With every new technology that promises to make firms better, faster, cheaper, safer—whatever the promised benefit—there is the same question: to build or buy? To further confuse matters, the nature of AI also enables a third choice: build some parts and buy some parts. How does an executive decide which option is best?
I don’t think that AI is any different than previous technology cycles, and therefore we can take lessons from the past to help guide the decision. Executives in financial services need to think through a few considerations to determine which might be best for them, including:
Let’s consider how the answers to these questions can lead one to implementing AI via build vs. buy vs. a hybrid solution.
Let’s first discuss when it might make sense to buy an existing AI solution that suits your needs. We’ll start here because the majority of current AI adopters have purchased their solutions. Late last year Saifr published our own research that showed that most AI implementations rely on vendor-supplied AI tools and specialized RegTech solutions. Read the full report: From Caution to Action: How Advisory Firms are Integrating AI in Compliance.
Here are some cases where it is likely the best decision to purchase a market solution.
Need rapid time to market. If a firm urgently needs to modify their processes, buying is likely the best solution. Perhaps a new threat has been identified, a gap has been internally discovered or pointed out by a regulator, or other changes in circumstances require quick action. In situations where time to full implementation needs to be as short as possible, buying a solution in the marketplace makes sense. You can define your requirements, interview vendors, identify the best one, and maybe even run the new AI process in parallel with your existing process. All of this can be done long before you could ever create one internally and helps close whatever the gap was as quickly as possible.
Standard use case. If your use case is the standard variety in the industry, why divert internal resources to reinvent the wheel? Hiring an off-the-shelf vendor solution makes far more sense if you don’t need something hyper customized or specialized. Working with a vendor can even provide insights and efficiencies that your team might not have come up with if working alone. And even if your use case requires some small customizations, vendors can often accommodate those—for example maybe you need the AI slightly tuned to only flag medium and high risks.
Lack in-house AI talent. Of course, if you don’t have the required AI experts internally, or you do but their priority is elsewhere within the organization, buying is the best path forward. By partnering with an experienced vendor, you can quickly gain access to the most sophisticated AI. Buying also can give you more flexibility as AI changes or new firms develop better ways to use AI to solve your problem. The ability to switch vendors to take advantage of new and rapidly changing AI could be a great benefit as you work through how best to take advantage of this new tech.
Minimal data to create models. Even if firms have AI talent in house, many don’t have the data needed to create robust, specific AI models for their use case. For example, access to ~20 years of proprietary, industry-specific data is what lead to the creation of Saifr within Fidelity Labs. We had access to data that most folks in the industry didn’t know existed. With that data, in addition to other data sources, we were able to build our marketing compliance review models that can review content and flag up to 90% of what a human compliance reviewer would. If you don’t have the data, you just can’t create effective models.
Need predictable costs. If your budget is such that you really need to know the full cost of embedding AI into your processes now and in the future, buying a solution and negotiating a longer-term contract might be the way to go. It will be more clear what your financial obligations are likely to be. If you were to create a solution inhouse, it might be harder to determine the total costs and how those costs might change over time.
There are obviously more factors that you can consider, but these are some key ones. Don’t be surprised if in thinking over these questions, you determined that buying is the way forward. As our research showed, vendor solutions thus far have dominated early AI implementations. Our interviews made clear that resource constraints can be a big factor.
While internal development may be somewhat rare, there are factors that can lead firms to take on that challenge. Let’s review a key few here.
Unique competitive advantage. If using in-house AI can help your firm create a competitive advantage, then building it in house can help to secure, and potentially extend, that advantage. For example, you might develop AI solutions that allow your customer service reps to excel beyond industry standards, or that make your back-office operations far more efficient, or that maybe understand data in a way that helps identify trends no one else can. If the AI can help you to rise above your competitors, it can make sense to build it yourself.
Data is proprietary or sensitive. You should consider the privacy and safety of your data throughout its entire lifecycle when making your decision to buy or build—from processing to storage and everything in between. If your data is such that even working with an outside vendor is too risky, you may have no choice but to build. Creating your own AI can ensure you have a closed loop to help safeguard your data. By building yourself, you can still get the advantage of AI without the possibility of exposing propriety, sensitive, or personal data.
Unique use case. There are situations where your use case would require such extensive customization to a vendor solution that it doesn’t make sense to buy it. Or your use case is so uncommon in the industry that no vendor has developed it. Not to say that it might not be developed in the future, but if you can’t wait, you have to create it yourself. And, if there is a larger market for it later, you could even consider selling it to others as an additional revenue stream.
Strong AI expertise. If your firm has AI experts, who can be focused on your use case, and you have the needed data, you can consider developing your own AI solution. You can create a team of product managers, AI experts, data analysts, developers, project managers, AI governance, and financial folks to help ensure that the solution developed meets the needs, complies with regulations, is delivered in the time needed, and makes economic sense in the short and long term.
Lower recurring costs. There are situations where developing your own AI solution can be more cost effective in the long run. Perhaps the costs to develop your use case are more in the upfront development and not in the running of it long term, yet vendors aren’t priced that way. You should consider the long-term costs of in-house development and maintenance vs. working with a vendor.
Maybe as you consider the points above, you aren’t clearly in one camp or the other. You see benefits to building your own and buying off the shelf.
There is also a hybrid choice where you build what is unique to your firm and use vendor AI models to plug any holes. Perhaps you don’t have the AI data or expertise to create effective models, but your system is so unique that you can’t use off-the-shelf solutions for the entire use case. Maybe the use case is just one part of a customized system that manages your phone representatives, or your trades, or other complex processes.
Hybrid solutions can involve using APIs from AI expert firms to provide the benefits you need. You can also use offerings like the Microsoft Azure AI model catalog to gain access to models. For example, four of Saifr’s AI content review models are available within the Microsoft Azure AI model catalog, making it easy for developers to incorporate them into their solutions. With this collaboration, firms can get the benefits of Saifr’s sophisticated AI to help them produce more compliant content much faster while helping reduce risk. Going hybrid can deliver the best of both worlds.
From an emerging technology perspective, AI isn’t much different from what we've seen in the past—for medium-sized and smaller institutions, it is most likely preferable to buy this new technology, not build it.
However, for larger institutions with more resources, it is not always a clear-cut case. There are instances where, given the time to value that they’re trying to realize, it might be easier and better to buy. However, there are some use cases that are so specific to a business, where the build decision makes the most sense. So, it’s more a case-by-case basis for larger firms.
Hybrid solutions are also a great choice for some firms. They can develop parts of the solution that are highly customized or need to be private and still build in the AI capabilities they want. The hybrid approach enables firms to access the benefits of vendors, that have the data and deep AI expertise to create sophisticated AI models and maintain them.
The bottom line is, AI is here to stay. Whether you build, buy, or go hybrid, we believe it’s time to get started with it, otherwise you may risk falling behind your competitors.
The opinions provided are those of the author and not necessarily those of Fidelity Investments or its affiliates. Fidelity and any other third parties are independent entities and not affiliated. Mentioning them does not suggest a recommendation or endorsement by Fidelity. The information provided regarding AI applications is for informational purposes only and is not intended to constitute a recommendation, development, security assessment or advice of any kind. Please perform your own research. Saifr's products and services are not intended to replace the user’s legal, compliance, business, or other functions, or to satisfy any legal or regulatory obligations. All compliance responsibilities remain solely those of the user and certain communications may require review and approval by properly licensed individuals.
1206412.1.0