💡 Avatar-Guided Business Data Onboarding: Conversational Data Readiness for Non-Technical Users

[github-actions[bot]]

Summary

When starting a new workflow that needs business data, the avatar conversationally guides users through connecting their actual data sources — explaining what data will improve results, asking permission, and verifying data quality in real time. Instead of a technical "Add MCP Server" settings screen, the avatar says "To help you with budget analysis, I'll need to see your spreadsheets — can you show me the folder where you keep them?" The avatar inspects data quality during onboarding and flags issues conversationally ("I see 47 files in this folder — 12 are PDFs, 8 are spreadsheets, and 3 seem to be duplicates. Want me to work with all of them?").

Market Signal

Goldman Sachs 2026 survey: 45% of small business AI users cite lack of technical expertise as their primary challenge. McKinsey: 46% of adoption friction is skills gaps. 28% of small businesses cite data readiness specifically as a barrier. EU: 70.9% of enterprises cite lack of relevant expertise as the primary reason for not adopting AI (Eurostat 2026). Meanwhile, enterprise data onboarding wizards (Salesforce, Power Platform) remain form-filling experiences designed for technical teams. No competitor offers avatar-guided conversational data onboarding — this is an open gap in the market where TalkTerm's embodied interface is uniquely suited.

User Signal

Existing idea #66 (Conversational MCP Onboarding) focuses on a tool marketplace — discovering and connecting MCP servers as third-party tools. Data onboarding is fundamentally different: it's about helping users make their OWN existing data (spreadsheets, documents, reports) accessible to the agent, not about installing new tools. The PRD's User Journey 3 (Sarah's PDF failure) directly illustrates this gap — Sarah doesn't understand why a scanned PDF can't be read, and the avatar must explain and offer alternatives. Data readiness is a prerequisite for every meaningful workflow — solving it conversationally unlocks the rest of TalkTerm's value.

Technical Opportunity

MCP's new stateless architecture (2026-07-28 spec) makes lightweight data connectors deployable as serverless functions, reducing the infrastructure overhead for supporting multiple data source types. TalkTerm's workspace selection (FR52-53) already handles file system access scoping. Extending this to conversational data discovery leverages the existing architecture — the agent's file read/create/modify capabilities (FR13) and MCP tool invocation (FR14) provide the infrastructure. The DataOnboardingFlow domain service interface can use the existing confirmation gate pattern (FR20) for data access permissions, and preference memory (FR51) to remember data source configurations across sessions.

Assessment

Dimension	Score	Rationale
Feasibility	med	Core local file discovery is straightforward using existing file system tools; expanding to cloud sources (Google Drive, SharePoint) requires MCP connectors. Conversational flow is primarily prompt engineering
Impact	high	Directly addresses the two largest AI adoption barriers simultaneously (45% technical expertise gap + 28% data readiness). Unlocks workflow value that is currently blocked by data inaccessibility
Urgency	high	Every TalkTerm user will hit data readiness friction on their first real workflow. Solving this conversationally is critical for the "first workflow completed in 25 minutes" success metric (PRD)

Adversarial Review

Strongest objection: Data quality issues — users connect messy, incomplete, or sensitive data, get poor results, and blame TalkTerm. Also, each data source type needs its own conversational flow, making the design surface enormous.

Rebuttal: Start with the simplest case: local file system access (folders of documents, spreadsheets, PDFs). The avatar inspects and reports quality conversationally — data quality conversation IS the feature, not a cost. It builds trust by showing the user exactly what the agent sees and letting them make informed decisions. Expand to MCP-connected sources (Google Drive, SharePoint) incrementally. The conversational interface means adding new data source types is primarily a prompt change, not a code change — the underlying MCP infrastructure handles the actual data access.

Suggested Next Step

Design the DataOnboardingFlow domain service interface in src/shared/types/ports/. Map the top 5 data source types by non-technical user frequency (local files, Google Drive, SharePoint, email exports, CRM exports). Prototype the conversational flow for local file discovery as the MVP — the avatar asks about project context, suggests relevant file types, offers to scan the workspace folder, and reports what it found with quality observations.