What can AI do for your company? With so many developments, it’s easy to feel overwhelmed by AI’s possibilities.
Stepping back from our work helping organizations to use AI to streamline operations, we've observed patterns across use cases, allowing us to develop a framework that categorizes AI skills.
This framework helps in two ways. First, it provides a structured way to understand AI’s broad capabilities. Second, we have found it helps organizations identify new opportunities that may not have been obvious.
We hope that you will find similar value in it.
A framework for AI skills:
Note: This framework covers knowledge tasks and excludes AI skills in the physical world.
1. Generation: Create and modify text, audio and visual content
You know what this is - the “Gen” in GenAI - and you likely already know about some of the many ways companies are using this capability:
Adore Me: Uses AI to generate on-brand product descriptions at scale, enabling fast content creation across international markets.
Duolingo: Uses AI to accelerate software development and enforce code consistency.
Mango: Uses AI models in place of human models to speed production of marketing images.
2. Conversion: Convert content between mediums and languages
You may have experienced improved closed captioning on various video platforms; it is now incredibly low cost to transcribe content from audio to text; and it’s bi-directional, with customizations around voice and intonation, along with the ability to provide accurate translations. Some examples:
Taco Bell: Uses AI to serve drive-thru customers in their native languages, increasing speed to order completion and delighting customers. This example covers both transcription from voice-to-text (to enter orders into the kitchen), along with multi-lingual translation.
TIME: Uses text-to-audio tools to create automated voiceovers for news articles and blogs.
3. Classification: Classify items by defined segmentation criteria
Even general purpose LLMs are great at classification -- try asking it to categorize a list of items. But when LLMs are fine-tuned for specific use cases, it enables incredibly consistent, accurate classification that can then automate work that would take us humans a mind-numbing eternity:
Doordash: Uses LLMs to add product tags to tens-of-thousands of grocery items, allowing users to filter on more attributes and increasing the accuracy of search results.
Grammarly: Uses AI to classify customer service messages based on their content, ensuring messages are routed to the right support agents and product managers.
4. Validation: Compare data sources to identify discrepancies
Mistakes are not only annoying, they are costly. That’s why companies add checkpoints into processes – prevention is the best medicine. The downside is that manual checkpoints drive an endless stream of repetitive work. But since LLMs are smart enough to spot tiny differences…
Dow: Uses AI to compare shipping invoices with purchase orders, flagging discrepancies such as incorrect fees and contract mismatches.
Doordash: Uses AI to compare in-app product descriptions with labels of actual products, reducing mismatches that lead to product returns and frustrated customers.
5. Inquiry: Answer questions using knowledge and data sources
As LLMs evolve, they are getting better at answering questions; see ChatGPT’s reduced propensity for hallucination. In parallel, advances in RAG (retrieval augmented generation) and fine-tuning have enabled companies to offer on-demand inquiry services:
Klarna: Uses AI to answer customer service chats, answering common questions and retrieving real-time information about customer orders.
PE firms: Use AI to accelerate due diligence, reviewing large volumes of information quickly to broaden and deepen the inquiries that can be completed in short time frames.
6. Synthesis: Distill insight from large amounts of information
Sometimes consuming the full source material matters; other times, a summary suffices. LLMs got famous generating text; they are also great at boiling things down:
Tegus: Uses AI to summarize financial reports and expert calls to distill insights with its AskTegus product.
Youssef + Partners: Uses AI to accelerate the work of poring through thousands of pages of legal documentation to shape arguments.
7. Recommendation: Make suggestions based on data
AI has long excelled at recommendations—Amazon and Netflix led the way in suggesting products to buy and shows to watch. What LLMs have changed is the cost and accessibility, making it far easier to build recommendation systems at a much smaller scale:
Ulta Beauty: Using AI to personalize outreach to existing customers.
Databricks: Combining LLMs with existing product metadata to help their customers develop recommendation engines.
8. Analysis: Recognize patterns; make predictions or optimizations
Similarly, AI has been great at analysis for a long time - ‘big data’ and ‘machine learning’ were the buzzwords du jour not long ago. What’s new is that LLMs can acquire and synthesize data in real time, and interpret unstructured data without feature engineering. As such, we see:
Instacart: Uses AI to predict the impact of live events on delivery demand and travel times.
Recursion: Uses LLMs to perform complex drug discovery tasks.
What AI skills would most help your company?
As we discussed in our framework for jobs, AI can automate many of the knowledge tasks that humans handle today, enabling companies to enhance efficiency, improve decision-making, and unlock new growth opportunities. Leaders are focusing on identifying manual processes that can be streamlined or fully automated using AI. By systematically evaluating the categories of work AI can address—through structured assessments, case studies, and real-world applications—organizations can uncover opportunities that, in retrospect, will seem like obvious areas for AI-driven improvement.
Which AI skill from this framework will be most impactful in your organization?