Embedded AI: The Questions Every CEO is Asking

Market Context & Challenges

• 2022 is characterized as a breakout year for AI, with ChatGPT identified as the fastest-growing app in history.
• CEO Zed Inam (Cresta) notes that contact centers suffer from 30–40% annual agent attrition, with some sectors reaching 80% due to low wages, high stress, and seasonal demand.
• Employee Net Promoter Scores (eNPS) in contact centers frequently fall below zero.
• Companies face critical decisions regarding data privacy, cost, accuracy, and the speed of AI integration.

Strategic AI Implementation Models

• Lazy AI: Defined as end-to-end automation of existing processes without reimagining the business model.
• Creative AI: Defined as using AI as a foundational capability to reimagine products, processes, and human-AI collaboration to achieve previously impossible outcomes.
• Cresta emphasizes the "bi-directional" value of AI: serving the customer while simultaneously feeding product and market insights back to the business.
• Unstructured customer conversation data is utilized to drive strategic decisions, such as pricing adjustments and new product launches (e.g., telecom pricing bundles).

Differentiation Through Context & Data

• Sourcegraph (Byung Liu): Positions "context as king," arguing that the quality of LLM output depends on the specific data fetched rather than the model itself.
• Kodi (Sourcegraph): Differentiates from GitHub Copilot by fetching broad, relevant code context (including organization-specific private code) rather than relying solely on local recent files or autocomplete.
• Sourcegraph differentiates via open-source transparency (Apache 2 license for Kodi) to facilitate a pluggable ecosystem and avoid vendor lock-in.
• Hex (Barry McCardle): Leverages thousands of historical SQL/Python queries and database schemas to construct personalized prompts and reduce model hallucination.
• Hex maintains a "human in the loop" policy, generating code for review rather than executing it autonomously to ensure safety and learning.

Security, Privacy, & Business Models

• Transparency: Cresta advocates for disclosing AI interactions to protect brand trust, noting that non-disclosure risks long-term reputation damage (e.g., AI accent masking).
• Data Moats: Companies are recognizing the value of retaining proprietary datasets longer to train models or provide context, though this introduces retention costs and security risks.
• Vendor Agnosticism: Sourcegraph enables customers to "bring their own model," supporting options from OpenAI, Anthropic, and Salesforce's Codex to mitigate security and pricing risks.
• Self-hosting is prioritized for security-sensitive enterprise customers to prevent code leakage.
• Model pricing volatility (e.g., OpenAI price hikes) necessitates strategies to integrate multiple models to manage cost exposure.

Future Outlook (2028 Projections)

• Contact Center Evolution: Industry experts predict a shift to a "touchless" workflow by 2028 where AI handles data entry, summarization, and system navigation.
• Role Transformation: Contact center agents will transition from reactive problem solvers to strategic relationship builders and account managers.
• Efficiency Metrics: Key performance indicators will shift away from "Average Handle Time" toward metrics focused on relationship quality and proactive engagement.
• Frictionless Digital: Routine inquiries will be handled by advanced bots (potentially with entertainment features), reserving human agents for high-value, strategic interactions.