What Is Task Intelligence?
Start with actual work, such as generating invoices, enriching leads or testing a website. Task intelligence helps your team examine the records, decisions and handoffs, then define the changed responsibilities and output. Use that analysis to scope a simulation around your enterprise. Prepare people for their responsibilities, evaluate agents on representative cases and agree who accepts the connected workflow.
Published source counts reviewed 2026-06-18: approximately 5M job-source tasks, 2,400+ companies and 81 industry segments. The separate workflow catalogue contains 7,113 workflows; the O*NET catalogue covers 894 occupations. These are different datasets, not one study population, a live refresh or customer outcomes. Check the data and source coverage.
Why does task intelligence matter?
A role title does not explain its records, decisions or handoffs.
A job title alone does not specify the work an enterprise needs changed. Financial analysts in different teams may use different records, make different decisions and follow different controls. Frey and Osborne's 2013 research estimated computerisation susceptibility for 702 occupations; it was an occupation-level model, not a task-classification intervention. For an enterprise project, examine the actual tasks, handoffs and exceptions, then test the proposed changes against agreed outcomes.
The figures below describe published research, not Nuvepro customer results or expected gains from an engagement.
high-growth startups in a field experiment on discovering AI use cases
Treated firms received information about how other firms reorganized production around AI. This was not a test of Nuvepro's classification method.
of the US workforce could have at least 10% of tasks affected by LLMs
An estimate of potential exposure, not observed adoption or proof that a workflow is ready to automate. Examine enterprise tasks and constraints before testing changes.
revenue reported for the information-treatment group in the startup experiment
Gains were largest in the upper tail. The study does not establish that Nuvepro's task classification causes revenue gains or predict an enterprise engagement's impact.
The research behind task intelligence
Research context, not promised customer results.
Published studies describe their own populations and conditions. Their findings are not Nuvepro customer results or guaranteed outcomes for your organization. The research below uses occupation-level models, task experiments and firm-level interventions. Read each on its own terms when considering a change to enterprise work.
The Future of Employment
Frey & Osborne
Estimated the probability of computerisation for 702 detailed US occupations and classified about 47% of employment as high risk. This was an occupation-level exposure model, not an observation of tasks automated or jobs lost.
This early occupation-level model provides context for later task-based approaches; it does not validate a particular enterprise workflow redesign.
Read the primary sourceThe task model of production
Acemoglu & Restrepo
Production requires completing tasks. Technology can augment existing tasks, automate them, or create entirely new ones. Each pathway has distinct effects on labor demand and wages.
Task-based economic models examine how technology changes production and labor demand. Nuvepro’s three planning categories are an application choice, not a taxonomy validated by this research.
The Jagged Frontier
Dell’Acqua et al. (Harvard/BCG)
The experiment with 758 BCG consultants found AI assistance helped on selected tasks inside its capability frontier and reduced correctness on a task outside it. The effects depended on the task and conditions.
Nuvepro's interpretation is to examine the actual work and test proposed AI contributions. The experiment did not test task classification as an intervention.
Read the primary sourceProductivity effects of GenAI
Noy & Zhang (MIT)
453 professionals saw 40% time savings and 18% quality improvement on writing tasks. Lower performers gained the most, compressing the productivity distribution.
Evidence of useful AI assistance on the professional writing tasks studied. The experiment does not validate Nuvepro's classification categories or establish readiness for other enterprise work.
Read the primary sourceGPTs are GPTs
Eloundou et al. (OpenAI)
Estimated that around 80% of the US workforce could have at least 10% of tasks affected by LLMs. Higher-income occupations showed greater potential exposure. The paper does not predict an adoption timeline.
This motivates examining actual tasks and enterprise constraints. Exposure estimates are not evidence that a workflow is ready to automate.
Read the primary sourceMapping AI into Production
Hyunjin Kim, Dahyeon Kim & Rembrand Koning (INSEAD/HBS)
In a field experiment with 515 high-growth startups, treated firms received examples of how other firms reorganized production around AI. The working paper reports 44% more discovered use cases and 1.9x revenue; revenue and investment gains were largest in the upper tail.
This supports exploring where AI fits across the work. The treatment was information about other firms, not a test of Nuvepro's task-classification method; the results do not predict an enterprise engagement's impact.
Read the primary sourceThe 30/40/30 pattern
A planning framing from The Agentic Enterprise (2026).
The Agentic Enterprise (2026) uses a 30/40/30 framing for automation candidates, human and AI collaboration, and human-led work, alongside its study of 2,143,500 tasks across six industries, 60 roles and 11 companies. Use the framing as a planning reference, not a measured deployment rate or a target for your workflow.
Consider an invoice workflow: preparing a draft, resolving conflicting delivery evidence and approving issue require different records and authority. Inspect those tasks directly. An occupation-wide percentage cannot establish whether the proposed billing workflow works.
Candidate rule-bound steps, such as validating a required field or transferring approved records. Agree permissions, exception handling and checks before considering operating use.
AI assists with drafts and analysis. People verify the evidence, resolve exceptions and approve decisions, such as invoice issue or accepting a code change.
People lead work such as resolving a sensitive customer dispute or taking responsibility for a safety decision. AI may assist within the permitted scope. The assignment follows the actual capability, policy and operating requirements.
Book reference: "The Agentic Enterprise" (Giridhar, Kashi, Rajan, 2026). Its framing and study population are separate from the dated source corpora above. See the NTI method and worked examples →
How does task intelligence work?
Five steps from examining the work to a project and evidence to review.
Map the selected workflow and its role contributions. Propose the division of work, design the handoffs and build or configure the change. Prepare people for their responsibilities and check what they and the agent produce before the operating owner decides the next step.
Document the workflows
Start with how work actually flows, not how the org chart says it should. Map the end-to-end processes (Procure-to-Pay, Order-to-Cash, Incident Resolution) so every task has a workflow context. Browse the underlying process reference catalogue for source records and coverage labels. Those reference patterns are separate from the published operating examples and practical guides in the enterprise workflow library.
Browse the process reference catalogue →Map role contributions to the workflow
For the selected workflow, record each participant’s inputs, actions, decisions and outputs. Compare approved records, job descriptions and interviews with the people doing the work. There is no required task count. Generated task lists need review, especially for exceptions and undocumented handoffs.
Classify each task
Classify proposed contributions as Automate (a bounded step within agreed rules and authority), Augment (AI assists while people review and decide), or Human-Only (work retained by people for capability, policy or operating reasons). A classification is a design hypothesis, not proof of correctness. Compare current public tools with enterprise-specific data and tools, then test the selected conditions.
Redesign the workflow
Use the analysis to decide what changes in the actual workflow, such as invoice drafting or website testing. Choose an appropriate approach, which may include agents, existing tools or deterministic steps. Define what people build, configure, review and approve, with exceptions and handoffs.
Prepare for the changed work
Connect the changed responsibilities to practical work. Functional and technical teams can build, configure, test and supervise AI at the depth their work requires. Shape projects and practice around the relevant enterprise systems, data and rules, then assess what people demonstrate. Measure operating impact separately after implementation.
See the full 5-step methodology →How is task intelligence different?
Connect work analysis with people, capability and system evidence.
A simulation needs several views of the work. Task analysis examines responsibilities and handoffs; capability evidence concerns what people can demonstrate; system records help inspect what happens in the process. Existing tools may support more than one view. Choose around the decision and check their current capabilities.
| Approach | Unit of Analysis | Core Question |
|---|---|---|
| Task and workflow analysisWork design | A task, decision or handoff | What should change, who does it and how is it checked? |
| Capability assessmentPeople | A person’s demonstrated responsibility | What can people perform, and what preparation is needed? |
| Talent and workforce planningRoles and teams | Role requirements and person evidence | Who should do the work, and what role changes are needed? |
| Process and system evidenceOperating records | An event, record or case | How does the case run, including exceptions? |
Three questions before you automate
For every task classified as automatable, ask these first.
A classification does not decide whether a task should be automated. Review operating impact, the value of human participation and the effect on the people involved. The questions below, drawn from The Agentic Enterprise, help owners examine those trade-offs before authorizing a change.
What is the actual economic value of automating this task?
Include build and integration work, reviewer effort, corrections, recovery and operating support. Compare the complete workflow with an agreed baseline. A wage-based estimate of time value does not establish savings.
What is the strategic cost of removing the human from this task?
Does this task teach people skills they need later? Does it maintain a client relationship? Does it build institutional knowledge? If yes, you may choose not to automate even if it makes financial sense.
What happens to the person when this task is removed?
Agree the changed responsibilities with the affected team. Consider whether people need preparation, different tools or a role transition, and whether the redesign creates new review work. Examine these consequences alongside the proposed automation.
From "The Agentic Enterprise" (Giridhar, Kashi, Rajan, 2026). See the operational framework → | What happens when you skip these questions →
What does task intelligence help you decide?
Workflow design, role preparation and impact hypotheses.
Task Intelligence informs the changed workflow, the work product and responsibilities. Your enterprise context determines the relevant systems, data, rules and exceptions for a bounded project. People build, configure and test that work, with scoped support and assessment. Operating results require your baseline, agreed measures and observation after implementation.
Workflow Reimagined
Examine where people and agents contribute. Propose responsibilities, exceptions and handoffs, then rehearse them through simulations before agreeing operating acceptance.
Workforce Reimagined
Choose a transition to a new role or prepare for AI within an existing role. Agree the responsibilities, then use simulations, GenAI Sandboxes and assessments to practise and assess human capability.
Impact to Test
Identify an impact hypothesis and the baseline needed to test it. Wage-based estimates describe potential time value; they do not establish realized savings, revenue or productivity.
Who uses task intelligence?
Four CXO personas. Four different questions answered.
Which roles need to change and what do my people need to learn?
Use a task map to review changed responsibilities with the people involved. Define what they should build, configure or decide, inspect current capability evidence and scope preparation where needed.
Which workflows actually change when AI gets deployed?
Trace one case through its records, approvals and exceptions. Decide which steps should change, how people and agents hand work over, and what evidence the owner needs before operating use.
What is the real financial impact of AI on the workforce?
Task Intelligence helps form impact hypotheses. Wage data can estimate potential time value, while realized impact requires your baseline, adoption costs and operating measurements.
What did the AI spend actually deliver?
Use the analysis to connect proposed work changes with preparation and agreed measures. Assess human capability separately from agent behavior and workflow acceptance. Confirm business results with operating evidence.
Frequently Asked Questions
Deep dives on Task Intelligence
Frameworks, comparisons and methodology. Read the part that matches what you are deciding.