AI Task Spotlight | Edition No. 12: Metrics on Demand
Each month, we spotlight a real-world AI task in Unit21. This edition is about the Unit21 MCP, which connects an AI assistant such as Claude to your Unit21 reporting. You ask a question in plain English and get the numbers back in seconds, without leaving the conversation.
Most Spotlights show an agent doing work inside the platform. This one is about getting answers out of it: the alert counts, rule volumes and SAR rates that every compliance team gets asked for, usually with a deadline attached.
Why one question takes two days
Monday, 9:02 AM. The board meets Thursday, and the chief compliance officer wants to know how many alerts turned into SARs last quarter, and why the backlog is growing.
The old way is familiar. Export the alerts. Clean the CSV. Build the pivot table. Check it twice. Two days later, you have a number.
Then the follow-up arrives. Which rules drove the volume? Is the backlog older than it was last quarter? How are the AI Agent's reviews holding up? Each one is a new export, a new pivot and another file named final_v4.
The analysts who could be working cases are rebuilding the same spreadsheet, and the answer is already stale by the time it reaches the meeting.
The question was never hard. Getting to the answer was.
Introducing the Unit21 MCP
MCP, the Model Context Protocol, is an open standard that lets an AI assistant connect to the systems where your data lives. The Unit21 MCP is a remote server: there is nothing to install. You give your assistant the URL for your Unit21 environment, sign in to Unit21 the first time, and the assistant can answer questions from your reporting.
You ask the way you would say it out loud, and the answer comes back with the numbers, ready to use. There is no export, no CSV and no pivot table in between. The assistant can also search the Unit21 product documentation, so a question about a feature or a setup step gets answered in the same conversation.
It works with Claude, ChatGPT and Cursor, and with any other assistant that supports remote MCP servers. We covered the MCP in our August product update and added report filtering in September.
What you can ask
The MCP is built for the questions that come up in every reporting cycle:
- Alert volume over time
- Alerts per rule and per queue
- Backlog aging
- Case to SAR rate
- Any date range, by agent or tag, by entity or country
- Alerts the AI Agent flagged as false positives
Because you are talking to an assistant, a follow-up costs a sentence instead of a rebuild. Ask for last quarter, then ask for the same view by queue, then narrow it to one country. The assistant keeps the thread.
The demo: four questions in a row
In the video, we asked four questions back to back in our demo environment, the way a compliance lead preparing for a committee meeting would.
- The funnel. How many alerts became cases, and how many of those became SARs. This is the number the board asked for, and the one that usually takes a day.
- The top rules. Which rules generate the most alerts, so you can see where the volume is coming from.
- The AI Agent. How the AI Agent is performing on alert review, including the alerts it flagged as false positives.
- The committee summary. A short write-up of the numbers, ready to drop into the board pack.
Each answer came back in the conversation, with the numbers on screen, in less time than it takes to open the export screen.
Built on your permissions
The assistant uses your Unit21 permissions. It can only reach the data and tools you can already access, and sign-in goes through Unit21 with OAuth.
If an administrator turns MCP off for the organization, or removes someone's permission, assistants connected under that account lose access immediately. It reads your reporting data and never changes an alert or case.
How to get started
Setup takes a few minutes. The full guide, including the server URL for each Unit21 environment, is in the Unit21 MCP docs.
- An organization administrator accepts the MCP Access Agreement and enables MCP in Unit21, under Settings, Integrations, MCP.
- The administrator grants each user the Use MCP read access permission in the permissions dashboard. Reporting tools also need the read:insights permission, plus access to alerts, cases or SARs for the data you want to query.
- Add the Unit21 MCP URL for your environment to your assistant. In Claude on Team and Enterprise plans, an Owner adds the custom connector for the organization first, then each person connects and signs in to Unit21. On Pro and Max plans, you add the connector yourself.
- Sign in to Unit21 when the assistant asks, approve access, and start asking questions.
Questions teams ask
Do we need to install anything?
No. The Unit21 MCP is a remote server. You add its URL to your assistant and authenticate.
Can it change our alerts or cases?
No. It reads your reporting data and the documentation. Your investigators still work alerts and cases in Unit21.
Why can I connect but not see reporting tools?
Reporting tools need the read:insights permission and access to the relevant alerts, cases or SARs. Ask an administrator to check your permissions.
Which environment URL do I use?
The one that matches your Unit21 environment. Production, EU and sandbox environments each have their own URL, listed in the docs, and connecting to the wrong one will not reach your organization.
Why this matters
Compliance reporting is rarely hard in concept. It is slow because the data sits one export away from the people who need it. Pulling data, cleaning it and rebuilding the same pivot every quarter is time your team could spend on investigations. For the numbers worth tracking each quarter, see our guide to AML case management metrics and KPIs.
When the answer is one question away, reporting stops being a project. The CCO gets the number before the meeting, the follow-up gets answered in the room, and the analysts who used to build the spreadsheet get their time back.
Metrics on demand.
About the series
AI Task Spotlight covers one task from Unit21's AI library in each edition. Sign up for a demo to see it on your own data, or subscribe to get the next Spotlight. Catch up on past editions: Edition 11 on structuring and Edition 10 on ransomware red flags.
Gal Perelman is the Product Marketing Lead at Unit21, where she spearheads go-to-market strategies for AI-driven risk and compliance solutions. With over a decade of experience in the fintech and fraud sectors, she has led high-impact launches for products like Watchlist Screening and AI Rule Recommendations.
Previously, Gal held marketing leadership roles at Design Pickle, Sightfull, and Lusha. She holds a Master’s degree from American University and a Bachelor’s from UCLA, and is dedicated to helping banks and fintechs navigate complex regulatory landscapes through innovative technology.
