A large part of expert work consists of steps that always look the same. We automate them so that the expert no longer has to “produce” every record by hand, but only checks and resolves the exceptions that genuinely need their judgment.
What institutions come to us with
- Staff repeatedly enter, retype and check large volumes of data — with many clicks and switching between screens.
- The digitization line has a bottleneck in manual scan cropping, the creation of structural metadata or output checks.
- The volume of unprocessed collections far exceeds the team’s capacity.
- The existing system does its main job, but lacks a specific tool: an editor, a review interface or a specialized task that a technician currently runs by hand with a script.
Decision pattern
We first split every process into three parts:
- Fully automate: regular and unambiguous cases.
- Pre-process and have it confirmed: a proposal that the expert only reviews.
- Leave to the expert: unusual and ambiguous situations. AI suggests, you decide.
How we work
- We analyze the whole process, not just individual screens — it often turns out that a significant part of the work does not need to be done by hand at all.
- We design the workflow and integrations. We explore integration via API, database or exports. For older and poorly documented systems we choose safer read-only solutions to avoid risks to data consistency.
- We deliver a tool with a review interface. In it, the user starts a task, enters parameters, follows its progress, sees errors and works with the result.
- We build modularly. Individual steps — cropping, OCR, metadata, identifiers, checks — can be replaced or extended without depending on a single vendor.
What you get
- Significantly less routine work — not just fewer clicks, but fewer steps that a person has to do.
- Higher capacity without a bigger team and stable, consistent output quality.
- New features without replacing the existing system.
- Fewer errors and better scaling as data volumes grow.
Where it already works
Cropilot automatically crops scans using an AI model. It handles the operation of large lines with tens of thousands of scans a day and offers an editor for reviewing exceptions. For atypical collections such as historical prints, an institution can fine-tune its own model: it is enough to manually correct a few hundred pages and the model processes the rest. Thanks to group and permission management, the work can be shared among several people, including remote work. AI Cataloging extracts bibliographic data from photographs of the title page and colophon, looks up authorities and prepares a draft MARC 21 record that the cataloger reviews and confirms. We are developing the tool together with catalogers from the Moravian Library.
Transferable to: corporate documentation, archives and records management and technical documentation.
