How Claimer reduced R&D report extraction and processing time by 95%

The Challenge

Early in 2023, the UK passed legislation requiring all consultancies and companies to submit extensive documentation, in addition to their normal tax submission, to claim R&D credits.

Claimer wanted to quickly come to market with a product that would automate the submission process for their users. The new product would extract the data from user-uploaded R&D reports, reformat the extracted text, and log it into the tax form on the government’s website.

Extracting data from lengthy - and often inconsistently-formatted - R&D reports presented a major technical challenge, so Claimer researched reliable document extraction platforms to power this functionality.

Claimer laid out four criteria for evaluating potential extraction solutions:

Short implementation timeline

• With a hard deadline for the product’s release, Claimer needed a solution that could be quickly integrated and deployed. This ruled out traditional OCR providers, like Amazon, Google, and Microsoft, which would require Claimer’s developers to build document processing infrastructure atop the engine.

We felt that we probably could have built document extraction functionality internally, but it would’ve taken several weeks. And knowing what I know now, it would have actually been significantly more than that.

Fully-featured API with a functional web-based user interface

• Claimer prioritized a fully capable API along with the ability to test in different staging environments on a web interface for optimal flexibility and ease of use.

Start-up friendly pricing

• While Claimer was willing to pay for a reliable solution, they needed to be cognizant of their budget as a startup. Other solutions were geared towards high document volumes with enterprise pricing, which were cost prohibitive.

Configurability for variable document formatting

• Extraction accuracy was crucial for the effectiveness of Claimer’s product. The team needed to be able to customize and configure extractions to guarantee high accuracy levels, even across highly variable user-submitted documents

Sensible was a clear front-runner when it comes to getting together a working solution as quickly as possible, and the ability to handle documents in a way that’s really configurable, all within our budget.

‍

The Solution

Where there is configuration and flexibility, there is power, so I was very happy with the level of configurability Sensible offered.

Sensible emerged as a front-runner across Claimer’s four criteria, and the team immediately jumped into the platform to get hands-on experience. The Claimer team quickly picked up SenseML’s syntax to create configurations for the first few document types thanks to the developer documentation. After some adjustments to account for report variances, SenseML’s anchoring techniques reliably returned the accurate extractions that Claimer required.

Claimer also worked closely with the Sensible team to identify a handful of new features that would bolster performance, including sub- and superscript annotation. Sensible developed five new features and fixes to support Claimer’s needs while supporting an aggressive launch timeline.

The Sensible team was so responsive to our needs. We had some pretty hard deadlines and Sensible absolutely delivered.

‍

The Results

Claimer’s automated tax submission product launched in August, meeting the team’s initial deadline that coincided with the UK’s new compliance requirement going into effect. Three months after the August launch, Claimer reached full automation for nearly all of its customers, and has extracted data from thousands of R&D reports.

With Sensible powering their document extraction capability, Claimer decreased R&D report submission and extraction time for their users by over 95%, going from 45 minutes per document to under two. For accountants and consultants in the middle of tax season, this kind of time-saving is an invaluable benefit. In less than six months, this accounted for more than 3,778 manual extraction hours saved.

Claimer realized that it would’ve taken weeks, if not months, to build their own document extraction functionality from scratch that delivered the same results as Sensible did out of the box.

These days, there's no point in building this functionality yourself if you can find a solution in the market that is cost-effective and sufficiently mature to work in a production environment where you're dealing with thousands of extractions a day. It just doesn't make financial sense to build it yourself. So, survey the market and look for something that exists — in this case Sensible fits the bill