How Budget Analysts Should Read BLS AI Projections

How should a budget analyst read BLS's AI discussion when projected employment grows despite automatable desk tasks?

Treat the signals as different layers. The BLS case study discusses plausible task change; the newer national table projects occupational employment; neither says what an employer will adopt or what will happen to an individual analyst. Map recurring work and watch for local evidence before making a costly career change. The dates matter. “Incorporating AI impacts in BLS employment projections: occupational case studies” describes research for the 2023–33 projections cycle, based on information available in June 2024. “Occupational projections and worker characteristics” covers 2025–35. These are different cycles, not simultaneous measurements of AI's effect. The earlier discussion considers how software might speed budget review or help visualize information; the newer table summarizes projected employment across the occupation. A task may take less time while employment grows modestly because tasks and headcount are measured at different levels, and the projection reflects many forces. BLS says projections cannot isolate one technology's effect; its AI exposure categories are supplemental, not employment forecasts (“Employment Projections Frequently Asked Questions”). Read growth neither as proof the job is safe nor evidence of impending loss. Ask which parts of your workflow may change, what your employer permits or plans, and what you can verify locally.

Sources: Incorporating AI impacts in BLS employment projections: occupational case studies; Occupational projections and worker characteristics; Employment Projections Frequently Asked Questions

What does the newer BLS projection actually say?

Read the table heading before the percentage. Occupational projections and worker characteristics labels this U.S. Bureau of Labor Statistics series 2025–35 and reports employment in thousands. The Budget analysts row (occupation code 13-2031) starts at 50.4 thousand employed in 2025 and reaches 51.3 thousand in 2035. Its percentage-change column gives 1.9 percent, while numeric change displays 1.0 thousand. Because the values are rounded to one decimal place, subtracting the displayed endpoints gives 0.9 thousand; that subtraction should not be used to “correct” the separately calculated and rounded change field. Read the columns as published, retaining their units and period. Together they describe modest projected national growth, not a measure of how many positions AI will create or remove. The row lists 2.9 thousand occupational openings as an annual average for 2025–35. Openings and net employment change answer different questions. Net change compares projected occupation totals at the start and end of the decade; annual openings estimate positions expected to become available in a typical year, including openings associated with workers leaving the occupation or labor force as well as growth. Thus annual openings can exceed the decade’s net addition expressed as a yearly average. They are not 2,900 newly created jobs each year, vacancies in a particular city, or a promise that any applicant can access them. For an analyst deciding whether to stay, move, or retrain, the number is national context, not a personal forecast. One BLS page can also show a different projection vintage from another. The live Budget Analysts Occupational Outlook Handbook page still displays the 2024–34 outlook, including 1 percent projected growth and about 3,100 annual openings. Those figures are not an alternative reading of the 2025–35 row; they refer to a different ten-year cycle. When citing current projections, use the explicitly dated 2025–35 table and identify that period. The Handbook remains useful for occupational descriptions, but its projection figures should not be silently blended with the newer table or relabeled as 2025–35 estimates. The table lists a 2025 national median annual wage of $91,640 and a bachelor’s degree as typical entry education. These are characteristics attached to national occupational data, not a salary promise or a statement that every employer requires the same credential. A reader comparing options should check local postings and pay ranges against their salary floor, location, and experience. The national projection cannot reveal whether a nearby agency is hiring, whether a department’s budget is expanding, or whether an office plans to change staffing. Record the table’s projection years, occupation code, units, and column name before quoting a figure. Keep employment change, annual openings, wage, and education in their proper roles. BLS’s current row supports a measured baseline of modest projected growth with recurring openings across the United States. It cannot establish that AI caused the outlook, that local opportunities will match the national pattern, or that an individual analyst’s position is secure. Those questions require evidence about the analyst’s region, employer, and actual duties.

Sources: Occupational projections and worker characteristics; Budget Analysts

Why do automatable desk tasks coexist with occupational growth?

Automating or assisting with selected tasks does not mechanically determine how many people an occupation employs. A task is one part of a job; employment projections estimate the net result for an occupation across many workplaces and influences. In its 2025 article “Incorporating AI impacts in BLS employment projections: occupational case studies,” the Bureau of Labor Statistics (BLS) described budget analysts as doing predominantly desk work, with computer software already central. It reasoned that further software advances could raise productivity without eliminating demand for analysts. That is a plausible way the signals can coexist, not proof that every analyst’s work will remain unchanged. The case study was built for the 2023–33 projections cycle, using the program’s interpretation of information available in June 2024. BLS expected AI to speed parts of the budget review process and provide visualization tools for presentations. The agency also identified analysts’ presentations, stakeholder questions, and discussions of the nuances and alternative paths in proposed budgets as work likely to continue involving human conversations. From that perceived continuing need for core tasks, BLS projected 3.9 percent employment growth for budget analysts over 2023–33, about as fast as the average for all occupations in that cycle. The figure describes that decade’s projection. It is neither a current adoption measure nor the newer 2025–35 estimate. The reasoning is about the bundle of work, not a claim that communication is permanently beyond automation. Reviewing figures or turning them into a first-pass visualization may take less effort, while deciding what a variance means, weighing alternatives, and explaining consequences can remain part of the analyst’s assignment. BLS’s account treated those latter activities as a continuing source of demand in its model. It did not measure how much time each task takes for an individual or how staffing changes after a task becomes faster. The case study is a projection judgment made at a stated time, open to revision as evidence changes. Several outcomes could follow from faster preparation, and the cited case study does not quantify their relative likelihood. An organization might need fewer labor hours for the same deliverable; use saved time to examine more programs or scenarios; shorten turnaround; or move analysts toward checking, interpretation, and explanation. These are possible pathways for productivity to affect work, not documented results from a budget office. Which one occurs depends on workload, tool quality and cost, information-handling rules, review requirements, and organizational choices. A capability to produce a draft is therefore different from observed use, and use is different from a staffing decision. The newer “Occupational projections and worker characteristics” table gives a separate national baseline: for 2025–35 it lists budget analyst employment rising from 50.4 thousand to 51.3 thousand, or 1.9 percent. It does not attribute that projected change to AI. The case study’s 3.9 percent and the table’s 1.9 percent cover different projection cycles; the earlier explanation cannot simply be carried forward as the cause of the newer figure. In its Employment Projections FAQ, BLS explains that its projections reflect multiple factors and cannot isolate one technology’s contribution. The current table is an aggregate forecast shaped by interacting assumptions, not a controlled estimate of what AI alone will do. For a worker, positive projected growth should not erase the possibility that particular duties will be redesigned, just as a task that appears automatable does not establish that an occupation’s employment will fall. The most defensible reading is narrower: in the older cycle, BLS judged that efficiency gains in some budget-analysis tasks could coexist with continued demand for the occupation’s broader work; the newer cycle still projects modest national growth, without assigning that outcome to AI. This supports neither “AI has no labor effect” nor “the job is safe.” It supports watching how tasks and responsibility change locally before treating a national projection as a personal forecast.

Sources: Incorporating AI impacts in BLS employment projections: occupational case studies; Occupational projections and worker characteristics

What does BLS mean by AI exposure, and what does it leave out?

BLS’s supplemental 2025–35 AI exposure categories are a relative comparison of how AI may overlap with some tasks across occupations. They draw on five external sources that include theoretical capability measures and measures of observed AI use mapped to occupational tasks. BLS is explicit about the boundary: an exposure category is not a forecast of employment growth or decline, a prediction that an occupation will be automated, a measure of productivity or wages, an estimate of adoption, or a worker replacement estimate. For a budget analyst, it is a prompt to inspect particular tasks, not a verdict about the job or the person. The category combines unlike evidence. Three inputs estimate theoretical exposure: they ask, in different ways, whether AI capabilities relate to occupational abilities or might reduce time spent on tasks. Two inputs use observed interaction data associated with occupational tasks or work activities. In Artificial Intelligence (AI) exposure categories, BLS explains that these observed measures do not directly establish that workers in a given occupation used AI on the job. They capture mapped interactions in the underlying data, not a survey of budget offices, employer policy, or an adoption rate for budget analysts. “Observed” therefore describes one input’s evidence type; it does not mean BLS watched an analyst complete a budget cycle. BLS places the five inputs on a comparable relative scale by converting each occupation’s score to its percentile rank among occupations covered by that source. It then forms separate summaries for the theoretical and observed dimensions and groups occupations into four categories: low, moderate, high, and very high. The labels say where an occupation sits relative to others in this classification. They do not express an absolute amount of exposed work. A “high” label is not a statement that most hours in every analyst’s week can be automated, and a “low” label is not a safe-from-change designation. This method makes the categories useful for one limited question: where might a worker look more closely for tasks that AI could assist with or complete? In budget analysis, that might lead someone to examine a repeatable first-pass summary, a calculation explanation, or a draft of a routine document. These are candidate tasks to investigate, not BLS findings about which tasks in the budget analyst occupation are currently automated. The category cannot tell whether a particular office has approved software, whether its data may be entered into a tool, whether the output meets the required error tolerance, or what share of a specific analyst’s time goes to that task. Nor does it show whether any time saved would be used for more analysis, shorter turnaround, reduced workload, or fewer positions. The distinction between exposure and employment projection is methodological, not semantic. Employment Projections Frequently Asked Questions says the exposure categories do not reflect the qualitative research underlying BLS’s employment projections or the factors expected to affect employment demand in the target year. The broader projection combines evidence and assumptions about many influences; BLS says it cannot isolate the impact of a single technology. The categories are supplemental context alongside that forecast, not a hidden AI-only forecast that contradicts it. A budget analyst should therefore use an exposure category as a question generator: which recurring step is digital and repeatable, what policy permits assistance, and who can check the result against authoritative records? Answering those questions requires the analyst’s own task map and workplace rules. The category alone cannot establish actual use, employer adoption, reliable performance, changing demand, or displacement. It must not change a career decision by itself; pair it with observed local evidence before deciding whether to test a bounded task, monitor a change in duties, or research another role.

Sources: Employment Projections Frequently Asked Questions; Artificial Intelligence (AI) exposure categories; Artificial Intelligence (AI) impacts on employment projections

Which steps in a budget analyst's workflow should be examined separately?

Map a budget deliverable from inputs to the decision it supports. Gathering figures and producing a first-pass summary may be easier to test for assistance than reconciling authoritative records, applying local rules, explaining a program tradeoff, or defending a recommendation. This is a hypothesis for examining work, not a claim that a system can reliably complete those steps. Occupational task lists identify places to look; they do not reveal duties, time, systems, or controls in one analyst’s job.

O*NET’s Budget Analysts profile describes connected work: analyze monthly department budget and accounting reports; advise on fiscal allocation and preparation; review trends; examine estimates for completeness, accuracy, and compliance; summarize budgets and recommend funding decisions; consult managers about adjustments; and communicate reports to stakeholders. The Bureau of Labor Statistics’ Budget Analysts Occupational Outlook Handbook likewise describes reviewing proposals, combining department budgets, monitoring spending, explaining requests, and helping managers consider alternatives. These descriptions show occupational breadth, not task frequency or difficulty in a particular office.

Consider an illustrative budget cycle. Departments send spending estimates and supporting explanations. An analyst gathers submissions, checks required records, compares amounts with accounting reports, and flags entries needing clarification. The analyst looks across patterns and proposed program changes, compares funding alternatives, and prepares a summary for decision makers. If a choice affects services or obligations, the analyst may explain assumptions, answer questions, and show why one allocation differs from another. This example maps work; it is not a report of an actual office or tool trial.

Examine each handoff. At intake, are source materials digital, complete, and approved for the intended use? During preparation, is the output stable enough to check against a defined rule, such as whether a total reconciles to the cited ledger or a required field is missing? At review, are exceptions visible, and can someone trace a statement to its source? At interpretation, which local policy, program context, or assumption changes the meaning of a comparison? At presentation, who must explain and defend the recommendation? These questions do not presume automation is authorized or suitable.

A polished summary can still be wrong: it might use a draft rather than the authoritative figure, mix reporting periods, omit an assumption, or overlook a local rule. An error can travel from an input to a trend comparison and distort the alternatives presented to a manager. Review should cover provenance as well as plausibility. Can the analyst identify the source record, period, calculation, and applicable rule behind each consequential figure or explanation? A readable output is not dependable if those links cannot be checked.

Distinguish repeatable preparation from interpretation as a heuristic, not a fixed boundary. Copying approved figures into a standard format may have explicit checks. Yet choosing the authoritative version, recognizing an unusual entry, and understanding how an allocation fits a program may require context. O*NET lists processing and verifying information, evaluating compliance, consulting managers, comparing programs, and communicating reports. These tasks coexist. A seemingly clerical step can involve judgment when an exception makes the usual procedure inappropriate.

The BLS Monthly Labor Review article, Incorporating AI Impacts in BLS Employment Projections: Occupational Case Studies, gives a historical rationale: for its 2023–33 budget analyst projection, it expected software to speed review and support visualization, while nuanced discussion of proposed budgets and alternatives would likely continue to involve human conversations. This projection rationale reflects research available as of June 2024; it is not evidence of current adoption at a particular workplace or a guarantee that communication tasks cannot change. It supports examining the chain of work rather than labeling the whole job.

A practical map can fit on one page: list a recurring deliverable, its source inputs, preparation and checks, interpretive decisions, exception owners, and the audience acting on it. Mark where records must remain traceable and local rules govern. Then identify one bounded preparation step for closer examination, keeping review and decision responsibility explicit. The aim is not to declare duties permanently protected. It is to distinguish a candidate for assistance from the checks, context, and explanation that make a budget recommendation usable. For each step, note what would count as a verifiable result and who is permitted to see the underlying information. If either answer is unclear, resolve that before treating the step as a candidate. This small map describes the actual workflow more precisely than an occupational title or broad task category can.

Sources: Budget Analysts; Budget Analysts; Incorporating AI impacts in BLS employment projections: occupational case studies

How should the two BLS products be compared without forcing a contradiction?

Read the three BLS signals by asking what each was built to describe. The 2025 Monthly Labor Review article, “Incorporating AI impacts in BLS employment projections: occupational case studies,” explains BLS’s reasoning for the 2023–33 cycle. Its budget-analyst discussion drew on information available in June 2024 and projected 3.9 percent growth over that decade. The occupational table for 2025–35 is a newer national employment baseline; its budget-analyst row reports 1.9 percent growth. A third product, “Artificial Intelligence (AI) exposure categories,” groups occupations by relative overlap between AI capabilities and some tasks. That category is not another growth rate. | BLS product | Period and unit | Useful question | Limit | |---|---|---|---| | Monthly Labor Review case study | 2023–33; qualitative rationale alongside a projection | Why might AI affect budget-analyst work while demand continues? | Not later employer adoption or a current standalone AI forecast | | Occupational projections table | 2025–35; national occupational employment | What total employment does BLS project under current assumptions? | Does not isolate AI’s contribution or describe one office | | AI exposure categories | Supplemental relative task-overlap grouping | Which occupations merit closer task examination? | Not employment growth, adoption probability, productivity, or worker replacement | The case study’s 3.9 percent and the newer table’s 1.9 percent are not a falling AI forecast. They cover different ten-year periods and cycles, with updated data and assumptions. The older article says software could speed budget review and support visualization, while discussion of budget nuances and alternatives was expected to continue involving people. This was BLS’s earlier reasoning, not evidence that agencies adopted AI or proof that communication tasks cannot change. The newer table answers a broader question about total occupation employment. It does not isolate technology’s contribution. BLS’s “Employment Projections Frequently Asked Questions” says projections combine quantitative and qualitative methods and account for multiple factors; technological change is assumed to proceed at an overall pace consistent with past experience. The “Artificial Intelligence (AI) impacts on employment projections” factsheet describes high uncertainty and annual updates as new data and analysis become available. The current projection is therefore a useful baseline, not a technology-specific forecast. The case study remains useful for understanding earlier task-level reasoning, with its June 2024 information cutoff in view. The exposure categories answer a different question. The BLS FAQ and “Artificial Intelligence (AI) exposure categories” describe a supplemental classification built from five external sources and relative comparisons across occupations. A higher category can prompt a worker to inspect tasks more closely; it does not say employment will fall. The classification does not distinguish augmentation from automation, and its source measures impose limits. It should not be placed beside 1.9 percent as though the two signals compete: one is relative task overlap, the other a projection of total national employment. The apparent contradiction dissolves when the units are kept separate, though uncertainty remains. Use the case study to understand a possible mechanism, the 2025–35 table for the current national baseline, and the exposure category as a prompt to examine particular tasks. None reveals what a specific employer will implement or how staffing will respond. For that, check local evidence: documented changes in duties, review requirements, approved systems, vacancies, or staffing plans. Treat BLS’s products as bounded context, then assess whether work and responsibility are changing where you are.

Sources: Incorporating AI impacts in BLS employment projections: occupational case studies; Occupational projections and worker characteristics; Employment Projections Frequently Asked Questions; Artificial Intelligence (AI) impacts on employment projections

What local evidence should change an analyst's next move?

Look for repeated changes in the work itself: recurring deliverables take less time after review, the volume or type of analysis changes, checking and correction duties expand, approval rules shift, or managers announce a concrete workflow or staffing plan. One demonstration that a system can produce a draft shows capability; it does not show that colleagues use it, that an employer has approved and integrated it, or that positions will change. A national projection is further removed from those local decisions. Treat each as a different kind of evidence, and wait for a pattern that bears on your actual responsibilities before turning concern into a career move. A short observation log can make that pattern easier to see. Across several reporting cycles, note the task and its inputs, time spent, revisions, error types, who reviewed the result, who remained responsible, and what happened to the work afterward. Did the task disappear, move to another person, take less time, or expand into more analysis? This is a practical record for your own decision, not a validated measurement method. Do not place budget records or sensitive material in an unapproved service just to test a possibility. Separate four questions as you record what you observe. First, can a tool produce a draft or perform a step under some conditions? Second, has someone actually used it for that task? Third, has the employer approved and integrated that use into a workflow, with permissions and review responsibilities? Fourth, is there evidence of changed employment, such as a revised position, an announced staffing plan, or vacancies being handled differently? A “yes” to one question does not answer the next. The BLS Employment Projections Frequently Asked Questions describes technology as affecting industry employment and occupational staffing patterns through different channels, while noting that the projection reflects combined factors and cannot isolate one technology. That is a reason to observe local implementation and staffing separately, not to infer either from a capability claim. The U.S. Bureau of Labor Statistics’ Artificial Intelligence (AI) impacts on employment projections factsheet also emphasizes uncertainty: projections depend on assumptions, are updated annually, and can differ from actual outcomes when assumptions do not hold. Meanwhile, O*NET’s Budget Analysts profile covers varied work, from reviewing reports and checking estimates to advising on resource allocation and consulting with managers. That broad occupational scope helps explain why one automated preparation step may not describe the whole role. It does not reveal your own task shares, your department’s workflow, or its plans. If a preparation step becomes faster, the result could be more time for analysis, shorter deadlines, additional output, fewer hours, or eventually a staffing change. These are possible responses, not outcomes established by a faster draft. The useful local signal is what actually happens next: whether review effort rises or falls, whether the analyst’s responsibilities shift, and whether leaders communicate a concrete change. A specific announced redesign or a change that threatens your pay or duties is a reason to research adjacent roles sooner, using local openings and your salary, location, and training constraints. When change remains hypothetical, a reversible, employer-approved task trial and continued observation are more proportionate than assuming either safety or imminent displacement. In this way, national evidence sets context while repeated workplace evidence helps decide what to do.

Sources: Artificial Intelligence (AI) impacts on employment projections; Employment Projections Frequently Asked Questions; Budget Analysts

What is a proportionate next step before retraining or leaving?

A sensible next move is a small, approved test when your workplace has not signaled a concrete redesign. Inventory one week of recurring work: preparation steps, inputs, reviewer, and resulting decision or deliverable. Choose one bounded step only if your employer permits the tool and data involved, the inputs are appropriate to use, and a reviewer can independently verify the result. Prefer an output that can be checked against approved records. This is a selection heuristic, not a claim that a tool performs the task reliably. Before trying it, agree on what a useful result means. Compare the output with source records and the ordinary process. Record preparation, correction, checking, and explanation time, plus errors, missing context, and extra review. A faster first draft is not a net gain if verification consumes the time saved or responsibility becomes unclear. Stop if inputs are not permitted, output cannot be checked, errors outweigh possible time saved, or no reviewer can stand behind the result. Keep a qualified person accountable for any recommendation affecting a budget decision. This test produces local evidence, not a forecast of employer action. Use the work itself to strengthen portable capability: understand the fiscal context behind a number, trace it to a reliable source, spot an assumption that changes a comparison, and explain a tradeoff clearly. These are ways to build on budget experience, not guarantees that particular skills are immune to automation. Monitor reporting cycles for changes in duties, review burden, deliverables, or responsibility. If a specific staffing plan, redesigned role, or sustained duty change appears, investigate adjacent positions using local postings and employer information. Compare requirements and compensation with your salary floor, location, credentials, learning time and cost, health, and family obligations before committing to a course, project, or larger pivot. The Budget Analysts Occupational Outlook Handbook and the 2025–35 occupational projections table provide national context; O*NET’s profile describes a broad occupation. None settles your local decision. A task checker can organize change-pressure signals, but its result is not a validated probability of displacement. A personalized roadmap can compare scenarios, but it does not guarantee employment or income. Start without either: test one permitted task and keep a concise record. Move from monitoring to adjacent-role research when local evidence changes, and consider substantial retraining only when a plausible option fits your constraints. A generalized exposure category or positive national projection alone is not that trigger.

Sources: Budget Analysts; Occupational projections and worker characteristics; Budget Analysts

Questions readers ask