How Answers Are Generated

Every answer you receive from AskBiz goes through this pipeline in under 5 seconds:

Step 1 — Intent classification: Your question is parsed to understand what you're asking. Is this a data query? A general business question? A tool request? This determines the processing path.

Step 2 — Data retrieval: For data-grounded questions, AskBiz queries your connected sources to retrieve relevant metrics. We don't send your entire dataset to the AI — only the specific data needed to answer the question. For example, 'What's my refund rate?' retrieves only refund records, not your full order history.

Step 3 — Prompt construction: We build a structured prompt containing: the retrieved data, your question, your business context (sector, currency, plan type), and our system instructions for how Claude should behave.

Step 4 — AI reasoning: Claude processes the prompt and generates a response. This is where the actual reasoning happens — synthesising data, applying business logic, identifying patterns, and formulating recommendations.

Step 5 — Confidence scoring: AskBiz evaluates the response against our confidence criteria: How complete was the data? How recent? How directly relevant to the question? This produces the confidence indicator (High/Medium/Low/Estimate).

Step 6 — Output formatting: The raw text response is parsed into the appropriate format — plain text, a data table, a chart specification, or an action card — depending on the question type.

Step 7 — Delivery: The formatted response is delivered to your interface with the confidence indicator, data freshness timestamp, and any relevant action buttons.

Not every question goes through the full AI pipeline. AskBiz's intent detection layer recognises certain question patterns and routes them directly to structured Business Tools:

• Questions about FX or currency risk → FX Risk Modeller
• Questions about landed cost, duty, or import costs → Landed Cost Calculator
• Questions about export markets or expansion → Export Market Scoring
• Questions about churn or at-risk customers → Churn Intelligence
• Questions about TikTok, Instagram, or social commerce → Social Commerce Intelligence

This routing happens because structured tools produce more reliable, auditable outputs for these specific use cases than open-ended AI generation. The AI is still involved in interpreting results — but the core calculation happens in a deterministic tool, not probabilistically.

Answer quality depends primarily on three factors:

Data completeness: The more data sources you connect, the better. Connecting Shopify + QuickBooks together enables cross-source answers (e.g. true profit per order) that neither source alone can provide.

Data recency: Stale data produces stale answers. Check the data freshness indicator on any answer. If your Shopify sync is 3 days old, revenue trend answers reflect 3-day-old reality.

Question specificity: Vague questions produce vague answers. 'How are sales?' is harder to answer well than 'What was my Shopify revenue in March vs February, broken down by product category?' The more specific the question, the more targeted the data retrieval and the more precise the answer.