The public LLM market is expanding quickly, but enterprise needs are evolving just as fast
A large language model (LLM) is an AI system trained on massive volumes of text to recognize patterns in language and generate human-like answers, summaries and recommendations.
The market for public LLMs (like ChatGPT and Microsoft Copilot) and enterprise AI is no longer defined by novelty alone. Gartner forecasts that worldwide AI spending will reach $2.59 trillion in 2026, up 47% year on year, and will grow further to $3.49 trillion in 2027, with particularly strong expansion in AI software, AI infrastructure and AI models. Gartner also says spending on AI models alone will rise from $32.6 billion in 2026 to $59.2 billion in 2027, reflecting how quickly model adoption is accelerating across enterprise workflows.
Public LLMs have become highly capable and widely accessible, but enterprise buyers are increasingly evaluating them not just on fluency, but on their ability to support governance, domain expertise, repeatability and measurable business outcomes.
That shift is happening because the market is both crowded and maturing unevenly. IDC’s 2026 outlook points to a transition from AI buildout to AI adoption, with enterprises moving from experimentation toward scaled deployment. The strategic direction of the market is also becoming clearer. Gartner predicts that by 2027, more than 50% of the GenAI models enterprises use will be specific to an industry or business function, up from roughly 1% in 2023. That is an important signal: organizations are moving beyond generic, public-purpose AI toward more specialized, domain-aware systems that reduce hallucination risk and better support real-world decisions.
This is the context enterprise CX leaders now face: a market full of rapidly improving public LLMs, rising investment and expanding vendor choice - but also one where trust, domain fit, governance, scale and operational actionability are becoming the real points of differentiation. In Customer Experience Management, that distinction matters, because value comes from turning insight into managed action across the customer journey.
Public LLMs answer prompts. Journey Management requires a context and differentiation
Public LLMs are useful for summarization, brainstorming and ad hoc analysis. But customer journey management is not a single interaction. It is an ongoing discipline that depends on data structure, shared context, prioritization, collaboration and follow-through over time.
Cemantica’s Journey Insights Hub sits inside the Journey Value Management Platform (JVMP), where customer and operational data can be unified, analyzed in journey context, and connected to decision-making, execution and continuous optimization. The platform integrates journey design, insights, orchestration, engagement and value measurement in one environment rather than treating insight and CX improvement as a one-off.
One of the core limitations of public LLMs is that they are, by definition, public. If every company is using similar general-purpose models to support CX strategies, then competitive advantage is lost. If a team is using AI, the success comes from its use when grounded in proprietary customer data, business taxonomies, journey logic and decision workflows.
That context is especially important in Customer Experience, where insight is often fragmented across journeys, surveys, calls, chats, reviews, tickets, CRM systems and operational applications. Journey Insights Hub integrates and structures those sources into a journey-centric intelligence layer, making insight more relevant to actual experience improvement work.
Domain-specific AI matters in Journey Management
General-purpose LLMs are not built with native knowledge of journey management methodology. They may produce plausible responses, but they do not inherently understand journey stages, persona frameworks, friction points, or best-practice journey design.
Journey Insights Hub reinforces a different model, with context-aware language AI, nuanced sentiment and concepts, prioritized insights and journey-aware action, all aligned with helping experience teams understand what customers mean and what the business should do next.
Because Journey Insights Hub is developed and maintained by Cemantica, it evolves in line with best practice and customer requirements over time, rather than following the generic roadmap of a public LLM provider.
Cemantica combines both quantitative AI that builds a map of the conversation with generative AI that provides more narrative interpretation, both approaches together can reveal different dimensions of meaning, for stronger insight and adoption by users.
How LLMs are transforming Customer Experience
LLMs are changing how organizations interact with customers, making it possible to deliver faster, more personalized and more accessible experiences across multiple channels. While public LLMs can support general-purpose interactions, their business value increases when they are connected to relevant customer information, preferences and operational context.
There are several important applications for Customer Experience teams:
1. Intelligent customer support
LLMs can power conversational assistants that respond to frequently asked questions, explain products and services, help resolve complaints and guide customers through processes such as returns or refunds. Unlike traditional rule-based chatbots, which rely on predefined responses, LLM-powered assistants can interpret natural language and generate contextual answers.
For example, rather than returning a standard return-policy statement, an AI assistant could explain the policy in relation to a customer's specific purchase and provide the next steps. When connected to approved knowledge sources and business systems, these assistants can also help customers complete tasks rather than simply provide information.
2. Personalized customer experiences
By interpreting customer behavior, preferences, feedback and previous interactions, LLMs can help organizations tailor communications and recommendations to individual needs.
Applications include personalized product recommendations, relevant offers, targeted campaign messaging and more contextual customer communications. When combined with customer analytics, this information can help organizations understand what different customer groups value and adapt experiences accordingly.
3. Multilingual customer assistance
For organizations serving international markets, language can be a significant barrier to consistent customer experience. LLMs can translate, summarize and generate content in multiple languages, helping businesses provide more accessible support across regions.
They can also help localize marketing communications and customer-facing content to reflect regional preferences and cultural context, rather than simply translating words. Human review remains important to ensure accuracy, appropriate terminology and cultural relevance.
4. Customer sentiment and feedback analysis
LLMs can help interpret unstructured customer feedback from surveys, support conversations, online reviews and social media. They can summarize recurring themes, identify emerging concerns and explain the context behind customer opinions.
However, generating summaries is only one part of understanding Customer Experience. Organizations also need consistent analysis across large volumes of feedback to identify sentiment patterns, changing customer expectations and the underlying causes of friction.
This is where Cemantica Journey Insights Hub adds another dimension. Its AI-powered analysis helps organizations understand customer language, sentiment and concepts across multiple feedback sources, connecting those insights to customer journeys and opportunities for improvement.
From individual interactions to journey-wide improvement
These applications demonstrate the potential of LLMs to improve individual customer interactions. However, delivering consistent Customer Experience requires organizations to look beyond individual conversations.
A customer support interaction, a product review and a survey response may all reveal different aspects of the same journey problem. By bringing these sources together and analyzing them in journey context, organizations can move from responding to individual issues towards identifying patterns, prioritizing improvements and measuring their impact.
Cemantica's Journey Insights Hub supports this broader approach by connecting customer feedback and operational data with the journey management process.
Understanding the different types of Large Language Models
Not all large language models work in the same way. They can be classified according to their underlying architecture, how they are trained and the tasks they are designed to perform. Understanding these distinctions helps organizations evaluate which AI approaches are appropriate for their business requirements.
There are three useful ways to understand LLM types.
1. Types based on architecture
The architecture determines how an AI model processes and interprets information.
- Transformer-based models: The dominant architecture behind modern LLMs. Transformers use self-attention mechanisms to understand relationships between words and other elements of a sequence. This allows them to process context effectively and perform tasks such as text generation, summarization, translation and sentiment analysis
- Recurrent neural network (RNN)-based models: These models process sequential information using an internal state that retains information from previous steps. Earlier language-processing systems used RNN architectures, including LSTM and GRU variants. They are less common as the foundation for modern large language models
- Hybrid models: These combine components from different AI architectures to address specific processing requirements. In enterprise applications, hybrid approaches can bring together language models, structured analytics and other techniques to balance contextual understanding, performance and reliability
2. Types based on training approach
The training approach determines how a model learns from data and how it can be adapted to specific business requirements.
- Self-supervised learning models: These learn patterns and relationships from large volumes of unlabeled data. For example, a language model can learn to predict missing words or tokens within a sentence, developing an understanding of language without requiring manually labelled examples for every training task
- Supervised learning models: These learn from labelled examples, where input data is paired with an expected output. This approach helps models learn to perform specific tasks, such as classifying customer feedback into predefined categories or identifying particular information within text
- Fine-tuned models: These are pre-trained models that receive additional training to improve their performance for a particular task, industry or application. For example, a general-purpose language model can be adapted to better understand customer service terminology or an organization's specific product information
These approaches can work together. A model may initially learn through self-supervised training before being fine-tuned using labelled data for a more specialized application.
3. Types based on capabilities
Models can also be distinguished by the type of task they perform.
- Generative models: These generate new content based on patterns learned during training. Applications include writing responses, summarizing customer feedback, translating communications and creating conversational interactions
- Discriminative models: These focus on identifying patterns, distinguishing between categories and making predictions based on input data. In customer experience, this can include classifying feedback, identifying sentiment and categorizing customer comments
For CX teams, both capabilities can be valuable. Generative AI helps explain and communicate insights, while analytical and classification models help structure large volumes of customer information into consistent, measurable outputs.
Why these distinctions matter for enterprise Customer Experience
The type of model is only one consideration when selecting an AI solution. Organizations also need to consider the data being analyzed, the context available to the model, the consistency of its outputs and how insights can be used in business processes.
For example, a generative LLM may be excellent at summarizing a collection of customer reviews. However, an enterprise customer analytics solution also needs to identify recurring concepts, classify feedback consistently, measure sentiment trends and connect findings to specific customer journeys.
Cemantica Journey Insights Hub combines structured natural language processing with generative AI capabilities. This hybrid approach supports consistent analysis of customer feedback while using generative capabilities to provide additional interpretation and make insights easier for teams to understand and act upon.
Scale, governance and repeatability
Public LLMs are often strongest in exploratory interactions with limited prompts or narrower datasets. Customer journey analytics work is different: it requires the ability to process large, continuous streams of customer and operational data and turn them into repeatable insight.
The Journey Insights Hub answers this need with capabilities of recurring classification, trend measurement, concept-level understanding and a hybrid analytic backbone designed for production use rather than purely conversational output.
Journey Insights Hub is designed for high-volume, ongoing analysis across surveys, tickets, calls, chats, reviews and operational data, where repeatability, consistent tagging, sentiment scoring and trend measurement matter.
Journey Insights Hub suits enterprise deployment with stronger controls around privacy, access and data exposure than shared public LLM environments, helping organizations reduce the risks that can come with sending sensitive journey data through public AI services.
AI systems need to be reliable, explainable, privacy-enhanced and secure. In a business environment; production-grade platforms like Cemantica, with defined controls, traceability and enterprise user management become more valuable than standalone prompt interfaces.
Reliability matters when insight drives business action
Public LLMs can generate impressive responses, but they can also hallucinate or produce information that is difficult to validate. That may be acceptable in low-risk exploration, but it becomes a larger issue when organizations are using AI to prioritize investment, redesign journeys or assign action across teams.
Journey Insights Hub is designed to minimize hallucinations through a more controlled, explainable, and repeatable analytic foundation, making its outputs more reliable for enterprise journey decisions than general-purpose public LLM responses. It uses structured NLP for the analytic backbone and LLM capabilities more selectively for summarization and deeper interpretation.
That approach is particularly relevant in enterprise AI for Customer Experience, where teams need outputs they can trust, trace and act on consistently over time.
At-a-glance: Public LLMs vs enterprise AI
The comparison below summarizes why journey management requires a different AI approach (Journey Insights Hub) than a general-purpose public LLM.
Criteria | Public LLMs | Cemantica Journey Insights Hub |
Primary purpose | Designed to generate answers, summaries, and conversational outputs in response to prompts. | Designed to support journey management, turning journey and customer data into structured insight and action within the Journey Value Management Platform. |
Management capability | Delivers a punctual answer but does not inherently provide management tools to track action and follow-through. | Connects insight to a broader platform for journey analysis, optimization, orchestration and value measurement. |
Competitive advantage | Uses a public model that is broadly available, so differentiation depends on what the user adds around it. | Builds advantage from customer-specific data, journey context and platform integration; rather than relying on the same shared public model everyone else can access. |
Adaptability to customer requirements | Evolves according to the roadmap of the public model provider. | It can evolve in line with customer journey-management requirements and platform priorities over time. |
Domain expertise | General-purpose by design, without native specialization in journey management methodology. | Built for journey work, with journey-aware structure, taxonomy and best-practice context. |
Data scale | Best suited to ad hoc exploration or narrower prompt-based datasets. | Designed for high-volume, ongoing analysis across surveys, tickets, calls, chats, reviews and operational data. Cemantica is stronger for scalable, recurring analytics. |
Repeatability and consistency | Outputs can vary from prompt to prompt depending on phrasing, context and model behavior. | Cemantica brings repeatability, transparency and explainability; with structured NLP providing the stable analytic backbone. |
Governance and enterprise control | Governance depends heavily on how the public tool is configured and used. | Built as a production-grade enterprise capability with stronger governance, control and platform-level management. |
Privacy and exposure risk | Public LLM use can create higher exposure risk when sensitive data is sent through broadly accessible third-party services. | Designed for enterprise deployment with stronger privacy, access and exposure controls. Reduced risk of external leakage and no exposure to external internet contamination. |
Hallucination risk | Can generate inaccurate or invented content, which creates challenges for enterprise decision-making. | Designed to minimize hallucinations through a more controlled, explainable and repeatable analytic foundation. With a hybrid architecture combining NLP and LLMs selectively. |
Best fit | General-purpose assistance, ideation, summarization and exploratory Q&A. | Enterprise journey intelligence, prioritization and action in a governed production environment. |