Knowledge work management - where can AI help?

Perhaps Sam Altman is right.? Perhaps agentic AI algorithms will someday perform the vast majority of the knowledge work required to run a modern corporation.? For the foreseeable future however, humans will continue to make critical business decisions concerning product design, manufacturing, sales and distribution as well as the investment decisions in spending and staffing that make those things happen.? Furthermore, corporate decision makers will continue to rely on the skills, experience and intuition of their non-agentic employees to ensure that their decisions are properly executed.?

?Agentic AI will undoubtedly improve knowledge worker productivity through the wholesale automation of ?many traditional activities.? But automation only addresses one aspect of on-the-job performance.? The other major challenge is efficient and effective time management.? Simply put: what is the optimum way of focusing the time and attention of highly skilled/highly paid knowledge workers on the spectrum of tasks they are expected to perform?

?There’s a curious contrast between the scientific ways in which corporations go about optimizing the return on their investments in manufacturing facilities, supply chain inventories, transportation & logistics, etc. versus the largely ad hoc ways in which they manage returns on their investments in human capital.? Most knowledge workers are exempt employees who are given broadly stated work assignments and left largely to their own devices in determining how to allocate time during their workday or workweek.? Agentic AI may deliver significant on-the-job time savings through automation but it has done little to date to assist knowledge workers in using such savings wisely.

The anatomy of knowledge work

?Knowledge work is messy.? Work assignments are commonly described in terms of desired business outcomes or deliverable artifacts.? Some assignments may be an inherent part of one’s job such as the quarterly close exercise that occurs every ninety days within finance teams.? Others may be performed infrequently or on an entirely ad hoc basis.? Many have notional timelines or resource constraints that are subject to renegotiation during the lifetime of an assignment.? Similar assignments performed in the past may merely serve as points of reference and fail to provide prescriptive templates regarding task planning and execution.? In fact, a knowledge worker may be explicitly instructed to bring some innovative “fresh thinking” to an assignment and not simply repeat past practices.? Finally, many assignments require assistance or coordination from co-workers over which an individual has little or no control.? In short, knowledge work is messy.?

?Individual work assignments are established through a waterfall process in which corporate strategies and initiatives are distilled into activities or projects that can be assigned to functional departments.? These departments, in turn, distill their assigned responsibilities into work packages that can be allocated to internal disciplinary teams.? Finally, these teams translate work packages into work assignments that can be allocated to individual team members.? ?

How have knowledge workers traditionally optimized their on-the-job productivity?

?Knowledge workers typically translate their assignments into a set of component tasks that are collectively required to produce a set of desired results.? Key steps in task planning and execution are displayed in Figure 1 and will likely be self-evident to any knowledge worker.?

Task Definition – What Needs to Get Done?

Assignments are deconstructed into one or more discrete work tasks.? Each task should have explicit completion criteria and may be assigned an explicit due date.? Some tasks may be performed serially, others may be performed in parallel.? Some may depend upon the availability or work products of others.?

?In practice, knowledge workers are generally quite adept at “chunking” work assignments into discrete tasks.? In many instances there’s an inherent and obvious logic involved in defining the discrete activities required to complete an assignment.? In other cases there may be constraints on the availability of key co-workers or specialized facilities that influence task definition.?

?Task Time Estimation – How Long Will It Take?

Time estimation is largely based on past experience.? If a task bears a strong resemblance to similar tasks performed in the past, the time to complete can be estimated with a fair degree of accuracy.? If the task is unfamiliar or infrequently performed, intuitive estimates of time duration are approximate at best.

?Task Scheduling – When Should It Happen?

In principle, scheduling should be based upon business and organizational priorities that have been clearly defined through the waterfall process described above.? In practice, task scheduling is far more complicated.? Accurate scheduling requires a realistic appraisal of the time that is actually available for work, as opposed to time allocated to other activities such as regularly scheduled meetings, training, holidays, etc.? Scheduling may also depend upon the availability of co-workers or the completion of tasks to which they’ve been assigned.?

?A universal best practice in task scheduling is to minimize the number of tasks an individual is expected to work on concurrently (i.e. minimize “work in progress”).? Multiple studies have demonstrated that human productivity is easily undermined by excessive multitasking.

Task Execution/Completion

Execution is complete when a task’s exit criteria have been achieved.?

Where do knowledge work management practices commonly break down?

?The foregoing description of knowledge work practices is far too clinical.? In practice, each step of the process depicted in Figure 1 is subject to error.? Each step has strong dependencies on the preceding steps.? Errors propagate down this chain of events resulting in situations where the initial objectives of a work assignment may need to be modified.? It’s not uncommon for success to be declared based on what has been accomplished to date, not on what was supposed to be accomplished when the assignment was initially defined.

?Common problems encountered in knowledge work practices are as follows.

?Overly Optimistic Time Estimation.? This can be particularly acute if a worker has limited knowledge or prior experience in performing the work required to complete a task.?

?Additional Work Discovery.? Workers may encounter new work that must be performed before a task can be successfully completed.? ??

?Dependencies on Others Introduce Unanticipated Time Delays.? Schedules may be disrupted when co-workers fail to make themselves available at required times or fail to deliver the work products needed to complete a task.? Tasks may also be delayed due to constraints on the availability of specialized facilities or infrastructure.

?Excessive Over Scheduling.? Individuals frequently make time commitments without truly accounting for their prior work commitments, personal work habits or true on-the-job availability.

?Incessant Workplace Distractions.? The insistent demands that email, text messages, application notifications, infrastructure alerts, etc. place on knowledge worker attention can easily compromise an individual’s ability to sustain focus on the task at hand.

?Unplanned external interruptions may also disrupt on-the-job productivity.? Corporate priorities may be redefined based upon competitive pressures or financial issues.? Acquisitions of new lines of business or the sale of existing assets may also disrupt work plans.? However, productivity losses stemming from these types of decisions are obviously not the fault of individual workers.

Can AI help improve knowledge work management?

?In principle, AI technology may deliver substantial benefits to knowledge work management in addition to its ability to eliminate certain types of knowledge work altogether.? Task mining tools commonly employed by robotic process automation platforms collect vast amounts of information concerning an individual’s use of business applications, desktop applications, collaboration tools, social media channels, etc.? In theory, AI algorithms trained on such data should be able to determine or at least advise a knowledge worker how to best define, estimate and schedule specific work tasks.

?With the use of AI, work can be managed on a much more personal basis, taking into account an individual’s efficiency in performing similar tasks in the past; the time of day and work locations that are most conducive to performing certain types of work; the number of tasks an individual can reasonably be expected to work on concurrently; and the potential risks of additional work discovery based upon personal past history and the past experiences of co-workers.? By comparison, the current use of AI to summarize meeting discussions, generate emails and coordinate schedules seems absolutely primitive!

?The advances in knowledge work management envisioned here come at a price.? Individuals will need to expose their most intimate work habits to AI systems if they are to achieve the productivity gains referenced above.? Many will understandably consider this to be an intolerable invasion of privacy.?

?This next generation of productivity gains – after the agentic automation wave has crested – may only be achieved when AI models can be hosted on individual devices and the personal data they require is retained on those devices as well.? If knowledge workers can be assured that their employers are not microscopically monitoring their on-the-job behaviors, they may be willing to participate in this second revolution in knowledge work productivity.?

Originally posted on the Forbes CIO Network

?