By connecting AI (Claude or ChatGPT) to Questetra BPM Suite (hereafter “Questetra”) via an MCP connector, you can reference Questetra Case data from the AI chat and have it summarize, aggregate, and analyze the data. This article explains how to phrase your questions so the AI can effectively analyze Case data, with concrete prompt examples.
For the connection setup itself, see the companion articles “Connecting ChatGPT and Questetra via MCP” and “Connecting Claude and Questetra via MCP.” This article assumes you have already completed the connection using one of those methods (it is not specific to any particular AI product).
Introduction: AI Reads Cases by Calling “Tools”
When connected via MCP, the AI calls read-only tools published by Questetra to retrieve data. It does not write data or process tasks. For Case analysis, the following tools are primarily used.
| Tool | Role | Use in Analysis |
|---|---|---|
| list_process_models | List/search Apps | Identify the target App for analysis |
| list_process_data_definitions | Retrieve Data Item definitions for an App | Understand the meaning and type of each item |
| list_process_instances | List/search Cases | Filter Cases by conditions |
| get_process_instance | Retrieve details of an individual Case (Data Items + step history) | Deep-dive into a single Case; review progress and Token Current Position |
| get_document | Supplementary documents such as search condition references | AI verifies how to specify search conditions |
Users do not need to be aware of SQL or APIs. Simply ask in natural language, and the AI will call the necessary tools in sequence and compile the answer.
Basic Flow of Analysis
The AI generally reaches the target Cases in the following order. Keeping this in mind helps you give more precise instructions.
- Identify the target App (list_process_models) — Determine which App’s Cases you want to examine.
- Understand the Data Items if needed (list_process_data_definitions) — Check item names and types such as “Amount” or “Status” to use as criteria for filtering and aggregation.
- Filter Cases (list_process_instances) — Extract target Cases by conditions such as status, time period, Title, or Data Item values.
- Deep-dive into individual Cases (get_process_instance) — Retrieve detailed Data Items and history for specific Cases of interest.
If you specify the target App by name at the beginning, subsequent queries will go more smoothly (e.g., “For Cases in the ‘Approval Request’ App…”).
Prompt Examples (Case Analysis)
Below are example questions. Enter them in your AI’s chat with the connected connector enabled.
Status Check / Inventory
How many Cases are currently running in the "Approval Request" App?
Show me a list of Cases started in the last week in the "Invoice Issuance" App.
Aggregation / Classification
For the "Approval Request" App, tell me the number of Cases by status (Running / Completed).
In the "Expense Report" App, count the Cases completed this month by originating department and sort them in descending order.
(The AI retrieves the list and counts the results. Since there is no server-side aggregation mechanism, narrowing the scope by App, time period, etc. produces more stable results when dealing with large numbers.)
Filtering / Extraction
Extract Cases from the "Approval Request" App where the amount is 1,000,000 yen or more, and list them with their Title and amount.
(When filtering by an item like amount, the AI first checks the Data Item definitions and then performs the search. If you know the item name, explicitly stating it makes the process more reliable.)
Deep-Dive / Summary of Individual Cases
Summarize the Data Items of Case pXXXXXXX.
In the "Approval Request" App, select the Case that has been running the longest and explain its progress so far (who processed which step and when, and whose turn it is now).
The details of an individual Case include not only each Data Item but also the step history (operator, processing date/time, and the current step). This allows the AI to answer questions like “explain the progress” or “who is it currently waiting on.”
Checking for Stalled Cases
For running Cases in the "Approval Request" App, tell me which step each is currently at, with the count per step.
Trend Analysis
For Cases started this quarter in the "Cancellation Request" App, tell me the trends (changes in volume and notable characteristics).
You can continue the conversation. It is effective to drill down based on results, such as “Now show me the top 5 by amount in detail.”
Tips for Effective Analysis
- Specify the target App first. Stating “which App’s Cases” at the beginning reduces the chance of the AI misidentifying the target.
- Use item names to specify criteria. For filtering and aggregation criteria, use the item names displayed on the Questetra screen (e.g., “Amount,” “Requesting Department”). Internal field names such as
q_string0cannot be retrieved, so item names are the reliable approach. - Narrow the conditions when the volume is large. List retrieval returns the total count (the total number of matching Cases), so questions like “How many are there for XX?” can be answered in a single query. On the other hand, enumerating or aggregating the contents requires retrieving the list. While each retrieval defaults to 100 items with pagination support, having the AI page through many pages to process all items is unreliable (there is no server-side aggregation mechanism either). Narrowing the scope by time period, status, App, or Data Item values produces stable answers.
- Be explicit about aggregation definitions. Words like “running,” “completed,” or “this month” can be supplemented with “from when to when” or “which status exactly” to reduce discrepancies.
Limitations and Caveats
- Do not take numbers at face value — verify them. AI may produce plausible-sounding answers that “go along with” the question. For important counts and amounts, cross-check against Questetra’s list and aggregation screens.
- There is a practical limit on how many items can be aggregated. For simply checking “How many XX are there?”, the total count is included in the response regardless of volume. However, when enumerating and aggregating contents (e.g., by department or amount range), the AI receives and counts the list data, so roughly up to 100 items is reliable. Several hundred is possible depending on conditions (but as retrieval limits and AI output limits are approached, items may be missed). Volumes exceeding 1,000 are not suited for AI-based aggregation. When the volume is large, narrow down by time period, status, or Data Item values, or use Questetra’s built-in Case/Task aggregation features.
- You can only see Cases within your own permission scope. The data that can be retrieved is limited to the scope of the authorized user’s permissions. The same question may return different counts for different people.
- Some information cannot be retrieved. With the current tools, internal field names of Data Items (e.g.,
q_string0) and detailed permission information per step are not returned. - Read-only. The AI will not update Cases or process tasks (read access only).
Once the connection is set up, all you need to do is ask in natural language. Start with a simple question like “How many running Cases are in the XX App?” and gradually drill deeper based on the results. You can leverage AI as a knowledgeable consultant for your Questetra data.
