Shape Your CX Roadmap With These Eight StepsShape Your CX Roadmap With These Eight Steps
An overwhelming array of AI possibilities for contact centers makes a structured evaluation approach imperative. Use this one as your starting point.
July 8, 2024
The impetus to evaluate what AI, and GenAI in particular, can do for customer service operations remains at an all-time high. Yet, organizations are grappling with the disconnect between vendors' touted capabilities and what can realistically be implemented. At Enterprise Connect 2024, someone challenged me to come up with practical recommendations for implementing AI in the contact center.
The challenge is that there are many use cases — over 25 to choose from, which I will detail in a follow-up article. Most companies only have the capacity for one or two proofs of concepts. The choice is even more challenging given that a sizeable percentage fails to progress beyond this initial phase, and those that transition into production are challenged to scale. Last year, Peter Bendor-Samuel, Everest's CEO, went as far as predicting that 90% of the pilots started will not make it to production in 2024. With such an overwhelming array of AI possibilities and high stakes, a structured evaluation approach is imperative. I recommend the following 8 steps to establish a prioritized shortlist and thoroughly assess your options.
Anchor your roadmap into your most pressing needs by selecting the most important Jobs To Be Done (JTDBs). If you struggle with prioritization, it likely indicates a need for better visibility, suggesting to start with understanding the what, why, and how of customer inquiries.
Once you have your "longlist," assess for each your data readiness. To avoid making the undertaking overwhelming, I recommend looking at the five main data buckets: knowledge, interactions, activities, outcomes, and customers, and performing a rapid assessment using a "good, bad, or okay" scale. As part of this stage, you can flag data improvement low-hanging fruits by marking them using a "can/should be improved" 4th option.
Map use cases to the type of AI model used: statistical, predictive, or generative. Statistical models analyze historical data patterns, predictive models forecast future events, and generative models create new data instances. Understanding these distinctions helps identify potential risks, transparency, and interpretability challenges.
Determine oversight requirements. AI recommendations can go directly to customers, be mediated by humans, or remain employee-facing. These 3 levels will define the type of oversight required.
Evaluate your options’ risks in terms of errors, compliance, and trust. You should include a legal review at this stage.
Define success metrics by translating your overarching revenue, cost, and CX goals into 1-2 specific, measurable targets for each use case.
Evaluate implementation approaches: build vs buy. With new models popping up constantly, building has become a more popular option: it gives you access to the latest innovation and lets you stay in control. However, it requires having access to development talent. Furthermore, you don't want to underappreciate commercial applications. They come with prebuilt models that can fast-track your time to results. AI can even come packaged as a read-to-use feature of some applications. Examples include speech, interaction summarization, or workforce management (WFM). It is important at this stage to not just evaluate the effort required to do a proof of concept (POC) and put it into production. You also want to keep an eye on the maintenance requirements, including the ability to make rapid changes and drive continuous improvements.
Address change management implications for employees and customers. Change fatigue is real and needs to be factored in to ensure proper adoption by agents. Identify process changes that require customer guidance.
These eight steps should now allow you to plot the various use cases on the table on a 2-axe chart, the vertical showing the potential business impact of the use cases and the horizontal one representing the effort required. You can now select pragmatically your first or next use cases. Proofs of concept will allow you to review and adjust their ranking.
