Since gen AI has been trained on vast amounts of human generated content, it can mix and match images, text, audio, and sometimes even video to create new content from existing information. This could be audio generated from your company's meeting notes, or creative images like “astronaut on a giraffe”.
Context learning: Even though gen AI is trained on generalized data to produce a large language model (LLM), you can augment it with your own data. This data could be your company’s historical information that could be fed to gen AI in a process called RAG (Retrieval Augmented Generation) to make the responses factual and decrease hallucinatory responses. As an example, you can provide live guidance within support or sales workflows based on historical conversations.
Language understanding: This is the ability to understand human-like questions and statements. For example, you can provide a human-like prompt to make the AI behave in a friendly, funny, or even angry manner.
Data augmentation: Did you ever feel like your data is scattered across the board and you had to manually consolidate data from multiple sources? Well, gen AI could be used to enrich your data from multiple sources. If you had a city name, but if you need the state of the city, gen AI could fill that up for you.
Pattern recognition: Gen AI could recognize patterns from both structured and unstructured data. For example, it can derive interesting insights from unstructured meeting notes, or it could understand structured APIs, and learn how to make specific API calls a.k.a agents.
Transfer learning: LLMs can be fine tuned to provide information in a specific style, format, and representation. For example, a consumer company might call Diabetes as sugar, whereas a medical company might call Diabetes as “Diabetes mellitus”. You can train LLMs to respond with company or industry specific terminologies.
Well, those are a few strengths of gen AI as it stands today. But along with those strengths come limitations and boundaries. Let’s look into a few limitations.
Limitations of Gen AI
Original creative work: It’s great at creating content from existing data, but it is also limited to that data. Humans on the other hand can come up with contrarian ideas, which were not available in historical data.
Complex reasoning: While gen AI demonstrates proficiency in tasks related to knowledge, it falls short in executing tasks that inherently demand arithmetic calculations and logical reasoning.
Ability to generalize and specialize: Yes, it can understand natural language queries, but it is heavily tied to the data it is trained on. It typically struggles with providing accurate answers for both high level general knowledge type questions and deep domain specific answers. Though there is some ongoing research to make AI versatile, the costs, latency, and accuracy of its performance are still questionable.
High stakes decision making: It can definitely augment data, but it still cannot be trusted to make decisions on behalf of humans. It still needs a human in the loop to continuously review and improve responses.
Data quality and quantity: Pattern recognition is one of the high value strengths of AI, but its performance is based on the level of data richness. If the data is garbage then the output will be garbage.
Ethical decisions: AI is great at learning from input and feedback of humans, it thrives at learning specific styles of industries. But it doesn’t have a moral and ethical compass, it cannot automatically conform to social and political norms.
Now that we have a set of strengths and limitations, let’s define a few criteria, where gen AI can be applied effectively.
Checklist for Determining Gen AI Applicability
Human in the loop
Is there a human supervising the AI? For example, having a support chatbot that provides refunds to customers without a human reviewing it might lead to false refunds. A good application would be text to SQL, where a human can verify the process, and provide continuous feedback to the AI.
Constrained environments
Will the AI be given a limited set of conditions? AI can go haywire if let loose without any limitations. This could be good for creative art, but for enterprise businesses, the more specific the requirement, the better is the accuracy. For example, accuracy of email drafts for a specific email template with a specific historical data source performs better than giving free reign to AI.
Accuracy tolerance
Can intermittent low accuracy be tolerated? Gen AI cannot always perform at high levels of accuracy. If the problem space has the expectation of >= 99% accuracy, then it might not be a good candidate for gen AI. For example, it would be dangerous to get advice from AI on medications for an illness. On the other hand, marketing content generation reviewed by a human is a safer bet.
Data transfers
Is there grunt data transfer from one system to another? Is there a manual repetitive process where data from one system needs to be changed into another format to integrate into another system? Then gen AI could be used to automate it by training on the formats, the data structures, and API methods. For example, reading data from a SaaS tool, changing the format with specific constraints, and pushing it into placeholder positions within a word document.
Analysis and insights
Is there an opportunity to reveal insights from data? Gen AI is excellent at scouring through unstructured data and revealing interesting insights that we couldn’t think ahead of time. This is especially useful for business teams to run ad hoc queries.
What are the potential use cases?
Based on the strengths, limitations, and criteria, let’s go through a few potential use cases for enterprises. At a high level, we’ve divided the types of applications into three categories.
Discovery: These are applications which are mostly replacing traditional keyword based search to find relevant information. Examples include generating responses based on internal company policies and documentation.
Generation: These are applications where new content is generated based on the user's context, including taking actions on behalf of the user. This could be an employee finding the number of vacation days left, and then applying for a leave.
Insights: These are analytics on top of structured or unstructured data. For example, a senior manager can run an ad-hoc query to get “top reasons for sweater returns in month of May in a bar chart”, without ever asking an analyst to generate a new report.
We’ve further divided the use cases by specific departments within an enterprise.
Sales
Product pitching: Increase sales throughput by providing product information on finger tips. By keeping gen AI in sync with your product documentation, sales people can always make a personalized product pitch based on the customer.
Strategy guide: Get suggestions on the next best action from your sales playbook to drive sales and increase revenue.
Meeting summary: Automate sales meeting summaries formatted as per your company policies so that your salesforce can save time, get key takeaways, and follow-up to close deals.
Lead enrichment: Give consolidated information in one place to make it easy for sales people to get customer insights before reaching out to them.
Sales summary and insights: Get broad and deep insights on sales performance to ensure your salesforce is continually learning, adapting, and performing at their best.
Marketing
Market and user research: Let gen AI scour through your resources and provide insights based on your target audience. Teams can save time and formulate better go to market strategies to yield the best results.
Email/blog generation: Get live suggestions on your next blog or email based on your company guidelines and policies. Save time, get creative suggestions, and write better performing blogs.
SEO and web copy generation: Create personalized website copy and SEO metadata to drive traffic to your website and increase your customer engagement.
Website and media insights: Let marketing teams plug their data and ask questions like “What are the best performing ads last quarter for instagram stories”.
Support
Product intelligence: Your support agents will always be on top of your products when gen AI is synced with your product knowledge base.
Response guide: Get live suggestions within your support workflows to provide timely responses to your customers and increase CSAT.
Issue summary: Tired of going through issue threads? Get issue summaries in an instant to pinpoint the problem and come up with a solution.
Support agent insights: Track all your support metrics, provide helpful feedback, and continually train your support teams to provide great customer service.
Operations
Process knowledge: Have a question on a business process or workflow? Get instant answers generated from your process flow documentation and diagrams to increase operational efficiency.
Workforce guide: Onboard workforce such as hourly contractors or warehouse workers with contextual assistance provided by AI. Ramp up your workforce fast and increase productivity.
Capacity allocation: Monitor and automatically adjust supply based on demand. Plug-in your data sources and external conditions, train LLMs based on your company's historical examples, and get suggestions to allocate your supply effectively.
Business intelligence: Analyze complex data and make faster decisions. Executives can get succinct reports and summaries from operational activities.
Shipment insights: Unlock insights on product deliveries with ad-hoc queries. Figure out reasons for delays, get suggestions to fix, and derive insights that are locked up in data stores.
IT
Code research: Get an understanding of your code base with natural language queries, so that you can ramp up new developers fast and increase productivity.
Code and documentation generation: Streamlined coding assistance and documentation generation for efficient development.
Test case generation: Generate test cases from use cases and code base to ship great products with full test coverage.
Engineering analytics and insights: Let AI provide suggestions on improving system performance by scouring through your logs.
HR
Policy research: Get HR policy information without searching through portals, documentation and information repositories to provide accurate information for your employees.
Employee onboarding: Ramp up employee onboarding, save time, and provide the best employee experience by giving AI assisted onboarding.
Candidate summary: Get consolidated candidate information ensuring you are hiring the best candidates for the job position.
Performance reviews: Get insights into employee performance, so that manager’s can save time and provide data driven feedback to employees.
Recruiting insights: Understand your hiring funnel to speed up your hiring process and reduce costs.
Those were a few use cases based on the state of the art advances in AI. Did I miss any use cases? Let me know in the comments.
Conclusion
AI is not a universal problem solver, but leveraged in the right way, gen AI has several use cases across departments such as sales, marketing, support, HR, operations, and IT. Now, integrating gen AI into all of those workflows requires a great deal of infrastructure and development time. What’s the effective and efficient way to integrate AI? That’s the topic of our next blog, which explores the problems with integrating gen AI into your company workflows and the optimal way to solve it.