Conversational AI fragmentation and finding the sweet spot
Chief Evangelist at Human First, Cobus Greyling asked last August, “Will Conversational AI Implementations Become More Fragmented?”
Count me in the resounding “yes” camp on that one. As I shared in my response to Cobus’ question at the time, “Going, going, gone are the days of the monolith CAIP.”
With the ever-increasing emergence of vendors across niche areas of the CAI technology landscape, from speech-to-text to natural language understanding and beyond, the odds are high that building incredible CX will mean pulling from this growing list of world-class technologies.
One might even find use cases entirely comprised of wholly disparate, loosely coupled services acting in concert with one another through intelligent orchestration. Bring your own NLU. Bring your own STT. Bring your own TTS. Bring whatever you’ve got, or whatever you think is best for the specific task at hand.
It better be one hell of an orchestration engine…
But even if it is, and even if you could pull together solutions composed of different services and technologies, across different platforms, networks, and cloud providers. Is that really the optimal path?
I’ll be honest, I’m skeptical. Perhaps I’m becoming a bit too long in the tooth. Jaded and cynical from decades of solution design and implementation. But in my experience, fragmentation can very quickly spiral into chaos.
Gartner called out in a recent publication in regard to CAI adoption:
“While the benefits of conversational AI are compelling, the technology is still maturing. A fragmented vendor landscape and complexity of deployments will result in measured adoption through the next two years.”
So yes, the fragmentation is real, and in many ways, it is the natural evolution of our industry; one that is transitioning beyond a small cadre of behemoth platforms to a much larger collection of what Mr. Greyling calls “vertical vectors”. Niche products and services that support the wider CAI use-case or solution ecosystem.
As exciting as this explosive growth is, that fragmentation if left unchecked, will eventually start to erode performance, scalability, and speed-to-market. You simply cannot eat all there is at the buffet.
The sprawl of services, systems, and applications across disparate networks and platforms will, without fail, introduce increasing complexity. Even the most elegant and sophisticated CAIPs designed with this multi-vendor, multi-tech model in mind, will provide only limited abstraction and unification.
At the end of the day, you will need deep knowledge and expertise in each component. And thus, you’ll need a larger more diverse team. You’ll need to manage vendors, maintenance, upgrades, and upkeep across a multitude of suppliers. As you scale your solutions up, as you scale your component library up, so too will you scale up the management and maintenance overhead.
To be clear, I am not arguing in favor of a return to the “all-in-one” platform (if such a thing even exists anymore). There are too many incredible solutions out there that demand we architect across multiple vendors and applications.
But as with most things, the extreme ends are dangerous whichever side you happen to be on. Consistency, stability, and manageable scale lie somewhere closer to the center. The happy medium. The “goldilocks zone” if you will.
Recently I found myself on an analyst call discussing this very topic. Weighing the pros and cons of hyper-extended use cases and this burgeoning field of niche CAI tech vendors and solution providers that are laser-focused on a specific industry or a specific area of technology. “Specialization leads to optimization,” he said. I couldn’t have agreed more.
In Deloitte’s latest “The State of AI in the enterprise” they note “83% of responding high-achieving organizations use two or more types of ecosystem partners.” and offer the suggestion; “Keep things complicated. Too few external partnerships can make it difficult to part ways with vendors if needed in the future.”
I think they make a good point, but I also think there is a huge difference between “2” and “more”.
One does not reduce the risk of vendor lock-in by simply adding more vendors ad infinitum, and even if that were the case, the trade-offs are risks that well outweigh the reward.
My advice, look for platforms and adjacent technologies that have taken the time to truly partner and integrate.
Look for a history of proven interoperability and scale. Look for a composition of a select few, where the synergy is evident, and the commitment is strong. Look to consolidate where you can and seek the happy medium between best-in-breed and a unified architecture that trends towards a more limited scope of technologies and finds a scalable path to optimization.
Don’t get me wrong, I understand the appeal of platforms that are happy to be put anywhere (my cloud, your cloud, their cloud, no cloud) and leverage any CAI adjacent service (this NLU, that NLU, my STT, your STT).
But don’t let this seeming flexibility fool you. You simply cannot optimize for everything. Fragmentation left unchecked will lead to chaos, and your path towards scaling ROI will get steeper and steeper.
Keep your vendor list short. Looks for proven track records. Look for real partnerships across those vendors and make sure their architectures are tightly aligned and ripe for optimization.