Assembling Your AI Strategy: A Guide

It seems companies today are rushing to prioritize AI projects. However, in a survey we ran, we found that only 9% of organizations said they had managed to ship AI products to production. What's holding them back?

There are multiple factors that create roadblocks to shipping generative AI apps. Even a team that's highly-motivated to release solutions may find itself stymied because it doesn't know how to overcome a given technical or legal issue.

A comprehensive AI strategy provides your organization with a firm foundation for the success of its AI initiatives. It acts as a road map that teams across your company can use to ship new solutions with confidence. 

In this article, we’ll dive into why you need an AI strategy, as well as how to assemble and put it into practice.

Why you need a comprehensive AI strategy

An AI strategy serves several purposes:

It prioritizes projects and funding.  With an AI strategy in place, teams can clearly see which use cases are top of mind for the organization and where the company sees the greatest value from AI investments. This can give teams that are unsure where to begin a jumping-off point for brainstorming and discussion.

It removes roadblocks to starting projects. Teams might be hesitant to pursue AI projects for a number of reasons. For example, without policies around responsible use of AI, they may hesitate to release solutions that could endanger customer privacy or even violate local data regulations. Or, they might not know what an acceptable quality bar is for a production release. A strategy can help teams address these potential blockers early in the design process.

It provides a roadmap for AI implementation. Not every team will have the resources to create everything they need—including guidelines and tools—to power AI projects. Creating an AI strategy and associated support systems means every team doesn't have to bootstrap themselves from scratch. 

Tips for assembling your AI strategy

There's no right or wrong way to go about assembling an AI strategy.  However, as an AI development platform company that works with many customers both small and large, we've seen a few patterns emerge.

Here are a few big buckets we’ve found it helpful to focus on when working with our customers:

• Identify priorities
• Drive your strategy from lines of business 
• Create a data strategy
• Select tooling and create capabilities
• Hire the right people with the right skills
• Define your approach to responsible AI

Let’s look at each one of these in detail. 

Identify priorities

Identifying overall priorities and potential opportunities gives your team guidance on where they can make the best investments. These priorities should be regularly revisited and updated as your AI capabilities mature.

For example, our surveys show that 39% of organizations identify enhancing customer service as an initial stage for starting out with AI projects.  Indeed, applications like chatbots are a common AI use case, with a wealth of examples and documentation available to support them. 

That makes something like a customer service chatbot a great way for your company to become familiar with: 

• The technical steps involved in building AI apps; and 
• The steps involved in shipping a successful AI application to production and supporting its ongoing maintenance 

Once you've tackled this or similar entry-level use cases, you can shift your attention to more advanced capabilities. One popular example is real-time decision-making —for example, using customer demand data to personalize sales offers you present to customers.

Drive your strategy from lines of business

In our State of AI report, we found a lot of companies are still driving AI projects from their IT departments. We consider this an antipattern. While IT plays a central role in both architectural and tooling support, it risks becoming a choke point if they have to lead every initiative.

For mid-size to large companies, we’ve found a better strategy is to form a Center of Excellence (COE) that assists teams with tooling, strategy, data, and other facets of GenAI apps. Line of business teams are much closer to a business's problem and the needs of their customers. That means they can more effectively discover new and innovative use cases for AI. This bottom-up approach, in the long run, leads to greater innovation.

COEs can best support teams by creating both tooling and architecture as well as communicating best practices. Such best practices may include:

• How to use LLMs in the highest performance, most cost-efficient possible
• Choosing the best technical implementation for a given use case. For example, when using retrieval-augmented generation (RAG) to provide contextual data, COE engineers can guide teams on when to use a graph versus vector approach
• Emphasizing human-in-the-loop approaches and other techniques for improving data and response quality

Create a data strategy

GenAI applications typically require large volumes of high-quality data to be successful. This is true even if you're leveraging a commercial or open-source large language model (LLM) and adding context with techniques such as RAG.

Our work and research have found that, as an organization's AI capabilities mature, the lack of high-quality data quickly becomes a blocker. Over thirty-eight percent of companies we surveyed who identified themselves as further forward in their AI maturity arc ranked data quality as a significant issue.

Creating high-quality data for AI requires having firm and consistent data standards, along with the tooling required to preprocess, transform, prepare, select features, and reduce datasets. To learn more about this topic, check out our guide on getting data AI-ready.

Select tooling and create capabilities

The COE can also help enable teams by vetting and sponsoring toolsets that a variety of AI developers can adopt to build apps quickly. These may include:

• Low-code and no-code tools that simplify application development for everybody and enable first time AI developers to be productive
• Scalable database systems, such as a vector database, to ensure that AI solutions can grow to meet rising demand

The goal here isn't to restrict teams that are farther along in their AI maturity from adopting cutting-edge tools. Rather, it's to provide a baseline capability that all teams can leverage regardless of their maturity level.

Hire the right people with the right skills

No matter where a team or company is in their AI journey, many companies we've talked to say that finding the right people is their biggest obstacle.  Meeting the rising demand for AI solutions means hiring aggressively for the right skills:

• Programming language proficiency (.e.,g knowledge of Python, Node.js, SQL)
• Knowledge of data structures, data processing algorithms, and data transformation tools and techniques to enable building RAG pipelines or fine-tuning LLM models
• DevOps and system reliability engineering (SRE) skills for building CI/CD release pipelines and managing solutions in production

Not everyone working on AI apps needs all of these skills. That’s an almost impossible demand. However, you should make sure that teams building AI applications are well-rounded enough to face any challenge that might come their way.

Define your approach to Responsible AI

Responsible AI means following a set of principles that establish trust in AI solutions with customers and stakeholders. It consists of three pillars:

• Accuracy - Verifying that data is timely, consistent, and correct
• Security -  Addressing baseline application and data security as well as security risks unique to Gen AI apps, such as the abuse of LLM for fishing or spearing attacks, or data set attacks such as data poisoning
• Compliance - Ensuring that personally identifiable information (PII) isn’t included in data sets and can’t be coaxed out of an LLM

Just as every new application should undergo a security review, every new GenAI app should undergo a Responsible AI review prior to release to ensure conformance to the company’s standards. 

The right tools to power your AI strategy 

Implementing an AI strategy requires more than just having a set of policies and procedures in place. It requires tools to make prototyping, developing, and releasing scalable GenAI applications easy.

The DataStax AI platform is a scalable, serverless AI development system that combines our scalable Astra DB vector database technology with no-code/low-code approaches to developing GenAI apps powered by Langflow.  Contact us today to learn more about how DataStax can power your AI strategy.