# AI Augmentation and Human Judgment Across 199 U.S. Professional Occupations

> O*NET 31.0 evidence across 199 occupations maps information-analysis, judgment-governance and coordination-translation activities into the ACT-3 role-redesign model.

- Canonical page: https://mtfinstitute.com/insights/ai-augmentation-work-activities-199-occupations-onet-31/
- Content type: Article
- Editorial category: Research & Reports
- Publisher: MTF Institute of Management, Technology and Finance
- Author: MTF Institute Editorial Team- Published: 2026-09-21
- Updated: 2026-09-21
- Language: English
- Topics: O*NET, AI Augmentation, Human Judgment, Occupational Research, Job Redesign

**Research Report MTF-RR-2026-09-21-01**

## Abstract

Which professional work activities are plausible candidates for AI assistance, and where should organizations preserve explicit human ownership? MTF Institute analysed complete O*NET 31.0 Importance ratings for twelve substantive work activities across 199 detailed occupations in Management, Business and Financial Operations, Computer and Mathematical, Architecture and Engineering, and Legal occupational groups.

The result is not an automation forecast. It is a work-allocation lens. Getting Information was important at level 4 or higher in 97.5% of occupations; Working with Computers in 89.4%; Making Decisions and Solving Problems in 86.9%; and internal communication in 81.9%. Management differed from the other groups: 23 of 55 management occupations met the report's human-context-dominant convention, compared with four of 144 occupations across the other groups. The evidence suggests that information handling may be widely assisted while judgment, coordination and accountability must be designed explicitly.

The report introduces ACT-3: split a role into information-analysis, judgment-governance and coordination-translation activity bundles; decide which tasks may be assisted, which require human review and which must remain human-owned; then monitor both output quality and decision consequences.

## The research question

Across 199 detailed U.S. professional occupations, how important are twelve work activities related to information analysis, judgment and governance, and coordination and translation? How do the patterns differ by occupational group, and how can leaders use them to allocate work between people and AI without pretending that an occupational rating predicts automation?

This question matters because AI discussions often jump from tool capability to job replacement. A role is a bundle of activities. The same occupation can contain data retrieval that is easy to accelerate, interpretation that requires context and consequential decisions that require accountable authority. A useful design therefore begins below the job-title level.

## Headline findings

1. Information acquisition and computer use are nearly universal in the selected professional groups. Getting Information had a median Importance rating of 4.43 and was at least 4.0 in 97.5% of occupations. Working with Computers also had a median of 4.43 and crossed 4.0 in 89.4%.
2. High information intensity did not remove the need for judgment. Making Decisions and Solving Problems had a median of 4.33 and crossed 4.0 in 86.9% of occupations.
3. Communication is part of the analytical system. Communicating with Supervisors, Peers, or Subordinates had a median of 4.25 and crossed 4.0 in 81.9%.
4. Management occupations combined stronger human-context activity. Their median coordination-translation bundle was 4.008 and judgment-governance median was 3.900, compared with coordination medians from 3.308 to 3.580 in the other four groups.
5. The ACT-3 descriptive convention classified 136 occupations as analysis-support dominant, 36 as blended augmentation and 27 as human-context dominant. These labels describe relative activity patterns, not technical feasibility or job-loss probability.

## Data source

The source is the [O*NET 31.0 Database](https://www.onetcenter.org/database.html), August 2026 release, published by the U.S. Department of Labor's O*NET program. O*NET provides occupation-level ratings for work activities. The database is available under Creative Commons Attribution 4.0.

MTF Institute used two source tables:

- Occupation Data, for detailed O*NET-SOC codes and titles;
- Work Activities, for occupation, activity, scale and value.

The archived package contains the exact source release, transformation code, occupation-level results, activity summary, group summary, cross-tabulation and method note.

## Population and inclusion rules

The population is every detailed occupation in five O*NET-SOC major groups:

| Major group | Valid occupations |
|---|---:|
| Management | 55 |
| Business and Financial Operations | 46 |
| Computer and Mathematical | 36 |
| Architecture and Engineering | 55 |
| Legal | 7 |
| **Total** | **199** |

An occupation was valid when it had a non-suppressed numeric Importance value for all twelve selected activities. All 199 occupations in the chosen groups met that rule. Each occupation received equal weight. The design is a census of selected O*NET groups, not a sample of workers or vacancies.

## The twelve coded work activities

The activities were selected before calculating results and organized into three practical bundles.

### Information-analysis

- Getting Information;
- Processing Information;
- Analyzing Data or Information;
- Working with Computers.

### Judgment-governance

- Evaluating Information to Determine Compliance with Standards;
- Making Decisions and Solving Problems;
- Developing Objectives and Strategies;
- Monitoring and Controlling Resources.

### Coordination-translation

- Interpreting the Meaning of Information for Others;
- Communicating with Supervisors, Peers, or Subordinates;
- Coordinating the Work and Activities of Others;
- Developing and Building Teams.

These are substantive activities, not page labels or keyword counts. They describe work that can inform role design. The bundles are MTF analytical constructions, not official O*NET categories.

## Calculation method

For each occupation, the analysis calculated the arithmetic mean Importance value for each four-activity bundle. It then calculated a human-context index as the mean of the judgment-governance and coordination-translation bundle scores.

The descriptive allocation convention was:

- **Human-context dominant:** human-context index at least 4.0;
- **Analysis-support dominant:** information-analysis at least 0.35 points higher than the human-context index;
- **Blended augmentation:** all other complete occupations.

The 0.35 difference and 4.0 threshold are transparent MTF conventions. They are not O*NET standards and do not define whether an occupation can be automated. They make the cross-occupation comparison reproducible.

The report also calculated medians, arithmetic means and the share of occupations with an Importance value of at least 4.0 for each activity. No employment weighting was applied.

## Overall activity results

| Work activity | Median importance | Mean | Share at least 4.0 |
|---|---:|---:|---:|
| Getting Information | 4.43 | 4.414 | 97.5% |
| Working with Computers | 4.43 | 4.395 | 89.4% |
| Making Decisions and Solving Problems | 4.33 | 4.314 | 86.9% |
| Communicating internally | 4.25 | 4.219 | 81.9% |
| Processing Information | 4.05 | 4.035 | 58.3% |
| Analyzing Data or Information | 4.04 | 4.020 | 53.3% |
| Evaluating Information for Compliance | 3.94 | 3.910 | 48.7% |
| Interpreting Information for Others | 3.69 | 3.685 | 19.6% |
| Developing Objectives and Strategies | 3.46 | 3.454 | 11.1% |
| Developing and Building Teams | 3.35 | 3.338 | 14.1% |
| Coordinating Others' Work | 3.24 | 3.303 | 15.1% |
| Monitoring and Controlling Resources | 2.83 | 2.849 | 3.5% |

The most important result is the combination. Information-heavy activity and decision activity are both widely important. Organizations should not infer that because retrieval or synthesis can be accelerated, the surrounding decision has also been automated.

## Group comparison

| Occupational group | Information-analysis median | Judgment-governance median | Coordination-translation median | Human-context dominant | Blended | Analysis-support dominant |
|---|---:|---:|---:|---:|---:|---:|
| Management | 4.093 | 3.900 | 4.008 | 23 | 21 | 11 |
| Business and Financial Operations | 4.286 | 3.561 | 3.580 | 2 | 8 | 36 |
| Computer and Mathematical | 4.440 | 3.490 | 3.464 | 2 | 1 | 33 |
| Architecture and Engineering | 4.218 | 3.570 | 3.570 | 0 | 5 | 50 |
| Legal | 4.110 | 3.315 | 3.308 | 0 | 1 | 6 |

The Legal group contains only seven detailed occupations, so its median should not be treated as a broad statement about all legal work. The group comparison is descriptive.

Management is the clearest exception to a simple “information work equals automation” story. Its information-analysis median remains high, but coordination and judgment are much closer. A management workflow may benefit from faster analysis while becoming more dependent on explicit decision rights, escalation, communication and team consequences.

## Occupation examples

The highest information-analysis scores included Statisticians (4.798), Bioinformatics Technicians (4.760), Fraud Examiners, Investigators and Analysts (4.700), Actuaries (4.680), Data Scientists (4.655), Operations Research Analysts (4.655) and Business Intelligence Analysts (4.638).

These roles are natural candidates for strong analytical assistance. That does not mean the full role is replaceable. A fraud examiner may use automated anomaly detection yet still evaluate evidence, document a case and operate within legal controls. A data scientist may accelerate code and documentation yet still define constructs, validate models and explain limitations.

The highest human-context index included Chief Executives (4.456), Information Technology Project Managers (4.444), Natural Sciences Managers (4.284), Chief Sustainability Officers (4.274), Medical and Health Services Managers (4.271) and Human Resources Managers (4.221). Management Analysts also met the human-context convention at 4.079.

These examples show why job title alone is insufficient. An IT project manager sits in the Computer and Mathematical group but has a very high human-context pattern. A reliable allocation must inspect the activities, decision consequences and interfaces of the local role.

## ACT-3: a role-redesign model

### A — Analyse the activity, not the title

List recurring tasks and map each to one or more bundles. “Prepare weekly revenue forecast” contains information extraction, definition control, scenario analysis, manager interpretation and resource decisions. Do not classify the entire task as automated or manual.

For each task record:

- input and data rights;
- output and user;
- frequency and deadline;
- consequence of error;
- current controls;
- O*NET-like activity characteristics;
- accountable owner.

### C — Choose the human-AI configuration

Use four configurations:

1. **Human performed:** the person completes the task because consequence, confidentiality or contextual judgment outweighs assistance value.
2. **AI assisted, human executed:** the system proposes; the person checks, edits and performs the action.
3. **AI executed, human approved:** the system completes a bounded task; the person verifies defined evidence before release.
4. **Automated with monitored exception:** the system executes a low-consequence, stable process and routes exceptions to an owner.

Configuration four requires stable definitions, tested controls, monitoring and a safe fallback. It is not the default for professional judgment.

### T — Test output and consequence

Measure more than speed. Test:

- factual and calculation accuracy;
- completeness and consistency;
- false-positive and false-negative consequences;
- sensitivity to changed inputs;
- privacy and security compliance;
- user understanding;
- escalation quality;
- decision outcome and unintended effects;
- correction and incident rate.

A faster report that changes the wrong decision is not a productivity gain.

## Worked redesign: management analyst

O*NET places Management Analysts in Business and Financial Operations. In this dataset the occupation had an information-analysis score of 4.385 and human-context index of 4.079, meeting the report's human-context-dominant convention.

Consider a cost-reduction diagnostic.

| Work component | Bundle | Recommended configuration | Control |
|---|---|---|---|
| collect approved process documents | information-analysis | AI executed, human approved | source inventory and access rules |
| summarize recurring cost drivers | information-analysis | AI assisted, human executed | reconcile to ledger and sample source passages |
| identify affected teams and constraints | coordination-translation | human performed with AI note support | stakeholder validation |
| generate scenario calculations | information + judgment | AI assisted, human executed | formula review and sensitivity analysis |
| recommend option and trade-offs | judgment-governance | human performed | accountable decision memo |
| communicate implementation impact | coordination-translation | human performed | owner, feedback and escalation record |
| monitor approved savings | information-analysis | automated with monitored exception | metric contract and variance threshold |

The redesign accelerates document handling and scenario construction while preserving human authority over trade-offs, organizational impact and recommendation. It also makes validation visible.

## Implications for analyst development

The overall medians suggest four durable capabilities.

First, analysts need strong information systems: data retrieval, processing, analysis and computer fluency. Second, they need decision reasoning because problem solving is widely important. Third, they need communication because analytical evidence must travel through organizations. Fourth, they need governance because compliance, definitions and resource consequences cannot be inferred from generated output alone.

A portfolio should therefore show:

- a clear decision and metric contract;
- reproducible data preparation;
- baseline and sensitivity analysis;
- human review of AI-assisted work;
- an executive recommendation;
- monitoring and change control.

## Implications for managers

Managers should resist two symmetrical errors. The first is treating AI as irrelevant because leadership is interpersonal. Management occupations still had an information-analysis median above 4.0; assistance can improve preparation and visibility. The second error is treating faster information as a substitute for authority and coordination. The management group's stronger human-context pattern shows why decision rights and communication remain central.

Before scaling an agent, ask:

- Which activity is being assisted?
- Who owns the decision?
- What source may the system use?
- Which test must pass before release?
- Who can override the output?
- What harm or trade-off is monitored?
- When does the workflow expire or require reapproval?

## Limitations

This study has important boundaries.

- O*NET ratings describe occupations in the United States and may not transfer directly to another country, employer or local role.
- Importance is not time spent, automation feasibility, employment volume, wage value or performance impact.
- The data combine source dates across occupations and are not a real-time vacancy sample.
- The twelve activities and three bundles were selected by MTF Institute. Other defensible selections may produce different patterns.
- The allocation thresholds are descriptive conventions, not validated predictors.
- Equal occupation weighting gives a rare occupation the same influence as a large one.
- The report does not observe actual AI adoption or outcomes.
- Group comparisons do not control for industry, seniority, regulation or organizational design.

The findings should guide task-level inquiry, not justify layoffs, hiring guarantees or claims that a profession will disappear.

## Reproducibility and archival package

The research package contains:

- searchable PDF report;
- `occupation_results.csv` with 199 valid occupations;
- `activity_summary.csv`;
- `group_summary.csv`;
- `archetype_crosstab.csv`;
- `analysis.json`;
- `analyze_work_activities.py`;
- method note and README;
- exact O*NET 31.0 source archive identification.

The script filters the five major groups, selects non-suppressed Importance values, checks completeness, calculates bundle scores and writes all results deterministically.

The complete public research package, including the searchable PDF and frozen source archive, is preserved in [Zenodo record 10.5281/zenodo.22866337](https://doi.org/10.5281/zenodo.22866337).

## Authorship and review

**Author:** MTF Institute Editorial Team.  
**Method and analytical review:** MTF Institute Research Review.  
**Publication date:** 21 September 2026.  
**Report number:** MTF-RR-2026-09-21-01.

Review covered source attribution, completeness, calculations, threshold disclosure, interpretation boundaries and reproducibility. O*NET and the U.S. Department of Labor have not approved or endorsed this analysis.

## Practical next step

Select one recurring professional workflow. Break it into activities, apply ACT-3 and pilot the narrowest low-consequence assistance configuration. Measure accuracy, correction, cycle time and decision consequence before expanding. Preserve a human owner even when execution becomes automated.

Professionals who want structured practice in data preparation, analysis, communication and responsible use can review MTF Institute's [Professional Certificate in Data Analysis](https://mtfinstitute.com/programs/professional-certificate-data-analysis/#enroll). Evaluate the current curriculum against the evidence you need to produce; a professional certificate does not by itself determine employment or automation readiness.

## Conclusion

The 199-occupation census shows that information work, computer use, decision making and communication are all important across professional occupations. The practical conclusion is not that AI will replace the roles. It is that information-heavy components offer broad assistance opportunities while judgment, governance and coordination require explicit design.

ACT-3 makes that design testable. Analyse the activity, choose the human-AI configuration and test output plus consequence. Organizations that follow those steps can pursue faster analytical work without confusing generation with accountability.



## Citation

When citing or summarizing this material, link to the canonical HTML page: https://mtfinstitute.com/insights/ai-augmentation-work-activities-199-occupations-onet-31/