Embedding AI into Operational Workforce Planning: A Ten-Step Framework

Headcount planning today is largely bottom-up. Business units submit FTE requirements based on their strategy, skills and business needs are assessed with varying levels of consistency, the existing workforce is reviewed, and judgment sets the final number. The impact of AI, whether it automates work or augments the people doing it, is not systematically built into those decisions.

Our latest Draup | Etter research paper, Embedding AI into Operational Workforce Planning, sets out a ten-step framework that embeds AI into workforce and budget planning. It also sets out the key questions every regional and functional leader should answer when they submit a workforce plan.

My central message is simple: the future of workforce planning is not just about determining how many people we need, but understanding how human allocation, AI and redesigned workflows can collectively deliver business outcomes.

Starting with Responsible AI guidelines and the job family architecture

The first step is the precondition for the other nine: clear Responsible AI guidelines. Without them, nothing works. The guidelines decide which work AI is allowed to touch and under what controls. That means defining permitted, restricted and prohibited uses of AI, setting where human oversight is mandatory and, where EU employees or candidates are in scope, aligning with the EU AI Act. Every later step is filtered through these guidelines.

The second step starts from the current job families, roles and levels. Planners identify strengths and gaps in the architecture, flag where it is not flexible enough to add evolving roles, and fix those structural gaps before new roles are introduced.

Breaking jobs into atomic and judgmental tasks

AI acts on tasks rather than jobs, so the third step decomposes each job. Atomic tasks are discrete, repeatable and rules-based, which makes them candidates for automation. Judgmental tasks need context and decision-making, which makes them candidates for augmentation. The task map draws on job descriptions, actual task mapping, surveys and value stream maps, and is then validated with the business.

The first two action items in the framework build on this split: identify which tasks can be automated and quantify the capacity this frees up, and where AI augments people rather than replacing work, use that to plan reskilling and upskilling.

Checking the tech stack and AI economics

What AI can do in practice depends on what is in place, so the fourth step reviews the foundation models in use or approved, the cloud providers, and the legacy systems and how easily they integrate.

The fifth step tests the economics, on the principle that machines should not cost more than the work they take on. Planners estimate expected token spend by job family or job role, compare it against the cost of the human effort it replaces or augments, and count only the use cases where the economics hold.

Redesigning processes, reallocating people and implementing new workflows

The sixth step is process redesign, and its rule is to avoid automating broken processes. Each function, such as customer service, is treated as a set of processes and workflows. The process is fixed first, AI is applied second, and quick wins are prioritized.

The seventh step is resource allocation: deciding which employees need to move, and move faster, to the new way of working. It requires understanding which capabilities the organization is betting on. Consuming the strategic plan is crucial because it tells you where people should go.

The eighth step deals with the tensions that new ways of working create across functions. An organization may identify a need for a GenAI Platform Engineer, for example, while Compensation cannot yet support the role. Without processes that resolve these tensions quickly, new workflows stall.

Building a skills catalog and tracking AI-native and startup peers

The ninth step is a catalog of the skills the organization needs to build, with each skill linked to the associated courses. The tenth assembles a peer set that includes AI-native companies and startups that may disrupt the plan, and tracks the skills they have, so the organization can plan proactively rather than reactively.

Ten questions to answer with every headcount plan

Each step comes with one question that every regional and functional leader should answer along with their FTE submission. For the tech stack, the question is which models, cloud services and legacy systems limit what we can deploy. For AI economics, it is what our expected token spend per job family is, and whether it is below the cost of the work. For resource allocation, it is which people need to move, how fast, and to which capabilities we are betting on.

The actions companies are taking to put the framework into practice

Companies putting the process into practice are taking ten concrete actions. They include deciding which foundation models are off-limits, whether for regulatory, security or commercial reasons, and standing up a small forward-deployed engineering team to test how reliably agents perform in real workflows before scaling them. They also include modeling token costs at the level of individual tasks and workflows, not only in aggregate, and building a three-year view of which skills are likely to be disrupted.

Draup can facilitate a working session with your regional and functional leaders to run these ten steps against your current FTE plan.

To get the full report, subscribe to the newsletter and our representatives will reach out to you.