5 questions to ask before building your own AI solution for CX
The build vs. buy debate in software is nothing new. With the rise of large language models (LLMs), spinning up prototypes that can complete tasks has become easier than ever. However, developing an AI application for CX that’s scalable and production-worthy demands extra consideration. In this post, we’ll explore key questions to ask when deciding whether to build or buy generative AI solutions for CX.
While the right answer depends on industry, use case, and business goals, it’s important to carefully weigh your options.
Why building an AI application for CX requires extra consideration
The reason for this has a lot to do with the nature of generative AI.
Unlike most software development where the goal is to create a deterministic system with the same, expected outcome every single time, generative AI generates new information and content based on patterns, often producing varied responses that can be difficult to predict.
This variability is what makes gen AI applications so powerful, but also adds complexities in maintaining consistency and control.
Differing from standard applications where the workflow is predefined, generative AI models require continuous monitoring, training, and refinement. There is a misconception that generative AI just needs instructions. But in fact, they need much more, especially if you are training on your own data.
The flexibility makes them ideal for tasks that benefit from creative problem-solving or personalized engagement, but it also means that development doesn’t stop at launch. This, compounded by the fact that CX applications can ultimately impact a brand’s relationship with its customers, brings additional layers of considerations when deciding whether to build the application yourself or to partner with a vendor.
While the following is not an exhaustive list, here are some top questions you should consider when making a decision.
Is building AI applications part of your business objectives?
If building AI applications aligns with your business objectives, you’re likely prepared for both short-term and long-term investment in development and maintenance that will be absolutely critical for AI applications, ensuring the solution doesn’t become a burden on your resources.
In contrast, if developing your own AI applications isn’t a core objective, partnering with a vendor may be the more sensible approach. Vendors are equipped to provide solutions that fit your needs without the complexities of internal development, allowing you to focus on your primary business goals.
Based on McKinsey & Company’s recent estimates, building and maintaining your own foundational model could cost up to 200M with an annual recurring cost of 1-5M. Using off-the-shelf solutions or fine-tuning existing models with sector-specific knowledge can dramatically reduce this cost.
What’s your timeline for deployment?
If your deployment timeline is urgent—perhaps due to a competitor already leveraging AI solutions—working with an AI solution provider can help you hit the ground running and accelerate implementation. A vendor with solid experience in enterprise AI development will help you avoid unnecessary trial and error, minimizing risks associated with AI safety - such as AI hallucinations.
Instead, leaders should strongly consider partnering with gen AI solution providers and enterprise software vendors for solutions that aren’t very complex or [industry] specific. This is particularly critical in instances where any delays in implementation will put them at a disadvantage against competitors already leveraging these services.
- McKinsey & Company, in "How generative AI could revitalize profitability for telcos"
Do you have the internal expertise, resources, and infrastructure to build, scale, monitor, and maintain an AI solution?
While widely available large language models (LLMs) have significantly accelerated the process of building a working prototype on a laptop, this does not equate to having a production-ready, scalable solution that can effectively address your business needs [link to last mile]. Additionally, the demand for ongoing maintenance means that development does not stop after the application is launched.
Developing AI applications requires a skilled team knowledgeable in machine learning, data management, and software engineering, along with the necessary technological resources and datasets. Even choosing the right LLM to use for the best fitted use cases would require a good understanding of how various LLMs differ. Experience working with AI solutions is also crucial for successful deployment in an enterprise context.
In the last two months, people have started to understand that LLMs, open source or not, could have different characteristics, that you can even have smaller ones that work better for specific scenarios
- Docugami CEO Jean Paoli, in CIO “Should you build or buy generative AI”
If your organization lacks this expertise or the infrastructure to support such a project - not just at the prototype stage, but also to scale, monitor, and maintain the solution in the future - it may be more prudent to consider vendor solutions that can provide the capabilities you need without overextending your internal teams.
Also consider, if problems arise, do you have the manpower and expertise to investigate the root cause and implement long-term fixes, or will you be crossing your fingers and hoping the issues don’t recur? A strong internal team capable of addressing challenges as they emerge is necessary for ensuring the reliability and effectiveness of your AI application for CX. Without this capability, you risk operational disruptions and even diminishing trust in your brand.
What ROI do you expect from your AI application? And how comfortable are you with the associated risks?
When considering your expected return from an AI investment, it’s essential to balance potential returns with the associated risks.
Agent-facing projects generally carry lower risks, as the AI solution won’t directly interact with customers. This allows for more trial and error, with agents able to provide feedback on the AI application’s performance. That said, such solutions might yield only incremental gains in agent productivity with less or no impact on the customer experience, and do not take advantage of the full capabilities of gen AI.
In contrast, customer-facing gen AI applications can offer a much better return because they can directly improve customers’ ability to self-service and, in some cases, resolve their issues directly. Here are the kinds of results a customer got when deploying generative AI agents with the capability to resolve Tier-one tasks.
Allowing AI to handle a broader range of tasks introduces added complexity. While there are risks—such as AI hallucination, where the system may generate incorrect or irrelevant responses—these challenges can be managed with the right approach. Having a strong internal team that can tackle this, or choosing an experienced vendor with a well-informed strategy for handling AI behavior, will ensure guardrails are placed around customer-facing interactions — so you can get the most out of your AI applications with confidence.
Making a thoughtful decision
Ultimately, the choice between building or buying an AI solution should align with your organization’s long-term vision. Each option carries its own set of challenges and opportunities, and taking the time to assess your specific needs can set the stage for success.
Considering the evolving landscape of AI, it's not just about deploying technology, but also ensuring that it fits well into your operational framework and readies you for the future. With careful evaluation, you can make a choice that enhances your customer experience. Whether you decide to build or partner, the key is to stay focused on your goals and embrace a strategic approach to generative AI applications.