Josh Bersin 谈任务智能：能否把每个岗位拆解成任务交给 AI

There’s a big flurry of interest in Job-Task Analysis (an old discipline) to break down jobs into tasks, categorize and inventory these tasks, and then “Voila,” you can build AI agents to automate these tasks.

It sounds like a great idea. So data providers like Lightcast, Draup, SAP (TechWolf), Reejig, and Workday (using Stanford data) are busy building these huge “Job-Task” data sets so you can now look at your company or department as a “basket of tasks” and apply AI to build just the right agents to streamline work. (And also make decisions like centralizing some tasks or building better learning to cross-train people for such tasks.)

As you’ll hear me describe in the podcast, this kind of work is extremely valuable for regulated, structured, safety-related, and operational jobs. If we know that every airline pilot has to master the task of “lowering landing gear at the right speed and time” it makes perfect sense for the aircraft to do this (or tell the pilot) automatically.

Similarly if every sales person has to update Salesforce in a 2 minute “task” we should build an agent to do it accurately. So there are task automation opportunities in every business area, every function, and every job.

The challenge is that these “tasks” don’t always stand alone: they’re related to broader job roles and business processes. So taking an inventory of tasks is great (we know all the tasks an HR Business Partner does, for example), but we have to think strategically about which of these are suited for AI, which are suited for regular IT automation, which could be outsourced, and which are very “context sensitive” and may need to be treated differently.

How To Do This Well In Today’s AI World

For many years job-task analysis was an industrial project. We did interviews and surveys so we could design, improve, fix, and document a series of linear, industrial, important tasks. Whether it be manufacturing a burrito or a car, each step has a purpose, a right and wrong way, and a set of tools, skills, behaviors, and constraints.

Call center agents, for example, have to know how to handle dozens of inquiries and this demands rigorous, hands-on training and role playing. In a safety or operational job we need simulations and certification (platforms like Kahuna specialize here) or else there could be injury, accident, or loss. I wrote an entire book about this years ago.

Today, while all this type of work continues, we have two changes.

• First we can use AI to find and catalog the tasks, so we may not need so many interviews. For example I asked our AI “digital twin” to explain our process for selling and provisioning a corporate membership. It read all the emails in our team, listened to meetings, and told me the “real tasks” we do.
• Second, we can use AI to automate and redesign these tasks with agents! Once you know what a fairly standard process is (ie. closing the books each month, provisioning a new customer, etc.) you can use our AI agent methodology to “build” an automated solution.

Now two new things are possible: you can analyze work much more quickly so you decide where to focus, and you can then build agents iteratively yourself without hiring an IT team or consultant to help.

We call this Dynamic Work Design, and Galileo can coach you through it. (You can ask Galileo to do it for you.) And now you can redesign the process for speed, accuracy, and scale. Our HR 2030 Multi-Agent design tools will help. This can be a business-driven process focused on outcomes, not just task optimization.

Five Steps

In our interviews with clients we find a hierarchy to this work. First, what is the business goal for this job or work? Second, what are the broader product, service, or customer outcomes we’re trying to achieve? Third, what are the steps or tasks and are they “rigid or mandatory” or are they “judgemental and optional?” Fourth, how do these individual tasks relate or chain together? And Fifth, what human or technical or judgement skills are needed?

In other words, just listing the tasks and knowing the “size” of each is not enough. We need to re-engineer these things in the context of outcomes, customers, important workflows, and the subtlties of “doing the task well.” And when this is done well we can use it for recruiting, training, performance management, and AI agent development.

What we want to prevent is a project to do task analysis for its own sake, because a huge list of tasks may be out of date in a few months. I think that AI, with all its learning capabilities, may make this domain more dynamic and useful than ever before. But don’t think of it as a project in itself, like building a skills taxonomy without a business purpose.

Here is an example of how Workday’s Work Intelligent agent works.

We are doing case studies in this area now – please contact us if you’d like help or share your work intelligence project.

PS: Our new book Superpowered explains this process with many examples.

Additional Information

How To Do Job-Task Analysis

Workday Introduces Work Intelligence to Help Customers Design Work for AI

Lightcast Work Intelligence

Draup Task and Work Intelligence Guideboooks

TechWolf (SAP) Work Intelligence

Reejig Operating System for Work

SAP Signavio: Process Mining for Work Redesign