AI for data, finance and legal: make the results reviewable

At the monthly meeting, one person blames churn, another blames discounts, and finance mentions a payment whose timing does not match. An AI can turn those guesses into a polished report before anyone has established the cause.

Data, finance and legal work all involve analysis, but they require different checks. Data needs consistent definitions; reconciliation needs balances and evidence; contract review needs clauses and a company position. We will use fictional business materials to show how role plugins can make those checks visible.

These community selections adapt Anthropic’s Knowledge Work Plugins. Onevium independently maintains the packaging, with the Apache-2.0 license, pinned revision and modification record included.

The selections install local analysis and drafting instructions without database, accounting or contract-system connectors. All figures, company positions and contract text below are fictional teaching materials. The reference answers are checks for your run, not measured customer outcomes.

The open-source license does not cover model usage or external business-system fees. These exercises need files and a configured model; they do not require a database or accounting connection.

Save this as sales.csv:

period,segment,orders,revenue
previous,small,80,8000
previous,large,20,6000
current,small,100,10000
current,large,10,3000

Both periods are complete months in the same currency. Segments do not overlap; assume no tax or refund adjustments. Ask analyze to compare revenue, orders and revenue per order by segment, show its calculations and avoid causal claims. Then ask validate-data to review the analysis.

You can check the arithmetic: revenue falls from 14,000 to 13,000, about 7.14%. Orders rise from 100 to 110. Revenue per order falls from 140 to about 118.18, while the segment-level figures remain 100 and 300. Averaging those two segment averages would give the wrong overall result.

The mix changed: fewer high-value orders and more low-value orders. This table does not explain why. A churn claim needs customer and renewal evidence; a discount claim needs pricing evidence. The upstream data plugin includes validation methods that help keep conclusions within what the inputs support.

Give reconciliation the following simplified period-end example, with all amounts in one currency:

Bank balance: 10,200
Book balance: 10,050
Deposit in transit: 500, supported by a receipt; bank credit next day
Outstanding cheque: 700, supported by its issue record
Bank fee: 50, deducted by bank and not recorded in the books
Assume no other differences in this exercise.

Request separate bank-side and book-side adjustments, with no ledger changes or approved journal entries. The reference calculation is 10,200 + 500 - 700 = 10,000 on the bank side and 10,050 - 50 = 10,000 on the book side.

Each adjustment also needs its type, amount and supporting record. Remove the fee evidence and repeat the exercise: the result should retain an unexplained difference of 50, rather than invent an adjustment to force agreement.

variance-analysis can help compare budget and actual results, but the team must supply consistent periods and materiality rules. Sample upstream thresholds are not your approval policy. Qualified professionals must review formal financial work. See the finance plugin.

For a fictional procurement exercise, prepare these two documents:

Tell review-contract that you represent the buyer, are comparing only these two provisions, and have not specified a governing jurisdiction. Ask for the original text, company standard, deviation, business impact and questions for counsel.

A useful finding explains that the draft requires an earlier renewal decision and asks whether procurement can complete its internal review in time. The difference alone does not establish illegality. For export, discuss retrieval and continuity concerns without inventing a legal entitlement to thirty days.

The upstream legal plugin relies on company playbooks and notes the jurisdictional context of its default examples. Your legal professionals should maintain the applicable standards. This fictional comparison is not contract advice.

Try a deliberate gap:

• Give the data workflow an incomplete month and check whether it flags the comparison.
• Remove reconciliation evidence and check whether it retains the unexplained difference.
• Omit the company playbook and check whether contract review acknowledges the gap.

These are acceptance conditions you can run, not a claimed model pass rate. Keep the inputs, model, output and review findings together so later changes can be evaluated.

All three workflows benefit from traceable evidence. Their professional review standards remain different.

Open Plugins in Onevium’s left navigation. Find Data analysis essentials, Finance essentials or Legal essentials in Onevium picks. These community selections retain the upstream author and license; Onevium maintains the adaptation.

Select your practice project in Scope, then click Add beside the package. In the installation window, check its name, version, publisher, included skills and selected scope. Read the contents, then choose Install and enable. Back in the directory, confirm that the package is enabled for your project before starting a new conversation turn.

If the new packages are missing, open Sources → Check for updates and wait for a successful directory refresh. You do not need to download a ZIP, import JSON or run a preparation script.

Open the installed package details and find Use in conversation. Use the complete name shown there—for this article, /onevium-data-essentials:analyze—followed by the task and inputs above. You can also explicitly ask the AI to use that installed skill. Check that the run loads the intended skill and reads the supplied material, then review the answer using the criteria in this article.

The companies and figures are fictional. The support case includes observations from one actual invocation; the other examples provide reference answers, not evidence that every skill has been tested in business use. See the plugin guide and Skills guide for more controls. The community repository provides source code, licenses and adaptation records.

Start small: four CSV rows, a simplified reconciliation or two fictional clauses. Open the output and verify its calculations and references before increasing the scope.

The plugin brings a working method into the conversation. People still maintain company rules and select the right inputs. For customer-facing work, continue with marketing, sales and support workflows.