Ontology AI

Ontology AI is the natural-language evidence analysis panel that opens on the right side of the EXAWin+ screen. Users can explore deals, activities, customer relationships, Bayesian judgment, ontology graphs, official documents, and blog/theory documents in natural language.

This feature is not a simple chatbot. It is designed so actual work data accumulated in EXAWin+, ontology relationships, public document assets, and theory documents participate together in evidence-centered analysis.

Location: Top header → AI-Agent icon → Ontology AI panel

What It Can Do

Ontology AI interprets user requests within the current permission scope. A request may be about a specific deal, the purpose of the current screen, the company's project portfolio, official documents, or blog/theory documents.

To analyze a specific deal, include both project code and deal name in the request, or select the project in the Ontology AI panel. For example: "Explain why P(Win) became lower for project code [code], deal name [deal name], with activity and signal evidence." You can also ask for product explanations such as "Summarize EXAWin's Bayesian judgment structure for a customer explanation." If you are looking at a screen, you can ask about that screen's purpose and review order.

When possible, the answer separates work data, document evidence, and graph evidence. The user can therefore receive explanations grounded in the actual system and official documents, not only general statements.

Panel Structure

Click the AI-Agent icon in the top header to open the panel. The panel includes an input area, request type guide, session area, scope selection, project selection, and answer area.

The session area separates analysis flows. Users can start a new flow or revisit previous flows.

The request type guide helps users understand what kinds of analysis are available.

The answer area can include natural-language explanation, evidence lists, and graph evidence. When graph evidence is available, users can inspect which objects and relationships support the answer.

Scope Selection

Ontology AI separates the answer scope. Choosing the right scope improves answer quality.

All knowledge uses work data, official documents, and blog/theory documents together. Use it when product understanding, deal judgment, and theory explanation are mixed.

Current screen sends screen-identification context such as current path and title. It does not mean every visible datum or the currently viewed project is automatically analyzed. Use it when asking what to check on the screen or how to use the screen.

Current project focuses on the project selected in the Ontology AI panel or a project identifiable from the URL and connected data. If the project is not clear, include project code and deal name in the request.

All projects reads the company portfolio in aggregate. Use it for delayed deal types or projects over threshold.

Official documents limits the answer to official homepage documents and user manuals. Use it for customer explanation, internal training, or feature usage.

Blog / theory uses blog and theory documents. Use it to understand Bayesian engine, silence, momentum, or Auto Tuner concepts more deeply.

General request is used for general knowledge separated from product work data.

Request Types

The panel provides request types as a guide.

Bayesian calculation covers P(Win), threshold, k, Bayesian update, stage, and signal calculation structure.

Time decay and momentum covers time decay, silence penalty, and momentum P(Win).

Document-grounded theory and manuals uses official documents, manuals, blog posts, and theory documents.

Portfolio and deal analysis covers projects, customers, stages, probability, and risk comparison.

Customer and activity relationships covers customer, contact, activity, attendee, and buying-center relationships.

Evidence graph shows how answers connect to documents, business ontology, and graph evidence.

Auto Tuner and analysis evidence explains what Auto Tuner learns and which analysis results it uses.

Full ontology coverage covers company, user, customer, deal, activity, signal, and snapshot relationships.

Good Request Examples

For a specific deal, include project code and deal name.

"For project code [code], deal name [deal name], explain why current P(Win) decreased using activity and selected signal evidence."

"For project code [code], deal name [deal name], suggest questions that must be confirmed before the next meeting."

"For project code [code], deal name [deal name], combine recent activity, Bayesian state, and official documents into a customer-facing evidence summary."

For portfolio-level analysis:

"What types of deals currently have high delay risk across all projects?"

"Separate deals that passed threshold from deals that remain blocked."

For product and theory understanding:

"Summarize EXAWin's Bayesian judgment structure for customer explanation."

"Explain why silence penalty is needed based on the theory documents."

"Explain what Auto Tuner learns from completed won/lost history."

How To Read Answers

Do not read only the final sentence. Check which evidence supports the answer.

For work-data answers, the evidence may include the deal, customer, activity, signal, or snapshot. For document-grounded answers, the evidence may include manuals, official documents, blog posts, or theory documents. If graph evidence is provided, inspect which objects and relationships are connected.

If the answer is not enough, narrow the follow-up request. Examples: "Summarize only the activities behind that judgment," "Separate official document evidence from actual deal evidence," or "Convert this into questions for the next customer meeting."

Security And Permission Boundaries

Ontology AI answers only within the user's permission scope. It does not answer about other companies, other tenants, or unauthorized projects.

Ontology AI also does not directly modify data. Requests such as "edit project code [code], deal name [deal name]," "delete this activity," or "change the signal value" are not execution targets. Data changes must be performed by the user in the appropriate operating screen.

This boundary is a safety feature. Sales data and Bayesian judgment affect company operations, so the natural-language panel is not allowed to arbitrarily change data.

Using It With Decision Console

When Decision Console needs more explanation, open Ontology AI and analyze the evidence. The project must be selected in the panel or the project code and deal name must be included in the request.

If Decision Console shows a bottleneck, ask: "For project code [code], deal name [deal name], explain which activities and signals created this bottleneck."

Before adopting a recommendation, ask: "For project code [code], deal name [deal name], what customer questions should be confirmed before executing this recommendation?"

Recommendation adoption, follow-up activity recording, and customer contact results must still follow the Decision Console and Activity War Room workflow. Ontology AI explains judgment and organizes analysis; it does not write operating records on behalf of the user.

Good Habits

Make requests specific. Avoid vague requests such as "How is this deal?" Use a request like: "For project code [code], deal name [deal name], explain why P(Win) decreased based on recent activities and selected signals."

Choose the scope deliberately. Use current project for real deal evidence, official documents for product explanation, and blog/theory for conceptual understanding.

Before sending an answer to a customer, review it against company language standards and the actual sales context. Ontology AI helps explanation, but customer communication remains the user's responsibility.

Read Next

To understand ontology principles, read Ontology Operating Principles.

To review deal judgment and AI analysis together, read Decision Console.

To inspect object relationships in read-only mode, read Ontology Inspector.