AI Report Analysis

AI Features · 6 min read

The AI report analysis panel is available across all reports in Evrykit — inventory, sales, purchasing, accounting, finance, point of sale, and any future module. This article explains how to use it effectively.

Opening the AI Panel

When viewing any report, look for the AI Analyze button in the report toolbar (it has a purple gradient and a lightbulb icon). Clicking it slides up the AI panel from the bottom of the report preview area.

The panel overlays the report rather than replacing it, so the data stays visible for reference as you work.

The First Question

When the panel opens it is empty and ready for your first question. The AI has already been given context about what type of report is displayed. You do not need to describe the report — just ask what you want to know.

Effective first questions by report type:

Inventory — Stock on Hand

"Which products are below their reorder level and which warehouse is most affected?"

Sales — Sales by Customer

"Who are my top 10 customers by revenue this quarter and how do they compare to the same period last year?"

Purchasing — Purchase Orders

"Are there any overdue purchase orders and what is the total value at risk?"

Accounting — Account Ledger

"Summarise the movement on this account and highlight any entries that look unusual."

Accounting — Profit & Loss

"What are the three biggest drivers of the change in gross margin compared to last month?"

Point of Sale — Session Report

"Which payment methods generated the most revenue and what was the average transaction value?"

The AI streams its response in real time — you see text appearing as it is generated rather than waiting for the full answer.

Conversational Follow-Ups

After the initial response, keep the conversation going. The full message history is sent with each request so the AI maintains context:

You: Which products are most at risk of stockout?
AI:  The following 5 products have stock below their reorder level...

You: For the top 3, what was the average daily consumption over the last 30 days?
AI:  Based on the stock movement data visible in the report...

You: Draft a purchase requisition summary I can send to my procurement team.
AI:  Here is a brief summary suitable for a purchase request email...

There is no limit on follow-up turns within a session.

Switching Models Mid-Session

You can change the AI model at any point using the Model selector in the report sidebar. The next message you send will use the newly selected model. Previous responses in the thread are unchanged.

This is useful when you want a quick initial scan (Gemini Flash) followed by a deeper narrative summary (Claude Sonnet).

Downloading Your Analysis

Once you have a useful conversation, you can save it in two ways:

Download as Markdown

Saves the full conversation thread as a .md file. Each message is clearly labelled (You / AI Analyst) with a date stamp. Use this to:

Download as PDF

Generates a formatted PDF with the Evrykit header, conversation bubbles styled by role, and your company branding context. Use this to:

Both download options appear in the sidebar once the first AI response has been received.

Tips for Better Results

Be specific about time periods. If the report covers a date range, mention it explicitly: "In Q1 2025, which..." — the AI can only see what is currently rendered in the report.

Ask for structured output. Requests like "Give me a table of..." or "List the top 5 with bullet points" will return markdown-formatted output that renders clearly in the panel and downloads well as PDF.

Use follow-ups for action items. A good pattern is: (1) ask for analysis, (2) ask for root causes, (3) ask for recommended actions. Breaking it into three turns gives better depth than a single complex question.

Switch models for different tasks. Flash models are fast for scanning — use them first. Switch to Sonnet or GPT-4o for the final narrative or when you need more nuanced language in a downloadable report.

Rate Limits

If you see a "Rate limit reached" error, your API key has hit the provider's request quota for the minute. Wait 60 seconds and try again, or switch to a different model/provider. To avoid this:

← Previous
Configuring AI API Keys
Next →
Choosing the Right Model