To have the maximum impact, AI should be incorporated into existing workflows and processes rather than create entirely new ones. For example, an AI prediction about a prospect’s propensity to buy a particular product should appear alongside the prospect’s information within an organization’s CRM system.

The current “best” AI model for a use case may not stay that way in the future. As international data protection laws evolve and models become newly available across global regions, flexibility in choosing AI models is crucial.

It is imperative that enterprise leaders know how their data is managed and stored, and that they understand the legal implications of using generative AI models. Model providers that have first been vetted and approved by the enterprise’s legal, cyber, and AI teams should be prioritized.

Enterprises must consider the distribution of their data across systems. The challenge is not simply gathering the data, but in consolidating and validating it from various sources. For example, linking customer IDs across multiple systems can be a daunting task. The reliability of data from different sources must be considered. If the majority of your reliable data resides within Salesforce, for example, it would be logical to use built-in Salesforce Einstein AI capabilities. However, for enterprises with federated data, the choice between Salesforce Einstein AI and enterprise AI capabilities becomes more complex.

Evaluate your need for control in AI applications. Determine whether your scenario demands high customization to extract maximum value. Traditional approaches that rely on data scientists crafting bespoke models offer extensive control, but pre-built models offer efficiency. For example, Salesforce Einstein AI offers easily pluggable models, enabling swift deployment and immediate value realization.

Time is a strategic asset. Salesforce’s packaged solutions for rapid deployment and immediate benefits is one option that today’s leaders choose, particularly if speed aligns with their business strategy. Custom-built solutions offer unique tailoring, but can extend the time to operational value.

Factors such as data, security, regulation, and compliance require careful consideration. Some organizations prefer a customized solution for enhanced control and specific compliance requirements, versus a standard preset solution such as Salesforce.

Evaluate the reliability of a software platform’s developmental roadmap against your AI needs and timelines. Consider whether your in-house AI initiatives can meet demand without delays, and whether to engineer bespoke solutions or to leverage configurable, ready-to-use AI functionalities such as the ones from Salesforce.

This involves determining your team’s readiness to drive AI innovation and whether they prefer to engineer their own products, or quickly configure and move to the next project. The choice between Salesforce AI and enterprise AI can hinge on these factors, intertwined with your financial decisions and the organization’s culture.

Assess the effectiveness of your newly integrated AI. Before launching large-scale projects, it is important to test-and-learn with smaller use cases assessing AI’s effectiveness and to ensure your AI program delivers the outcomes you envisioned.

Your AI is only as potent as the data it is fed. Generating sales emails without customer-specific data – possibly external data, for example, will result in generic templates. The higher the quality and richness of the input data, the higher the quality of the output. Therefore, having a data strategy that supports your AI roadmap and use cases is crucial.

AI product capabilities are evolving rapidly. For example, Salesforce now provides AI updates at a much faster pace, meaning the capabilities available today will likely be surpassed in three to six months. To maximize the output from AI, you may need to discard some content or work over time as better product features emerge. Success lies in balancing current needs with future potential.

It is important to foster your team members’ growth into prompt engineers, or to hire externally. It is becoming an expertise all on its own. Extracting value from AI— particularly from Large Language Models (LLMs)—requires robust prompt engineering skills and strategies.