Migrating existing contracts to a CLM: method, OCR and pitfalls

Start by knowing what you have. Contracts are rarely in one place: shared servers, email inboxes, e-signature tools, filing cabinets, a tracking spreadsheet, sometimes an old software tool.

List these sources, with an owner and an approximate volume for each. An imperfect count is enough at this stage. It helps size the project and spot the areas no one knows about.

Then set a simple objective. Do you first want to track deadlines, find contracts, or prepare a broader analysis? The answer determines which contracts to process first and which data to migrate.

Migrating everything at once is rarely necessary. A batch approach lets you correct the method before extending it.

• First batch: high-stakes contracts in force, or those with an upcoming deadline.
• Second batch: the other contracts in force, by type or by team.
• Third batch: expired contracts that must remain accessible, with a lighter level of data.

Also decide what will not be migrated: duplicates, drafts that were never signed, working versions. This decision is made with the legal department, because a document left out of the new tool is not necessarily a document to destroy.

Data is only useful if someone uses it. Define a short record, then add fields only when a use justifies them. The core usually covers:

• Identification: contract type, title, parties and the group company concerned.
• Dates: signature, effective date, end date and notice deadline.
• Renewal: none, automatic or express, with the length of the notice period.
• Tracking: status, internal owner and team concerned.
• Links between documents: initial contract, amendments and appendices.
• Specific: amount, price review index or other information specific to your activities.

Write a rule for each field. How do you enter a company that has changed its name? Which date do you keep when an amendment changes the term? Without a rule, two people will fill in the same record differently. The filter or reminder you expect will then rely on inconsistent data.

OCR, or optical character recognition, turns the image of a document into text. It makes a scan searchable and usable. It does not understand the contract: that step belongs to extraction, which comes afterwards.

Its quality depends on the document. The Google Document AI documentation, for example, recommends at least 200 dpi and states that 300 dpi and above generally give the best results. It also notes that downsizing lossy formats, such as JPEG, can reduce accuracy, and that for handwritten documents accuracy depends on the quality of the handwriting. These benchmarks apply to that tool and vary between products.

In a real portfolio, the difficult cases are well known:

• copies of copies and old faxes;
• skewed or poorly framed pages;
• handwritten notes, dates added by hand and signatures that overlap the text;
• stamps and annotations in the margins;
• price tables and appendices in small print.

Good practice is therefore to test before deciding. Take a sample that reflects the diversity of the collection, including the worst documents, and see what comes out.

Automatic extraction saves reading time, but it can be wrong. Checking is part of the method, and it must be built into the schedule.

• Match the check to the stakes. Sensitive contracts are checked one by one. Routine contracts can be checked by sampling, if the team agrees.
• Check first the fields that trigger an action. A wrong end date or notice period directly distorts the reminder that depends on it.
• Measure on your sample. Record errors by field and by document type. This result is worth more than a general accuracy rate.
• Compare with the source text. The value must be comparable with the passage of the contract it comes from.

Pay particular attention to wording that requires a calculation, such as a notice period expressed in months before an anniversary date. The decision to keep a value or not remains with your team.

• Forgotten amendments. An end date read in the initial contract may have changed since. The applicable text is read together with all its amendments.
• Incomplete contracts. Missing appendix, unsigned copy, missing general terms and conditions: flag it rather than hide it.
• Contracts that have already ended. Without verification, they feed reminders and distort statistics.
• Forgotten access rights. Employment contracts or sensitive files must not become visible to everyone during the import.
• Double entry. If the old spreadsheet stays in use after the cutover, two references coexist. Set a date after which it is no longer updated.
• The backlog without the flow. Migrating existing contracts is not enough. Decide how each newly signed contract will get into the tool, including those signed elsewhere.

Do not destroy anything by reflex. Scanning a contract to make it accessible is one thing. Replacing the original is another.

Article 1379 of the French Civil Code provides that a reliable copy has the same evidential value as the original, and that reliability is left to the judge’s assessment. A copy is presumed reliable, unless proven otherwise, if it identically reproduces the form and content of the instrument and if its integrity is guaranteed over time by a process that complies with a decree. If the original still exists, its production may always be required.

Decree No. 2016-1673 of December 5, 2016 specifies this process for electronic copies: information on the digitization context, quality tests, a digital fingerprint, retention of traces and documentation. Importing a scan into a software tool is therefore not enough, on its own, to meet these conditions.

Contracts also contain personal data. If a provider processes it on your behalf, Article 28 of the GDPR requires a contract or another legal act. Check this point and the list of subprocessors before the import.

Docform imports your contracts in bulk into the contract repository, hosted in France. OCR makes scanned PDFs readable, then the AI suggests the parties, dates, term and notice period. You see the passage each value comes from, and your team validates.

Folders, tags, custom attributes and permissions then organize the portfolio. The rollout can start with a targeted scope, and migration is part of the onboarding support. Data remains exportable to Excel or CSV. The Trust page details the subprocessors and the GDPR framework.