The Case for Frontier Models in CX

A recent trend among some of our peers is the introduction of branded “proprietary” AI models. The positioning varies but it goes something like this: take an open-source model, add a sprinkle of reinforcement learning, brand it as a “proprietary” AI with strategic differentiation.  But the underlying strategy shows through; what’s happening here is a transition to cost optimization.

We understand the draw. If your goal is simply to deliver automation that is better than legacy AI technology, then yes we agree: Open source models have improved enough to outperform legacy AI technology.  We also believe that with the right setup, access and supervision, they can automate most of the work performed by your typical customer service representative.

But here is where our strategy at Crescendo differs: our ambition and our optimism for what AI can do next.  We didn’t set out to simply replicate legacy automation technology or replicate customer service representatives.  We set out on a path to maximize the impact of AI technology.  In our first two years of operation, we’ve been able to increase access to amazing CX for our customers by expanding coverage hours, expanding channels, expanding languages, and delivering AI powered expertise.  ….Oh, and we also lower costs.  That’s a footnote, not our primary goal.  

Frankly, we think that retrenching to “automate CX work” with low cost models shows a failure of imagination and a pessimistic view of where AI technology will be taking the global economy.

A difference in Product Architecture and Strategy

We’ve seen the deployments of some of our partners that use LLM technology to replicate essentially what was possible with low-code workflow tools.  The workflow is the outcome they seek, the only difference is that the steps are now written as prompts and executed by an LLM.  This may fit historical expectations for what CX automation technology should do but it misses the pivot that’s possible with LLM AI technology: exponential improvement.

At Crescendo, our strategy has been shaped by an assumption that the intelligence from AI will continue to improve rapidly and it will become more powerful and more capable than the agents or even the CX Consultants of the past.  While it may be uncomfortable to some, the recognition that computers and software can ‘outperform’ people at certain tasks is not new; this has been the case since the introduction of the calculator.  The difference with LLM AI is that we now have a technology that can perform language and reasoning tasks better than people.  This is where it takes imagination to shape the future of CX work, and this is how we are trying to contribute to the conversation of the evolving CX workplace.

As far as the CX automation use case; it became clear to us in 2025 with GPT-4o, we felt that we had all the intelligence we needed to perform most conversational CX work.  From that point on, the primary barriers to CX automation was no longer AI; it was access to data, integration into customer systems, and process definition.  This type of work drove much of our focus over the past 6 months.

Our engineering work progressed into tackling advanced CX conversations involving voice, image processing, and ultimately full multimodal interaction.  That was our shift from imitating human CX work and toward exceeding the capabilities of a person.  Our multimodal assistant can conduct a voice conversation, exchange messages, and process shared images simultaneously in one conversation.

From a research and development perspective, our focus has moved on to much more ambitious use of AI.  We extended the product into analysis and optimization work. What began with Data Assistant evolved into AI Insights.  Here, the role of AI is no longer just to execute customer experience, it is to advise on how customer experience should be improved.  In operator terms, the system is now recommending improvements to SOPs (soon, just writing them outright).  This work benefits from the power of frontier model AI and for this reason, we are always hungry for the next model release.  For example, we adopted and released our AI Insights feature with OpenAI 5.4 days after it became available.

Initially, our operational experts (our people) translated these insights into process changes. Increasingly, AI is beginning to take on more of that translation work.  The rapid emergence of Agentic Coding has resulted in a fast transition from AI making recommendations to AI doing the execution of work. 

At Crescendo, we now have AI that can do the substantial parts of the work done by a Forward Deployment Engineer. We have AI that analyzes AI performance and suggests improvements.  We have AI that writes code to produce business-specific performance metrics for all of our customers to offer insights.  We are moving toward AI that can write the code required to implement those suggestions in a loop of recursive self improvement.

Once those pieces are connected, you are no longer looking at a static automation platform. You are looking at an agentic self-improving system.  

Customer experience is entering an era where the winning platforms will not merely automate existing workflows. They will observe operations, identify friction, recommend improvements, implement approved changes, and measure the results.

That is the customer value proposition of frontier models in CX: not cheaper automation, but a faster rate of operational improvement. Crescendo customers do not just get an AI assistant that handles conversations. They get a system that learns from every conversation, turns those learnings into better policies, better integrations, better customer journeys, and better business outcomes.

Frontier models matter because they keep that horizon open. They allow the platform to improve as the underlying intelligence improves. They do not just automate today’s workflow. They expand what CX can become.