We moved from paper-pushing to bit-pushing. Now we're entering a new era of productivity.
By Vedika Jain on April 15, 2024
Technology promised to kill paper-pushing. But, we've just digitized the drudgery. We've moved from paper pushing to bit-pushing.
John Warnock, one of Adobe’s founders outlined his vision for the PDF:
“Imagine being able to send full text and graphics documents (newspapers, magazine articles, technical manuals etc.) over electronic mail distribution networks. These documents could be viewed on any machine and any selected document could be printed locally. This capability would truly change the way information is managed.”
The PDF, now over 30 years old, ushered society’s shift toward digitization. It made processes more efficient and less reliant on the real world. The shift from physical to digital document management reduced the need for printing, mailing, and manually filing paper documents. However, this digitization also introduced new forms of tedious work like extracting information from PDFs, inputting information into databases, managing complex digital filing systems, and more. We replaced “paper-pushing” work with “bit-pushing” work.
Historically, unstructured files were hard to mine. Partly because of the lack of semantic context in documents, but also partly because of the ambiguity, subjectiveness and judgement required to “process” the data before inputting it. This is the root cause of a lot of administrative work within companies.
The first wave of RPA turned structured data and well-defined processes into automated workflows. Traditional RPA struggles with tasks that have unstructured data or ambiguous workflows. This is exactly where LLMs “come into their own” (h/t Tom Blomfield). Further, traditional RPA is top-down set by management vs bottoms-up orchestrated by employees.
With LLMs, software has finally reached the place that has been hardest to mine for data: unstructured data. They enable “the fracking of information”. This is a killer analogy from Tomasz Tunguz: “If data is the new oil, then LLMs are the environmentally friendly fracking rigs, blasting value from unstructured text shale formations.”
The more old-school, unstructured-data heavy an industry, the greater unstructured data has been the barrier to technological disruption:
“In pretty much every old, large company, there are huge teams of people running manual processes. They’re hidden away from the end customer (hence ‘back office’ rather than ‘front office’), so we don’t tend to encounter them very often in our day-to-day lives.” — Tom Blomfield, YC RFS
LLMs finally bring these industries into the SaaS and enterprise automation fold:
“That left out the foundational industries that are primarily dependent on unstructured data (e.g. contracts, records, and multimedia files across text, audio, and images). Now, large language models are equipped to handle workflows with unstructured data, meaning AI can be the missing piece that finally brings technologically-underserved industries into the modern era.” — Greylock, Vertical AI
With LLMs, software has finally reached both unstructured data and less-defined workflows:
Customer studies from LLM-powered automation platforms like Luminai, Orby, and Tennr gives insight into how companies are using LLM-powered automation platforms:
Another implication is that software has finally reached the “source of truth” for a lot of system-of-record companies: documents. This opens up opportunities for startups to create the best “objects” in core company datasets like employees, finance, sales by combining structured and unstructured data, and unlocking use-cases that incumbents can’t. This opens up an opportunity for them to displace the dominant system-of-record: HRIS for employee data, CRMs for sales data, ERP systems for financial data.
Of course, administrative work isn’t new. We think timing is right to (actually) end bit-pushing as a result of greater pain-point, behavioral and technological shifts.
Greater need for integrations and system admin; With the explosion of tools adopted by each team, the need to keep data across various systems accurate while maintaining relationships/mappings accurate between different pieces of data (like employees to their roles and teams) creates a lot of administrative work at companies.
Greater complexity in multi-stakeholder workflows: There are more processes in the enterprise like procurement that require approvals from multiple teams within an organization including legal, compliance, security, IT and more, which creates a lot of administrative burden.
Greater security, privacy, and access concerns: With the vast amounts of sensitive data being stored and processed by software tools, the risks associated with data breaches, unauthorized access, and privacy violations are at an all-time high. Adhering to data privacy and security creates a lot of administrative burden for companies.
Most work is now happening in the browser. Automation platforms can now observe work happening in the browser to teach agents how to perform a task.
Expectation of “consumer-grade” experiences in the enterprise. Businesses users are demanding “consumer-grade” applications that meet expectations set by consumer technology companies and the prosumer revolution has served that demand.
LLMs enables automation of the “long tail” of manual processes in ways that were not possible before, either due to unstructured data, ambiguity or subjectiveness. LLMs are well-suited to solve this. LLMs can extract data in non-standard format and be fine-tuned to adapt to specific domains, workflows, or styles. This addresses specific requirements or preferences of different industries or users, enabling the automation of tasks that were previously too specialized. LLMs can be trained on broad datasets enabling them to handle a broad range of tasks that would be too specialized or niche for traditional automation tools.
Granular visibility into events in applications. Automation platforms can make use of the availability of individual process event logs from the underlying application databases (e.g. prospect interactions in CRM, payment receipts in ERP) to gain visibility into processes.
Horizontal platforms
Vertical platforms
We’re excited about startups that are building solutions for the following:
Horizontal
Department
LLM-enabled data extraction and workflow automation platforms for functional areas like HR, finance, compliance, etc. with the ambition of becoming the system-of-record.
Vertical
LLM-enabled vertically-focused data extraction and workflow automation platforms dedicated to real-estate, financial and healthcare, etc.
---
If you’re building a LLM-powered workflow tool with the ambition of building a system of record, please email us at ryan@weekend.fund or vedika@weekend.fund.
