# How Rely trusts Datalab to process massive data rooms in hours, not days

> Rely audits commercial real-estate transactions worth millions in misvalued assets. Running Datalab on-prem at 12.5 pages/sec lets them ingest 500,000-page data rooms before deals close — without resident documents ever leaving their network.

- Canonical: https://www.datalab.to/blog/datalab-rely-case-study
- Published: 2026-01-16
- Authors: Datalab Team

![Datalab Case Study - Rely](/images/blog/datalab-rely-case-study/rely-blog-image.png)

[Rely](https://www.tryrely.ai/) is building an AI-powered due diligence platform for commercial real estate transactions. Their product automates what traditionally takes weeks: auditing thousands of lease documents, applications, financial statements, property records and more to reconstruct accurate financial views of real estate assets.

The Portland, Maine-based startup is tackling a massive problem: when apartment buildings and large properties are bought and sold, thousands of resident documents must be manually reviewed to extract critical data. This process is expensive, time-consuming, and prone to errors that can cost millions in misvalued assets.

## The Challenge: Scaling Ingestion Without Sacrificing Trust

Rely's document processing problem had three dimensions that standard solutions couldn't handle:

- **Extreme volume with tight deadlines:** Typical deals contain 2,000 documents (20,000 pages), with portfolios reaching 500,000 pages. With 2-4 week due diligence windows, processing needed to complete in hours, not weeks.
- **Complex layouts that break standard OCR:** Real estate files include 10-100 page lease documents with embedded tables spanning multiple pages and unlabeled PDFs merging multiple documents. Standard OCR tools consistently broke on layout understanding and scanned documents.
- **High-stakes traceability:** A $100,000 cash flow error translates to a $1 million reduction in asset equity value. Rely needed accurate extraction with bounding box precision to cite back to specific sources.

After evaluating multiple solutions, Rely hit roadblocks with each:

- **LLM-based markdown conversion:** Limitations with time per output token, difficult to scale faster and expensive. Wrong tool for the job
- **Reducto:** Cost structure becomes prohibitive at standard data-room volumes
- **Alternative OCR solutions:** Layout detection was fragile or not robust enough to capture data points consistently and accurately

## The Solution: A Production-Grade Ingestion Backbone with Datalab

Rely began in September after trying "literally every possible doc processing thing we could find." For John, who had worked with OCR since 2019, Datalab was a breakthrough:

> "The combination of the speed, the accuracy, and just the ease of use—it kind of just offloaded that huge part of our problem to a product that has it solved so that we could focus then on the real guts of the problem, our business logic."

Rely deployed Datalab's on-premise Docker containers to convert documents into structured markdown with layout and positional data—creating a durable ingestion layer they could index, classify, and extract from without repeatedly resending raw documents to external systems.

## Key Outcomes

- **~12.5 pages/sec throughput** using distributed on-prem infrastructure
- **Support for massive scale**, including data rooms beyond 500,000 pages
- **Reliable processing** of 300+ page documents with complex layouts
- **Full pipeline ownership**, enabling fast iteration without reruns
- **Improved trust**, with outputs traceable back to source documents

## Impact: From Bottleneck to Competitive Advantage

Datalab transformed document processing from a limiting factor into a core competitive advantage:

- **95+% Time Savings:** Data rooms that previously consumed entire 4 week diligence windows can now be audited on the first day they're received.
- **3x faster throughput:** 12.5 pages per second versus 4 pages/second on DataLab's API service.
- **Unlimited scalability:** Scale up resources on-demand for 20,000 document deals, then scale down—throughput is user controlled rather than rate limited.
- **Complete pipeline ownership:** Iterate on classification logic and audit workflows without re-processing source documents.

## Looking Forward: Scale

Rely is already working with some of the largest property management firms in the US and plans to scale rapidly this year. With Datalab as their document processing foundation, the team is confident they can scale to enterprise demand without compromising speed, accuracy, or trust.