While everyone in the CFO community is aware of the potentially massive transformative effects of generative AI — especially implications for operating efficiency, which is always top of mind — there has been less practical advice offered. Below I outline some key considerations CFOs can use today to drive immediate business impact and get in front of the AI wave while continuing to build enduring businesses.

If implemented and leveraged correctly, generative AI can be a true game changer for a business, but as with everything else it requires a thoughtful, measured approach.

With so much focus on AI, boards of directors, investors, and management teams will expect that every company has a strategy to take advantage of the developments in generative AI. As a finance leader, you’re not just focused on reporting, you’re a thought partner to the entire executive team, and your team leads the way in execution. As a result, you not only need to help guide your company’s broader AI strategy across product, customers, and internal operations, but you also will be expected to outline how your finance and accounting teams are going to leverage the technology to improve internal operations.

For many finance leaders, this may simply mean adopting vendors or utilizing existing vendor tools that employ generative AI. The expectation is not that every CFO — or even tech company — is developing their own AI tools. However, you should have a sense for what quantum of productivity improvements are realistic by having conversations with your technology and operations teams. As vendors across your CFO software stack release AI products, have conversations with them and your FP&A team to understand productivity savings. At a minimum, these conversations are inputs to your “AI strategy.”

Developing and rolling out an AI strategy that extends beyond features rolled out from existing vendors is not something you should do haphazardly. If you are serious about using AI either internally or as part of your product suite, create a centralized team dedicated to it. In addition to technical folks from Engineering, Product & Design, this team should absolutely include Compliance & Finance (more on why below). This team should be responsible for quickly shipping new, high-quality AI experiences while discovering the right customer interaction patterns, internal or external.

The leader of this team should be a senior leader, preferably someone who has been at your company for a while and has a lot of institutional knowledge. Hiring an external “hot shot” rarely works, and it’s unlikely to work with AI. Over time, this team should dissolve themselves and empower all teams across your company to build on top of their efforts. At Brex, we have run this playbook a few times, most notably when we rebuilt our mobile experience, and it’s worked great. As a CFO, this is not an area where you should skimp. Again, if your company is serious about AI, this is a strategy that should bring about a lot of ROI in a short period of time.

If you’re a startup that is currently burning money, one of your most valuable assets is a high (and preferably increasing) gross margin, as this indicates future profitability to investors. Building AI applications generally requires a meaningful investment in cloud services and compute, which could hurt your margins. Costs can add up fast and make it harder to achieve profitability in the future. You should not automatically assume that the incremental cost of AI features is immaterial — examine the costs of any new AI features or products and plan to adjust your pricing accordingly.

Understand your role and ability to add value in the broader AI ecosystem. For instance, very few companies should be developing proprietary AI foundational models. You can leverage existing solutions and still offer something new and useful for your customers.

Company data has always been valuable, and the interaction between finance teams, business intelligence, and data drives important insights. However, with the rise of generative AI, the importance of proprietary data is paramount and potentially directly or indirectly monetizable.

Consider the proprietary data your company and product maintains (product usage, customer attributes) and its value in potentially training a model to perfect features or use cases of your product. Also consider what rights and access you are giving to vendors to your data, including companies offering generative AI models (see vendor lock-in below). Any use of customer data must be carefully considered from a legal and compliance perspective, as also covered below.

With the rise of SaaS and the cloud, many companies find themselves “locked-in” to the cloud vendor (e.g., Amazon, Microsoft, or Google) because so many of their applications have been built on top of the cloud offering. This complicates price negotiations and renewals, given the difficulty many companies would have if they migrated cloud vendors.

For generative AI, there is a similar concern, especially given the fast-changing nature of the space: today’s leaders may be tomorrow’s laggards. As you build out your offering, work with technology leaders in your company about the desire to avoid vendor lock-in to the extent possible. Invariably, there will be a tradeoff between speed of execution and careful planning around lock-in, but this is a tradeoff you should make explicitly and with eyes wide open.

While AI is not new for most companies (for example, at Brex we’ve used it to inform decisions on credit underwriting and fraud management for years), the last 6 months have brought about a step function change in potential use cases for the technology. And things are moving very quickly. In response, make appropriate investments in regulatory and compliance resources early on. You don’t want to get caught flat footed and miss important changes in policy and regulation, of which there are many.

For areas including financial data, intellectual property, and customer privacy especially, winning customer trust and taking careful precautions are paramount. If you’re using actual customer data in fine-tuning your large language model (LLM) for example, it’s both tricky and risky because without the right precautions, you have limited to no control over how this model will regurgitate the data. This requires caution and careful scrutiny, as an accidental leak or comingling of customer data will be the equivalent of the large data breaches that caused huge reputational damage last decade.

If your company wants to use generative AI to drive productivity to create a better customer experience, create well-defined policies that take into account the risks of using AI, potential use cases, and your company’s risk tolerance. It is best practice to treat these policies as public-facing documents, and a trend among many companies is to actually publish them (Brex plans to do so later this quarter). These policies are typically driven by legal and compliance departments, but take input from others including technology and operations. Implement quality control and apply human oversight and careful review of any outputs before they are used.