Skip to main content

How to use the AI Chat to build custom reports

Written by Cristian Henkel

Complif's AI Chat answers questions about your operational data in plain language, builds tables you can download as Excel and charts you can download as images.

Before you start

The chat works with your permissions. It only reaches information you can already see in the platform. Two people with different roles can get different results for the same question.

How to ask for a report

Write the question the way you would ask an analyst. You do not need field names or special syntax.

A question that produces a good report usually has three things:

  • The entity. People, accounts, cases, onboardings or requirements. This is what changes the result the most: the number of people and the number of accounts are different figures, so name the one you need.

  • The grouping. By analyst, by advisor, by status, by stage, by template, by risk, by country, by month.

  • The period. "This month", "the last year", "between January and March". Without it, the chat assumes the running total.

A complete example: "Give me the onboardings created this year grouped by analyst and company type."

Downloading the result

Tables in Excel. When the answer includes a table, the "Descargar XLSX" button appears above it. The file carries the full data, ready to continue the analysis in a spreadsheet or share it.

Charts as images. When the answer includes a chart, "Ver grande" and "Descargar PNG" appear. Useful for dropping the result into a deck or a report. The chat produces bar, line, pie and doughnut charts.

If you want a specific shape, ask for it: "show it to me as a pie chart" or "build a table with that".

Reports you can ask for

These are requests the chat resolves with precomputed metrics, so the numbers stay consistent.

Team workload

  • "How many cases does each analyst have assigned?"

  • "What is the average time cases stay open, by analyst?"

  • "How many active and pending accounts does each advisor have?"

Risk

  • "Pending onboardings grouped by risk matrix result."

  • "Account risk grouped by advisor."

  • "Which accounts moved from low to medium risk?"

  • "How many PEP people do I have?"

Documentation and due dates

  • "Which onboardings have missing documentation?"

  • "Which requirements are due this month?"

  • "High priority cases due this week, grouped by template."

  • "Number of requirements by template."

Portfolio analysis

  • "Which accounts generated the most cases?"

  • "Which accounts transacted the most relative to the cases they generated?"

  • "People who appear in more than one account."

  • "Accounts with more than four representatives."

  • "People grouped by economic activity."

  • "Accounts approved by month over the last year."

Getting better results

Name the right entity. "People by risk matrix result" and "accounts by risk matrix result" are two different reports with different numbers. Ask for one expecting the other and the result will look wrong without being wrong.

Narrow the period. A request with no dates over a large operation can be slow or too broad. If the chat says the query is too heavy, add a range or a filter.

Ask for the detail in a second step. First the aggregate ("how many by status"), then the list ("list the pending ones"). It is faster and easier to review.

Use the screen you are on. With an onboarding open, you can ask "what documents are missing in this onboarding?" without naming it. To step out of that context, say "across the whole organization".

If you need a grouping that does not exist, the chat says so. When there is no metric for that dimension it tells you, instead of improvising a number.

Working with your own files

You can attach files to the chat and ask it to work on their content. For example, hand it a spreadsheet with a list of companies and ask which ones already have an account in the platform.

  • Formats: PNG, JPG, PDF, TXT, MD, CSV, DOCX and XLSX.

  • Up to 10 files per message and 10 MB per file.

  • Up to 50 MB of attachments per conversation.

Always review the result

The screen itself warns you: Complif AI can make mistakes, so check the answers before using them.

Two quick checks before calling a report good:

  • That the total matches the number of rows listed.

  • That the entity and the period are the ones you asked for.

If a number does not add up, ask the chat to show you the detail behind that total. For decisions reported to a regulator, cross-check the result against the matching screen in the platform.

Did this answer your question?