No Jitter asked Derek Top, Principal Analyst & Research Director, Opus Research for his thoughts on a few questions related to generative AI trends in 2024 and how enterprises might select a Gen AI solution.
No Jitter (NJ): What were some of the major trends/developments in Gen AI during 2024?
Derek Top (DT): This year we saw an almost never-ending list of vendors announcing product offerings to create, deploy, and maintain GenAI-powered agents that can interact with customers autonomously across voice and chat channels. These AI assistants are integrated with more data sources, tailoring it to business needs and expanding use cases in customer care and CX. We can expect this trend to continue as GenAI drives even greater efficiencies by automating routine interactions, streamlining ticket resolution, and enabling smarter, real-time decision-making.
NJ: Launches of AI agents for customer service, internal knowledge workers, etc., became more prevalent in the latter half of 2024. What are AI agents and how do they differ from conversational AI or Gen AI-powered bots?
DT: The bottom line is that with AI agents, organizations don’t have to build out a workflow but instead just give an AI agent a goal and access to tools and then it figures out how to accomplish the task based on its prompt. AI Agents understand instructions in plain English and then carry out all the sub-tasks required to get the job done. AI agents simplify the process to retrieve and analyze data, even when it is stored across multiple enterprise systems. But accessing those enterprise systems – either through integrations with customer data platforms or workflow APIs – will remain a significant challenge. While the promise of agentic workflow automation sounds appealing to many CX organizations, the reality is implementations will require careful planning and lots of human-AI collaboration.
NJ: It seems like enterprises can 'standardize' on Microsoft/OpenAI or Google Gemini (or others) while some enterprise platform providers (e.g., ServiceNow, Salesforce) offer connections to all (or most) foundation models, and yet others like Zoom, Cisco Webex, RingCentral, etc., have their own models and most offer ‘federated’ approaches to model selection. And then there are the open-source model options.
How might enterprises approach this array of choices? What should they consider?
DT: Most organizations have to consider lots of variables: what’s your budget, preferred features, importance of safety, maintenance, speed and performance? The point is to evaluate your options with respect to your business objectives and ability to execute. In many cases, that might include building a model from scratch with your own proprietary of data.
What do you think is most important for enterprises to consider when it comes to Gen AI or AI more broadly?
DT: Enterprises must take proactive steps to mitigate the known risks associated with Gen AI proliferation. Concerns span from internal risks associated with LLM hallucinations and data governance to deliberate malicious attempts to inject prompts and access corporate data from external actors. We see a significant need for enterprises to address challenges such as ensuring robust data privacy and navigating the complexities of integrating new technologies into existing systems. Fortunately, there are both third-party solution providers and DIY approaches to help enterprises make informed decisions and take appropriate actions.
5) Crystal ball time: Any thoughts for Gen AI / AI developments in 2025?
DT: As AI assistants integrate with more data sources and thus begin to better understand customers’ needs while providing expanding utility across enterprise applications, we’ll see more bot-to-bot collaboration. AI agents will begin to establish relationships with brands and their sales or customer care systems. This allows brands to manage their data and governance their way, while having the freedom to choose and update their AI models. A unified AI architecture that combines intelligent agents, adaptable models, a composable data layer, and knowledge processing will enable CX organizations to deploy AI solutions quickly and securely.
