Verify accuracy of financial information
Most of the output can be produced by software today.
Software can pull the data, run the comparison and write up what it found. The accuracy depends almost entirely on whether it is wired into the real system of record rather than a stale copy, which is the problem the tools and auth layer exists to solve.
| How automatable | Mostly automatable |
|---|---|
| Kind of work | Analysis and evaluation work |
| Jobs that do it | 5 occupations |
| Tool categories that apply | Builders, Tools and auth |
| O*NET activity | Processing Information (4.A.2.a.2) |
What this looks like on the job
Not our paraphrase. These are real task statements recorded against this activity, spread across the 5 occupations that perform it.
- Examine records, reports, or other documents to establish facts or detect discrepancies.
- Review taxpayer accounts, and conduct audits on-site, by correspondence, or by summoning taxpayer to office.
- Examine budget estimates for completeness, accuracy, and conformance with procedures and regulations.
- Verify and inspect cash reserves, assigned collateral, and bank-owned securities to check internal control procedures.
- Review loan agreements to ensure that they are complete and accurate according to policy.
- Prepare, analyze, or verify annual reports, financial statements, and other records, using accepted accounting and statistical procedures to assess financial condition and facilitate financial planning.
What we would actually use
One recommendation rather than a shortlist, because a shortlist is just your problem handed back. This is what we would buy for analysis and evaluation work, and what it costs.
Start in the assistant with an export of the real data, because the accuracy problem here is a stale copy rather than a weak model. Make, at $9 a month, is the cheapest way to put the pull on a schedule once the analysis is worth repeating.
Chosen for the cheapest thing that does the job, not the best funded. Nothing on this site is sponsored, and the full landscape is there when you want to disagree with us.
The rest of the category
Context, not alternatives to weigh up. We name the layer rather than promise a named product does your specific task.
Matched from the kind of work, not from vendor marketing. Check any of them can reach the system your records actually live in — that connection, not the model, is where these projects stall.
How people actually do it
No workflow names this activity yet, but these automate the same kind of work, with the prompts to paste and what to keep for yourself.
Who does this work
5 occupations in the O*NET database perform this activity. Each one has a full breakdown of its other tasks.
Work that goes with it
O*NET groups these under “Evaluate the quality or accuracy of data”. In practice they tend to be done by the same person, in the same sitting.
Method. The activity, its taxonomy placement and the occupations that perform it come straight from the public-domain O*NET 30.3 database. The automatability band is ours: we map each of O*NET’s 41 generalized work activities to how much of its output current software can produce, assuming a person still reviews and owns the result. It is a coarse three-way judgment on purpose. A precise-looking percentage here would be invented.