Companies Automate Tasks - Not Workflows

Companies are now beginning the massive migration of deploying entire work processes to AI Agents that they have curated to their needs. Unfortunately, what is happen in real-time is that the companies and the firms are automating tasks. Tasks and the AI have a one to one relationship. An example would include:

• Compare these documents and provide suggestions to improve the contractual terms for our business.
• Check this pleading against the rules of the court, and ensure that we are in compliance with the statutes that have bearing on this matter.
• Send me a notification when a change occurs to this document from my client.

But these are merely tasks. They require the user to baby sit the computer to ensure that the task is done, done correctly, done with the right legal contextual understanding, etc. In this situation, AI is not moving more work through a flow, it is improving or speeding up a task that the worker has too many to do.

AI Founders on Youtube provide a clear process for how to not just speed up tasks but how to more substantially improve your business by automating the flow of work.

Step 1: You must fully map the workflow (document) the process.

This step is incredibly hard. Most fail at this. The reason companies and firms fail at this, is there are several steps that are not documented to start with, but the key knowledge worker is able to "think" the process out. For example, if the process is to get a contract signed. What exact steps are to be taken? Now, before you call it done, take each step in the map, and ask the following: What happens if this step doesn't meet the criteria for the next step, it fails in some small way? What happens then? AI Automation will not take this into account. Let me be more concrete. What if the Address, Phone Number, and Tax ID are incorrect on the contract. This can happen when the initialization of the AI Automation was kicked off by using the customer for the contract, named Bayer. But in this case, it wasn't Bayer, it was Bear. When this happens, a human knows that the contract information was from the wrong division, or practice area. So, additional documentation of the process is important, so that Automation Verification elements can be built into the new automated process. Even a single missed step in the documentation process can push errors through the system with speed, thus rendering the automation itself the problem.

Step 2: Remove any and all ambiguity from the process.

Lets be clear about the meaning of this step. Humans are able to make assessments about specific situations, and then based on historical track records, or reading a room, people can make a decision. Automated tasks run by an Artificial Intelligence, cannot make a decision, without all the context. All context when writing a process must be articulated into a matrix of options that have clear instructions. Let me provide you with an example. Using the Contract example above, where the Client was named Bear, not Bayer. A layer of verification by the AI could have prevented that error. But, what is not known by the AI is that Bear requires that every contract be reviewed by the Vice President of Sales upon completion (no other client contract needs this) because the Vice President will often times make a change to the fast pay discount based on the quantity ordered. This particular "special" circumstance must be documented, as does every possible "cul-de-sac" or "waiting room" that every workflow navigates as it journey's from start to finish. Failing at this part of your documentation will require you to rewrite, and rewrite your process until every possible iteration of the workflow has been documented.

Step 3: Once core documentation has been written, and ambiguous processes are folded in to the documented process, then you can Automate.

If you would like to better understand the entire process, AI Founders, a Youtube Channel has made a fantastic video on this. Click here to watch the 10 Minute video.