In AI, 12 months is a lifetime. While laptops and smartphones may last a few years, today’s AI models go stale much sooner. That’s a big problem for brands building contact centers around them.
Unfortunately, many are doing so, slapping generative AI (GenAI) on top of their contact center stack, using basic API integrations and point solutions. But without more intelligent design, these plans will fail fast.
A better approach is building a smarter foundation. That includes a model architecture that’s channel-agnostic, a platform that automatically adapts to the nuances of each communication channel, and a system built for easy plug-and-play with emerging AI components.
With such as system, a contact center’s AI system will grow with them, not age out.
Carlos Aragon, Senior Director at Sprinklr, shared the following five insights to help contact centers build toward such as system and futureproof their AI investments.
- Build a Flexible, Modular Architecture
Aragon advocates for an approach where a contact center can design a single AI model, which advances over time, and deploy it across all their contact center channels. That eliminates the need to establish, monitor, and optimize separate models for each.
“The platform should also support language input from any source (voice or text) and include an intelligent integration layer that automatically adapts to channel-specific requirements,” he added.
Consider social customer service, for instance. Here, the model should limit responses to 300 characters on Instagram or extend to 3,000 words on Facebook, without manual intervention.
Equally crucial is support for plug-and-play AI components, such as virtual agents, recommendation engines, or other models, enabling seamless integration and flexibility.
- Continuously Monitor & Retrain AI Models
Continuous monitoring and retraining of AI allows for feedback loops that improve how the system responds to customers.
Second, it ensures the AI is functioning properly and not producing inaccurate information that could lead to business risks, whether that’s making false promises or even violating regulations, which could lead to lawsuits.
Over the past 18 months, there have been many examples of this. From virtual agents misadvising customers, arguing within them, and – per one memorable example – trashing the company.
Some of these cases have resulted in legal challenges. That’s why regularly retraining models to prevent data drift is essential. Otherwise, model performance can degrade over time.
“You should also use tools that allow agents to give feedback when an intent fails,” added Aragon.
“Even simple feedback like a thumbs up/down helps, but ideally, agents should be able to provide specific comments like: “This failed because the system didn’t include X.” That way, the system can learn and improve automatically.”
Moreover, many contact centers may move to update knowledge bases, case guides, and process scripts dynamically, so that the AI provides the right answer next time around.
- Prioritize Data Privacy & Ethical AI
Ethical AI and data privacy is not only essential from a legal perspective, but also from an AI hygiene standpoint. One major consideration here for brands that use third-party AI models is to ensure the proper scrubbing of personally identifiable information (PII).
Noting this, Aragon said: “At Sprinklr, for example, we use internal AI models that detect and scrub PII, like names, addresses, and bank information, before sending prompts to OpenAI or other LLMs.”
In doing so, Sprinklr replaces sensitive data with placeholders, sends the anonymized request, and when the response comes back, verifies it for hallucinations. If it passes a quality threshold, it re-injects the original data and deliver the final response to the customer.
“This approach is only possible if AI is built into the platform from the ground up,” continued Aragon. “For platforms treating AI as a plugin or afterthought, it’s extremely difficult to retrofit that level of control.”