Connect with us.

Tell us how we can help and we’ll get back to you soon.

400-22 E. 5th Avenue, Vancouver, BC, Canada
Follow us on LinkedIn
Please provide a correct email.
Please provide a correct email.
Dropdown
Please provide a reason.
Please provide a correct email.
Send Message
Thank you for contacting LexSelect! A member of our team will be in touch as soon as possible.
Oops! Something went wrong while submitting the form.

Switching from Google to LexSelect

Customer
instaSpace
Industry
Legal technology
Use case
Parsing scanned contracts and legal documents for contract review
Primary integration
LexSelect API inside instaSpace’s contract platform
Contact us
Contact us

Overview

instaSpace is an operating system for legal teams in enterprise business units. It covers the full contract lifecycle, from intake and negotiation to approvals, execution and renewal tracking. Its clients include banks and government entities, and it processes more than 30,000 pages a month. About half of its users are in the GCC, where most contracts are Arabic or bilingual and scans are part of everyday legal work.

Every document has to be read before instaSpace can do anything with it, and general-purpose OCR kept failing on the files lawyers rely on most: templated legal forms, long scans and Arabic documents.

instaSpace now sends PDFs and scans through LexSelect. It returns the text, the structure and the position of every line.

In instaSpace’s tests, Google cut off 7 of 16 sample files, leaving more than a third of the pages unread. LexSelect processed every page with no cut-offs and recovered about 50% more text, including an 81-page Arabic exhibit bundle that it also split into 24 labelled sub-documents.
16-file test results: Google OCR cut off 7 files, while LexSelect read every file with content in full16-file test: Google OCR cut off 7 files, LexSelect read every file with content in full

The problem

In the month before the switch, 2.3% of PDF uploads failed extraction outright with the previous provider, and 1.9% of uploads across all formats.

Templated legal forms got cut off
Google’s OCR applies a content filter that stops transcribing when the output looks like reproduced boilerplate. Standard legal forms and pleadings trigger it, so files often stopped after the first page or two.
Arabic and long files broke
Other tools capped processing at 50 pages and returned Arabic in scrambled reading order. Very large scans were a challenge too.
Hard documents broke contract review
The previous parser gave instaSpace little control over document structure, so key-term extraction and table review struggled on documents that were hard to parse.
Engineering time went into infrastructure
OCR, page handling, language support and edge cases in scans all needed the team’s attention.

The solution

How instaSpace rolled it out

Arabic support was the one hard requirement. Rather than switch everything at once, instaSpace’s team:
1. Sandbox test
Tested LexSelect in a development sandbox on their own documents, starting with Arabic and bilingual contracts.
2. 16-file test
Ran a representative set of 16 files through it to check coverage and speed: scans and text PDFs, English and Arabic, 1 to 158 pages.
3. Production
Moved PDFs and scans to LexSelect in production, keeping the previous pipeline as a backup.

The LexSelect features they rely on

  • Full documents, no page cap. Long scans come back complete.
  • Arabic and right-to-left reading order. LexSelect detects the language and reading direction of each page.
  • Document structure. Document type, sub-documents, clauses and tables.
  • Line positions. The position of every line on the page, not just the text.
  • One API call, with webhooks. Everything comes back from a single call, with a webhook when processing finishes.

instaSpace is one of LexSelect’s early partners, and its feedback shaped what we built. Its request for Arabic grew into support for more than 200 languages, including right-to-left scripts, along with request limits per API key and per team member and workspace-wide logs. All of it is now available to every LexSelect customer.

Results

How it works now

Here’s how a document moves through instaSpace now:
User uploads a contract, LexSelect reads every page in one API call, instaSpace extracts key terms, and a lawyer reviews them in a table
User uploads a contract, LexSelect reads every page in one API call, instaSpace extracts key terms, and a lawyer reviews them in a table

Three documents that show the difference

All three were hard scans with complex layouts, the kind of files that broke other tools.

Templated legal forms and pleadings

English, scanned, 2 to 20 pages. These were the files Google cut off. LexSelect reads them straight through, however templated they are, at about 40 seconds for a 12-page form.

An 81-page Arabic exhibit bundle

Scanned, right to left, and one of the toughest files instaSpace tested. LexSelect read all 81 pages in the correct right-to-left order in 1.5 minutes. It also recognized that the file was a compilation of separate exhibits and split it into 24 labelled sub-documents, which instaSpace had not asked for.

A 158-page English contract

Scanned, and the largest file in instaSpace’s test set. At 158 pages, it’s more than three times the 50-page cap other tools hit, and the kind of very large scan that had been a challenge for instaSpace before. LexSelect returned the entire document in 7 minutes without difficulty.

Measured results

The 16-file test set covered scans and text PDFs in English and Arabic, from 1 to 158 pages.
Measure
Previous provider
LexSelect
Files with usable text
14 of 16
15 of 16*
Page coverage
63% (441 of 696 pages)
100%
Characters recovered
743,771
1,115,944
Processing time per page
About 0.75 seconds
About 2.5 seconds**
Median end-to-end time
27 seconds
74 seconds**
Time the user waits
Up to 45 seconds
About 8 seconds

*The 16th file was a blank 1 KB PDF.

**LexSelect takes longer per page because it reads every page and returns the document’s structure and the position of every line, not just the text. Of the 74-second median, only about 8 seconds keeps the user waiting.

What changed

For lawyers using instaSpace
Scans now arrive as fully usable documents, not just text. Because LexSelect returns the position of every line, the assistant highlights the exact lines on the scanned page when it cites a passage. That wasn’t possible with the text-only OCR they used before.
For the instaSpace team
Document extraction is no longer something they build and maintain, so engineers can focus on legal workflows, drafting and review.

Customer feedback