The initial excitement surrounding AI’s experimental phase is giving way to a stronger focus on real-world applications and return on investment, as technology retention among the enterprise is growing. According to the Summer 2024 Spend Report by Ramp, in Q2 2024, the average accounts payable expenditure on AI vendors surged by 375% compared to the previous year, indicating a trend towards larger and more prolonged investments in AI applications.
As spending in this category continues to grow, enterprises need a practical framework to ensure they are integrating AI where it will have the most impact. Inspired by proven strategies adopted by dozens of our portfolio companies focused on the Enterprise segment, I propose a straightforward, four-step approach for enterprise technology leaders:
A practical framework for AI integration
Step 1: Identify which tasks the proposed AI solution is most suited for.
Large Language Models (LLMs) like GPT-4 are powerful in many areas, like natural language understanding and generation, content creation, and code generation. However, there are a number of tasks where they have limitations or are not well suited – tasks related to mathematical precision and complex calculations, understanding and processing non-textual data, or reasoning or logical deduction.
As a first step, you need to take a first-principles approach to assess which specific tasks within your department the proposed AI solution is most suited for, and which tasks can be effectively handled by an LLM. Each AI solution will have a specific set of skills. Where do those skills apply best? In each team or function, list out all the ‘jobs to be done’ and rank them from areas where the proposed solution is most suitable to least suitable. If you are in charge of lending, for instance, LLMs may be helpful when it comes to loan underwriting and servicing but not loan origination.
Step 2: Evaluate the potential impact of each of the ‘more suitable’ areas.
For each of the suitable areas identified, ask yourself where the proposed AI solution would provide the biggest impact? What are the most acute problems? Where are the biggest areas of opportunity to make money, save money, mitigate a critical risk, or dramatically improve a process or experience? In other words, what is the business case?
Recently, Amazon CEO Andy Jassy shared some remarkable statistics regarding the impact of the company’s Amazon Q gen AI assistant on generally tedious software hygiene tasks by his team. They were able to save 4,500 developer-years of work and ship 79% of auto-generated code reviews without any additional changes for an estimated $260 million in annualized efficiency gains. Talk about a solid business case.
Start by listing out the suitable areas you identified in the first step. For each area, estimate the dollar amount in additional revenue, cost savings, or time saved. Also factor in opportunity costs. In the Amazon example above, how much more would Andy Jassy’s team be able to accomplish with the additional time saved from not having to work on updating foundational software?
Step 3: Evaluate the risk of the proposed AI solution getting it wrong.
As previously mentioned, different AI solutions will excel at performing certain tasks and struggle with others. For many use cases, the level and consistency of accuracy the solution can achieve is a crucial factor to take into account.
In a customer service use case for a fashion retailer, as an example, if a chatbot were to answer a question incorrectly or not know the answer to a query, it could at worst lead to a poor customer experience. However, the upside is creating massive efficiencies for a low-risk, high-cost center, potentially saving the retailer hundreds of thousands of dollars.
However, in regulatory or compliance use cases, a task being performed by AI incorrectly could lead to millions of dollars in fines, or worse, the organization in question losing their license.
For this step, make a list of tasks that the proposed AI solution is particularly good at handling, that would significantly benefit from the AI’s involvement without introducing any unexpected and serious risks. The emphasis is on identifying areas where the AI can make a clear and substantial difference without causing new problems.
If there are areas of the business that are high-risk but would benefit exponentially from the AI solution, consider ensuring there is a human-in-the-loop aspect and that every step has the opportunity for quality assurance.
Step 4: Experiment until you prove the value before scaling.
Once you have your shortlist, it’s time to run experiments. This should go further than prototyping to full-scale production. This can prove challenging, but is crucial to truly understand the benefit of the AI solution and ensure there are no catastrophic consequences.
For inspiration, look at Palantir, a company known for its cutting-edge AI and data analytics solutions that have revolutionized industries to the tune of $2 billion in revenue in 2023. Their secret weapon? AIP Bootcamps. These immersive and hands-on-keyboard sessions allow customers to build alongside Palantir engineers, all working towards the common goal of deploying AI in operations.
The team does a deep dive to understand how to apply AI to mission-critical operations, then develops initial use cases in AIP, and then trains and onboards users for roll-out. AIP Bootcamps involve full-scale production trials, including rigorous testing and adaptation, to ensure organizations can implement AI effectively over time.
Using this approach, they worked with ONE Mount Sinai department at ONE hospital to prove that they can get patients out 1.5 days quicker, free up 20 additional beds, and save $1.5 million. Mount Sinai has 8 hospitals in their network. These production trials are well worth their while.
An adaptable framework for any use case
Overall, this 3-step framework not only tests AI’s real-world applicability but also aligns key stakeholders to the project’s goals, ensuring a smoother transition to operational use and scaling AI solutions effectively across the enterprise.
As AI technology continues to advance marked by developments like smaller, more efficient models and the rise of open-source platforms, this framework can be used again and again over time, allowing organizations to adapt their AI strategies in response to new advancements, ensuring innovation and practical application remain at the forefront.
Navigating the complex landscape of disruptive technology requires a clear and thoughtful approach, one that helps focus on functions where AI not only fits but can also deliver transformative results, ensuring investments align with strategic business objectives and yield substantial benefits. By continuously evaluating and refining AI strategies within your organization, you can harness AI’s full potential to drive growth, innovation, and operational excellence.
Be First to Comment