Keboola is used as the main tool for data transformations, extraction, writing, and resolving all the data challenges.
In a world that is growing progressively digital, hours of human talents are wasted on reading and organizing receipts, scanning through printed contracts for the relevant clauses, and cataloging documents.
Traditional OCR solutions that automate document processing are slow, expensive, and error-prone.
Rossum developed a proprietary AI engine that automatically extracts meaning from documents - cutting through the operational hurdles of finance, legal, and logistics departments.
Rossum embedded machine vision and other machine learning solutions within their cloud-based platform to speed document pre-processing (deduplication, ingestion, …), content extraction, validation, and automated document processes end-to-end.
We talked to Josef Pithart, who is the Head of BI at Rossum to discover how data is used to fuel the development of such a state-of-the-art AI product and what are the operational data challenges behind the scenes.
Josef is one of the pioneer users of Keboola. He started using Keboola for automating ETL pipelines over 8 years ago!
As a data veteran with 12 BI years under his belt, he deployed data pipelines anywhere from fashion e-commerces like the pan-European Zoot to digital banking startups in Germany.
Then one day he was called by the leaders of Rossum with a specific challenge: to build the Business Intelligence capability for Rossum.
“They said: ‘We need to implement BI and you can choose the technology you want to do it on.’ Which to me was clearly Keboola. Immediately, I told them that Keboola is straightforward to use, it can do everything they need, it's not expensive, and one or two people can run it even for a big company, so it saves money.”
After Rossum made sure Keboola passes all the strict security standards and is certified according to data regulations, the team at Rossum started playing around with Keboola as the main tool for data transformations, extraction, writing, and resolving all the data challenges.
The main use case for Keboola at Rossum is the building of end-to-end BI reporting from scratch.
With Keboola, Josef and a product analyst are capable of building the end-to-end BI operations as a two-person team.
Keboola helps them extract data from AdWords, Facebook, Linkedin, Google Analytics, and SalesForce with the prewritten Extractors. They automate the data collection from 3rd party apps without needing to write or maintain additional code. Later they link the data with the backend and product usage data to enrich it. With Keboola, they also validate the data, before sending it to various visualization tools.
Rossum’s main advantage stems from machine learning. Their visual recognition algorithms outrank the competition at a fraction of the price for their customers.
But to keep ahead of their competitors, they need to keep innovating. This translates into analyzing the data, develop new features, and building POCs to squeeze out additional optimizations or new experimental features that delight their customers.
“We have people who know Python, they know how to dig into data, but they couldn't work on our internal databases, because their work would crash them. So they go into Keboola where they can start the machine, upload data, make a sandbox, do the analysis, and build models. When they have something done, they can just come in and it's easily reproduced in Keboola to make it work in production.”
With Keboola, Rossum’s data department can dive deep and analyze data on the same models as production and BI. With the use of Keboola’s Sandboxes, they abstract from the resources devoted to production to avoid burdening their customer-facing applications. And with the devoted data science tools, they can easily build production-ready models faster using the frameworks they already know.
Rossum uses Keboola to perform reverse ETL, a.k.a. send enriched data from their databases to other tools where the data helps them move the business needle.
The best example of that is their churn model. With the churn model, Rossum built an algorithm that predicts which customer is likely to say goodbye. By sending the churn risk into SalesForce, they can quickly determine the risk of an account churning and work on churn prevention.
Another business area where Rossum taps into the power of data is Account Potential. By building models that predict how much an account could buy and comparing the potential to the actual services and products each account has already bought, they can quickly identify which accounts can more easily be expanded and upsold to new services.
Beyond the three use cases, how does Josef plan to expand data operations at Rossum?
Josef plans to automate a lot of the data processes across the board.
One area of interest is their churn prediction algorithm. Instead of ingesting churn risk into SalesForce, he plans to automate notifications for account managers when a customer is at high risk.
This is just one of the predictive analytics in the plan for the upcoming future. Data and smart analytics will also be used to better predict sales, hire better candidates, and anticipate support issues.
All of the predictive abilities rest on the assumption that the data which is used to build the prediction model is accurate and valid.
With Keboola, Josef and his team can collect, clean, validate, and model data at scale. A feat, which would be hard to achieve on his own.
As Josef said:
“I figured that without Keboola, it didn't make sense to do it. Because the work that we can do with 1 or 2 people using Keboola would be done in the same time by 10 people without Keboola.”
Ready to discover what Keboola can do to help your business? Get in touch and let us start building a better data future.
.png)