Turning 4,000 jobs into a task-level map for AI.
WorkIQ: a task-based workforce-intelligence model that gave a $1B transformation a common, explainable language for where AI can augment work, and where humans should lead.
A $1B question, and no common language
A $1B, four-year transformation hinged on one question: where can AI augment the workforce, and where should people stay in the lead? Answering it the traditional way meant analyzing roughly 4,000 job profiles by hand, thousands of resource-hours of effort.
Even then, leaders had no common language to align a strategy on what AI can and should do. Without a shared, evidence-based way to talk about the work, every function had its own opinion and the transformation risked stalling in debate.
Decompose the work, score it, align on it
Decompose jobs into tasks
Every job is a bundle of tasks. Encoding roles at the task level turns fuzzy debates into a precise, comparable map of the actual work.
Score exposure, explainably
Each task gets an automation-exposure score grounded in O*NET research, so the model is auditable and defensible, not a black box.
A common language for AI
The task taxonomy gave the CEO, CHRO, and transformation teams one shared vocabulary to prioritize where AI helps and where human judgment wins.
Systems thinking, at speed
Built as a decision-grade model in six weeks, designed to do about 80% of the job-analysis work and to improve recursively from use.
From jobs to tasks to a workforce strategy
1Encode jobs as tasks
Each job profile is broken into its component tasks, creating a precise, comparable unit of work across the enterprise and a common language for what people actually do.
2Score exposure to automation
Every task is scored 0–100 for exposure to automation, grounded in O*NET work-context research, so scores are consistent, explainable, and defensible rather than a black box.
3Roll up to roles and org
Task scores roll up to roles, teams, and business units, revealing where work is human-centered and where it is highly exposed to automation.
4Model supply and demand
Task loads are combined with headcount trends to forecast supply and demand by skill and to simulate human-plus-AI workforce scenarios.
5Prioritize the plays
The model flags where to transform, develop, or redeploy. For example, roles that are both growing and highly exposed to automation are prime transformation targets.
6Deliver securely, improve recursively
Delivered in Tableau with row-level security so each leader sees only their organization, and designed to capture feedback and improve recursively from use.
What can AI do — task by task?
The starting point for job analysis: pick a role to see its tasks decomposed and scored for exposure to automation, from human-centered judgment to routine, automatable work. An interactive reconstruction using public O*NET tasks.
Where is demand rising, and how exposed is it?
The resourcing view: every role plotted by demand (headcount change) and exposure to automation. The top-right is where growth meets automation, the prime place to transform. Filter, and click a role for its recommended play.
Decision-grade in six weeks, a year off the timeline
Under my leadership, a workable, decision-grade model was delivered in six weeks. The CEO and CHRO judged this component alone worth $2M+ had a firm like Deloitte built it. Together with the tools and the capability to use them, the Head of Strategy, CHRO, and CEO estimated the effort accelerated the four-year, $1B transformation by roughly 12 months, saving thousands of hours and creating a real strategic advantage. On the resourcing side, redeployment over severance was modeled to save $3–5M a year.
Whether you’re modernizing people analytics, deploying AI responsibly at scale, or sharpening product and market strategy, I’d be glad to talk.
The dashboards and figures here are an illustrative reconstruction using synthetic headcount data and public O*NET task statements and automation indicators. This work was completed while employed at a Fortune 25 health insurer; the internal tool and its production data remain the property of the company.