Skip to content
Fast Company Executive Board

The Fast Company Executive Board is a private, fee-based network of influential leaders, experts, executives, and entrepreneurs who share their insights with our audience.

How to make your business AI-ready

The journey to AI readiness is complex, requiring strategic investment in both talent and technology to maximize its potential. 

Man looking at a tablet

[Images: fizkes / Adobe stock]

Profile photo of Subbu Iyer

Those of us in tech have been contemplating, experimenting with, and even incorporating artificial intelligence (AI) and machine learning into our applications for years. However, the explosion of interest in generative AI in the past year has left even the most forward-looking leaders unprepared. This surge has highlighted a significant gap between the potential of AI and its actual implementation in both the enterprise and digital natives.

Consider this conflicting comparison: In an early 2024 study by Tata Consulting Services, most enterprise CEOs and senior executives anticipated AI would have an equal or greater business impact than the internet (54%) or smartphones (59%). Yet nearly a quarter (24%) haven’t moved past the exploratory phase and another 29% are in the initial stages of “cleaning up their data and moving it to the cloud,” rationalizing that it can be used for AI later.

This disparity between the potential of AI and its slow incorporation into production can be summarized into two core areas of readiness: mindset and infrastructure.

BREAKING DOWN THE AI-READY MINDSET

In a March 2024 BARC and IT Market Strategy report, the most frequently cited obstacle to AI strategy was a lack of AI skills and expertise among developers. What makes this gap different from past talent shortages is that AI requires an entirely different approach to development.

1. Navigate the programming paradigm shift: You don’t program AI; you give it instructions in English through a large language model. This is called “prompt engineering.” People emerging from university are well-trained in this technique, but many of us already in the industry have a ways to go. 

2. Build a tolerance for uncertainty: Traditional programming is predictable: You write code, and it operates consistently based on the inputs. However, AI development is inherently unpredictable. The success of an LLM, for example, is predicated on its ability to use context to deliver relevant answers such that two people asking the same question might get different, more personally relevant results. 

3. Remember the importance of guardrails: The variability that makes AI powerful also makes it prone to “hallucinations,” where it draws conclusions with no basis. AI models rely heavily on “fuzzy” searches, which draw associations without the time-consuming process of keyword and metadata engineering.

Developers must create guardrails and engineer prompts to minimize hallucinations. For instance, how do you know your answers are right? ANN (approximate nearest neighbor) is a class of algorithms for finding fuzzy matches. But how close is close enough before the answer is wrong?

4. Be open to an integrated ecosystem: AI’s success relies on multiple components, often outside your control. This includes real-time data streams, management of multiple data models, existing applications, and even open-source LLM models themselves. 

Ultimately, aligning your development team for AI is critical. The traditional mentor/mentee relationship of experienced developers with new grads is now more of a partnership that integrates the knowledge of classical developers with up-and-coming ones who are more familiar with new AI technologies.

CONSIDERATIONS FOR YOUR AI FOUNDATION

Beyond mindset, foundational infrastructure readiness is essential for executing an AI strategy. In its survey, BARC and IT Market Strategy found that “only 4% of respondents state that their company architecture is already optimized for AI,” and nearly half (44%) planned to “overhaul existing data management architecture.”

The importance of real-time data: Real-time data is vital to establishing the guardrails mentioned earlier. Business-specific and current data provides the necessary context that helps keep an LLM grounded. If your organization can’t harness real-time data effectively, it won’t realize the full benefits of AI or could make decisions based on inaccurate AI-derived conclusions.

Optimize for scale and volume: In a traditional development environment, only a subset of data is relevant for a given task. However, in a generative AI scenario, all data is valuable—particularly real-time data—for context and accuracy.

More data, faster, and all the time results in more accurate answers. That can mean efficiently ingesting and rationalizing data from tens of thousands of internal and external sources in real time and making it immediately available to other applications to query, potentially by thousands or even millions of users simultaneously.

CHOOSING THE RIGHT DATABASE FOR AI

As your organization considers these requirements, a critical component of infrastructure is the database(s) itself. While several types exist, these requirements quickly narrow the field:

Standalone vector databases:

LLMs and other AI models transform inputs such as images, video, audio, JSON, and text into a series of numbers (vectors) that can be searched for relevance. Unfortunately, vector databases are relatively new to market and frequently lack enterprise-class backup, security, and performance. They can be useful for contained situations, like a proof of concept or an internal use case.

Relational databases:

The workhorse of the database world, relational databases are an obvious alternative for high-performance or high-scale environments. Many relational databases have added support for vector data indexing and search, providing AI developers with access to enterprise database features and SQL-specific capabilities, such as advanced filtering, sorting, and processing.

While often faster than vector databases, relational databases fall down on scale and speed as they struggle to handle searches on very large data sets or by large numbers of concurrent users. Relational databases are best suited for small- to medium-scale AI projects.

Multi-model databases:

Perhaps the best option is a database that can support many data models, including key value, graph, and vector, given the wide range of data sources ingested. An AI platform that can handle any AI data processing requirement with the appropriate model improves control over searches because it can feed more relevant data to the models, thereby generating more accurate responses. Depending on the database, it’s likely also to have the enterprise-class features and performance support that the other options don’t. 

PREPARING FOR AN AI FUTURE

As your organization seeks to close the gap between the scale of your AI ambitions and readiness, it is vital to adopt the mindset and infrastructure to help you succeed. The journey to AI readiness is complex, requiring strategic investment in both talent and technology to maximize its potential.