Caddi finds the work worth automating. And shows its evidence.

Activity in the tools you have connected, under the access you already have in each of them. It reads reports, counts and recent activity rather than walking through records one by one, so finding your processes does not turn into a spike in API usage. It does not write anything, and it does not reach a system you have not connected.

A run does not repeat on its own either. Once Caddi has read a set of tools it does not offer to read them again for 30 days unless you connect a new one, because an unchanged environment returns the same findings.

From what it measured: how often the work occurred, how long each occurrence took, over the window it observed. Every finding shows those three numbers, so you can redo the arithmetic yourself.

Read it as the cost of the process, not one person’s time. Where a tool is shared, the count includes everyone who touches that work, so the figure for a shared inbox is the team’s figure. It is a way to rank findings and start a conversation, not a forecast.

Sometimes, and that is useful. If Discover surfaces work you already automate somewhere else, it has found the real work, which is a check on everything else in the list. The findings that matter are the ones nobody had written down.

Because Discover reads with your access, not the firm’s. That is why the pattern is one run per team, by someone who does that team’s work.

No. The contents of your connected systems are never used to train models. OpenAI and Anthropic calls run under zero-retention terms, and accounts under a Data Processing Addendum exclude their content from model development entirely.

Yes. Each finding can be automated in Caddi, exported to Claude as a skill or exported to Microsoft Copilot as an agent, whichever is the best tool for the job. Caddi is the right pick for high-volume, high-accuracy work; Claude and Copilot for the judgment work in a person's day. The exported skill or agent reports each decision and API call back to Govern, next to every Caddi run, so leadership sees what is automated, where, and what is still done by hand.

Yes, it is AI process mining, but it does not start from event logs or tickets, which only show the work that is already tracked. It starts from hypotheses grounded in how firms like yours run and checks them against your systems, so it finds the untracked work too, with no event-log extraction or recorder on every desktop. The bigger difference is what comes next: process mining stops at the insight, while Caddi goes on to Automate, where the person who does the work builds the automation, and Govern, where every run and decision is logged. See what process mining is, or compare Caddi to Celonis, Power Automate Process Mining and Skan.ai, or see Caddi for process mining.

Every plan. It is on the free Individual plan for one user with no time limit, and full use of Discover is part of every Business and Enterprise pilot, where we run it across your team's tools with you. See pricing.