Gradient Ascent #3
Business process automation, workflow automation, DAGs
Welcome to the 3rd edition of Gradient Ascent. I’m Albert Azout, a prior entrepreneur and current Partner at Cota Capital. On a regular basis I encounter interesting scientific research, startups tackling important and difficult problems, and technologies that wow me. I am curious and passionate about machine learning, advanced computing, distributed systems, and dev/data/ml-ops. In this newsletter, I aim to share what I see, what it means, and why it’s important. I hope you enjoy my ramblings!
Is there a founder I should meet?
Send me a note at email@example.com
Want to connect?
Workflow automation is an overloaded term these days. You can quickly get subsumed in a blackhole of jargon 🙃. After all, much of the work that enterprises do can be reduced to workflows, so the subject covers a lot of ground.
Let’s start with workflows…
Utilizing software to model and execute enterprise processes is not a new concept and was catalyzed by Business Process Management Software (BPMS) in the early 2000s. At a high level, business processes are strategic while workflows are tactical and well-defined. Business processes need not be defined vis-à-vis workflows, yet workflows are always attached to a business process. And many BPMS solutions tend to encapsulate (either explicitly or via lower level capabilities) workflow management and workflow automation capabilities. The lines between the two are blurry.
Historically, BPM co-evolved with the industrial revolutions, as work products were systematized alongside new innovations and enterprises started to think very deeply about process optimization (i.e. Six Sigma). The revolutions proceed as follows: 1.0 was the steam engine and water power for machine-based manufacturing, 2.0 was electrification, 3.0 was information, and now 4.0 is AI-based automation. The first two revolutions generated and converted energy, and the last two (which began with cybernetics) replace energy with information.
Like energy, information can be processed by people, but to do it at an industrial scale, we needed to invent computers, and to do it intelligently, we now use AI [source: Causality for Machine Learning].
Today’s enterprise workflows transform information in a series of steps. Machine learning is a class of informational transforms (e.g. mapping an input to a set of output classes).
As for workflow automation…
Business Process Management evolved over time to focus on Business Process Automation, which is now called Digital Transformation. First we automate by replacing human repetitive work with software and inevitably we replace (aspects of) human cognitive work with AI/machine learning.
Whether you are automating task workflows or data workflows (dataflows)…
task workflows - examples of horizontal platforms include app-builders (AWS Honeycode, KissFlow, Appian), pipeline builders (Pipefy, n8n), and template-based systems (Smartsheet, Airtable). In RPA, a repetitive and discrete human task are candidates for automation (UIPath, Automation Anywhere). Apache Airflow is a well-known dev framework for task workflows.
dataflows - data pipelines for streaming or batch data. Machine learning is typically incorporated in a training phase or an inference phase as part of the pipeline. Apache Beam is a well-known dev framework for dataflows.
…the underlying workflow is typically represented a directed graph (with no circular references, called a DAG)*. For instance, the graph below represents a stream of sensor data:
Using DAG’s, we can reason about how/where ML can be embedded into task workflow to drive automation:
Using ML to decrease workflow latency - increasing the speed of information through a workflow by removing humans and/or collapsing nodes in the workflow
Using ML to increase throughput - via parallelization during workflow execution. A great example is computational drug discovery, where you can search large areas of the chemical space in silico.
Using ML for intelligent routing - by intelligently branching information downstream (e.g. making decisions). For instance, using ML to detect defects in manufacturing and alert operators.
Using ML to sense and understand - transforming physical signals or textual data into structured information and meaning, to be used in downstream tasks (i.e. computer vision, NLP).
Embedding ML into workflows results in many questions and complexities: (1) continuous model learning, autoML, deployment, and lifecycle, (2) model stability, observability (roots in control theory), and explainability, (3) security, safety, and risk (4) next-gen workflow orchestration, etc.
We can expect lots of innovation in these areas over the next few years…
Some interesting startups in and around this space
Building mission driven companies:
Great interview with roboticist Russ Tedrake:
Amazing book by Edward O. Wilson on evolutionary biology and the genesis of societies and language:
* UIPath in built on the Windows Workflow Foundation, which incidentally is a Finite State Machine not a DAG.
While the author of this publication is a Partner with Cota Capital Management, LLC (“Cota Capital”), the views expressed are those of the writer author alone, and do not necessarily reflect the views of Cota Capital or any of its affiliates. Certain information presented herein has been provided by, or obtained from, third party sources. The author strives to be accurate, but neither the author nor Cota Capital do not guarantees the accuracy or completeness of any information.
You should not construe any of the information in this publication as investment advice. Cota Capital and the author are not acting as investment advisers or otherwise making any recommendation to invest in any security. Under no circumstances should this publication be construed as an offer soliciting the purchase or sale of any security or interest in any pooled investment vehicle managed by Cota Capital. This publication is not directed to any investors or potential investors, and does not constitute an offer to sell — or a solicitation of an offer to buy — any securities, and may not be used or relied upon in evaluating the merits of any investment.
The publication may include forward-looking information or predictions about future events, such as technological trends. Such statements are not guarantees of future results and are subject to certain risks, uncertainties and assumptions that are difficult to predict. The information herein will become stale over time. Cota Capital and the author are not obligated to revise or update any statements herein for any reason or to notify you of any such change, revision or update.