Your workflows. A fresh AI assessment.
AI Automation Opportunity by Occupation
Find workflows worth testing with AI. Choose up to three tasks, add your working context, and get a practical shortlist: an approach, tools, checks, human review, and a first pilot.
- occupation profiles
- 923
- original task statements
- 19,265
- matched BLS occupations
- 772
AI proposes a workflow; a pilot establishes whether it works. Historical O*NET 27.2 research and BLS May 2023 wages remain available as supporting context.
01 / Choose your occupation
Loan Officers
O*NET 13-2072.00 · Business and financial operations · 30 tasks
Choose work your team actually performs. Generate a new AI assessment of possible approaches, required tools, output checks and a pilot to validate the idea.
Historical occupation research (optional)
These GPT-4-era exposure labels and BLS May 2023 figures are background research. They are not inputs to the fresh AI assessment or evidence of current model performance.
- Human exposure index · β
- 64.8%
- Median annual wage
- $69,990
- National employment estimate
- 321,090
- National wage-exposure index
- $17.6B
Higher than 95.2% of 923 research profiles.
BLS May 2023 · USD per year
Mean hourly: $40.62
Wage and salary positions · May 2023
Employment × mean annual wage × human β. A modeled index, not savings.
Wages and employment describe Loan Officers (13-2072). These estimates exclude self-employment. View BLS source.
Direct LLM index · α34.1%
LLM + software index · γ95.5%
GPT-4 annotation comparison · β59.1%
Research / Historical task labels
A task, not a whole job.
Original O*NET statements with historical human exposure labels. Candidate tasks still need testing on your data, with review effort included.
11 tasks
Direct LLM assistance
17 tasks
AI with tools or code execution
2 tasks
Below research threshold
- AI with tools or code executionTask 3408 · Core · Weight 2
Meet with applicants to obtain information for loan applications and to answer questions about the process.
- AI with tools or code executionTask 3409 · Core · Weight 2
Analyze applicants' financial status, credit, and property evaluations to determine feasibility of granting loans.
- AI with tools or code executionTask 3407 · Core · Weight 2
Approve loans within specified limits, and refer loan applications outside those limits to management for approval.
- Direct LLM assistanceTask 3410 · Core · Weight 2
Explain to customers the different types of loans and credit options that are available, as well as the terms of those services.
- Direct LLM assistanceTask 3416 · Core · Weight 2
Submit applications to credit analysts for verification and recommendation.
GPT-4 annotation differs: ai with tools or code execution. Annotation disagreement is not a confidence interval.
- AI with tools or code executionTask 3413 · Core · Weight 2
Review loan agreements to ensure that they are complete and accurate according to policy.
GPT-4 annotation differs: direct llm assistance. Annotation disagreement is not a confidence interval.
- AI with tools or code executionTask 3412 · Core · Weight 2
Review and update credit and loan files.
- AI with tools or code executionTask 3411 · Core · Weight 2
Obtain and compile copies of loan applicants' credit histories, corporate financial statements, and other financial information.
Showing 8 of 30 total tasks. Every task contributes to the index.
The linked page may have newer tasks than this O*NET 27.2 research snapshot.
From a proposal to evidence
Find the right starting point for your loan officers team.
Bring a proposed workflow to a scoped pilot. We can test representative examples, compare model and tool options, measure review effort, and estimate costs including integration.
Discuss a workflow pilotYour selected role carries through to the consulting inquiry form.
What the output means
Proposals first. Evidence from a pilot.
Fresh AI analysis: when you choose tasks and generate a shortlist, the server sends their original descriptions and your context to Google’s Gemini API. The response proposes an approach, tools and access, checks, human review and a small pilot. Its model and generation time are shown. Historical exposure scores and wages are not sent as capability evidence.
A proposal needs testing: this tool does not run Python, inspect your systems or validate a workflow on your data. AI can propose a script and a code execution environment, but an actual pilot must test correctness, review effort and operational fit. An unavailable or invalid AI response produces an explicit error.
Capacity uses your assumptions: weekly net hours = runs × (minutes per run × assumed gross reduction − review minutes per run) ÷ 60. The editable 10–30% starting range is an illustration, not a model prediction. Negative results stay visible. Optional annual capacity value uses 52 weeks and excludes implementation, runtime and maintenance. Do not add overlapping workflows.
Historical judgments, original tasks
923 occupation profiles and 19,265 task statements come from Eloundou et al.’s GPT-4-era research using O*NET 27.2. They are not a benchmark of models available today. Human annotations are the default; the GPT-4 annotation is shown separately.
Exactly how the index works
Core tasks receive weight 2; supplemental and unspecified tasks receive weight 1. α counts direct LLM assistance (E1); γ counts E1 plus software-assisted tasks (E2); β counts E1 plus half of E2. Each is divided by all task weights. The rubric asks about at least a 50% reduction in task completion time at consistent quality. These are task weights, not hours.
U.S. wage data and specialty limits
BLS May 2023 national figures match 892 profiles to 772 distinct detailed occupations. The other 31 profiles remain searchable with unavailable wage metrics. Specialty profiles share their parent’s wage and employment estimates. National employment rankings show each parent once. The wage-exposure ranking uses the general profile (.00), or the sole profile when unambiguous; it does not invent a specialty-weighted score.
Comparisons and scenarios
Percentiles compare human β with all 923 research profiles, counting strictly lower scores; ties share a percentile. They are not employment-weighted. The wage-exposure index is employment × mean annual wage × β, a modeled proxy. The workflow calculator instead uses your explicit task volumes, durations, review time and time-reduction assumptions. No index is a claim of cash savings.
Data sources, licenses, and downloads
This page includes information from the O*NET 27.2 Database by the U.S. Department of Labor, Employment and Training Administration (USDOL/ETA), used under CC BY 4.0. O*NET® is a trademark of USDOL/ETA. TerraNet Technologies has transformed the information into joined records, calculations, and examples. USDOL/ETA has not approved, endorsed, or tested these modifications.
Exposure labels: Eloundou et al., GPTs are GPTs, from the pinned research release. The MIT copyright and permission notice is retained.
Wages and employment: BLS May 2023 OEWS, public U.S. government data. Estimates exclude self-employment and other populations described in the technical notes. Hourly and annual wage fields remain unavailable when not published; top-coded medians are not treated as exact values.
All 5,538 supplied occupation scores reproduce within 1e-9. The BLS source workbook was verified against the official release byte-for-byte. Sources were validated October 18, 2026; newer wage releases do not update the research’s capability judgments.