AI Use Cases, Department by Department

In short: AI pays off fastest on work that repeats, follows known rules, and produces output somebody can check. That description points at a different job in every department: invoice matching in finance, first-pass screening in HR, quote preparation in sales. The scenarios below were selected on volume and measurability. According to McKinsey's 2025 survey, only 21% of organizations using generative AI have redesigned a single workflow; choosing a scenario is the first step into that 21%.

The test for choosing a scenario

Three questions decide whether a job suits AI:

1. How many times a month does this repeat? Work that repeats rarely never earns back the setup cost.
2. Can "correct" be defined? If it cannot, the output cannot be checked.
3. What does a wrong result cost? If the cost is high, human approval is mandatory — which does not make automation worthless, but it changes the design.

Scenarios chosen without answering these three tend to be interesting but low-volume.

Finance and accounting

Invoice and receipt matching. Matching an incoming invoice against the order and delivery note, and flagging differences in amount or line items. High volume, known rules, a definable correct answer. Metric to measure: minutes per match, share of invoices needing manual intervention.

Expense classification. Assigning free-text expense descriptions to the right accounting code. Metric to measure: correct classification rate, period close duration.

Where the human stays: differences above a threshold and first-time suppliers go to approval.

Human resources

Application pre-screening. Summarizing applications against the requirements in the posting and flagging missing information. The aim is faster reading, not the screening decision itself. Metric to measure: review time per application, days to first interview.

Answering internal policy questions. Answering leave, expense and procedure questions from the organization's own documents. Metric to measure: number of repeat questions reaching HR.

Where the human stays: every decision about a candidate. Screening decisions are not left to automation because of discrimination risk, and the lawful basis for processing personal data has to be defined in advance.

Sales

Quote preparation. Producing a draft quote from past quotes and the price list. Metric to measure: quote preparation time, share of inquiries answered.

Writing meeting notes into the CRM. Summarizing notes into the relevant record and proposing the next step. Metric to measure: CRM record completeness, hours a rep spends on admin.

Where the human stays: pricing and discount decisions, and the final text that reaches the customer.

Customer service

Classifying and routing incoming requests. Identifying the subject, assigning it to the right team and flagging urgency. Metric to measure: first response time, misrouting rate.

Drafting replies. A draft answer based on the knowledge base and past resolutions. Metric to measure: handling time per request, first-contact resolution rate.

Where the human stays: complaints, returns and compensation.

Operations and supply chain

Document reading. Extracting structured data from delivery notes, customs paperwork and contracts. Metric to measure: minutes per document, data entry error rate.

Demand forecasting support. Projecting seasonal demand from historical data. Metric to measure: inventory turnover, unfulfilled order rate.

Where the human stays: the decision to place an order and the choice of supplier.

Marketing

Content production and localization. Adapting existing content for different channels and languages. Metric to measure: pieces published, hours per piece.

Summarizing campaign results. Collecting data from several sources into one report. Metric to measure: report preparation time.

Where the human stays: brand voice, campaign strategy, and the final text that goes out.

Which department to start with

The choice of department is usually made by picking the most enthusiastic manager. A better test is this: start wherever the highest-volume work with the clearest definition of correct happens to sit.

Criterion	Why it matters
Monthly repetitions	Volume has to cover the setup cost
Definability of "correct"	Measurement and review depend on it
Data order	A scattered document pool eats a third of the project
A process owner exists	Without an owner, decisions sit waiting

Starting with the process that scores highest on these four is the fastest route to a first result.

Frequently asked questions

How many departments should start at once? One. In organizations that launch five processes simultaneously, none reaches a measurable result. The second starts after the first has been measured at 90 days.

Do the scenarios change by sector? The departmental logic does not change; the weightings do. In manufacturing, operations and supply chain come forward; in services, customer support and sales.

Can these scenarios be covered by off-the-shelf tools? Some of them, yes. Scenarios that require reaching into internal documents and connecting to your own systems are where off-the-shelf tools fall short.

How do employees react to these scenarios? The reaction depends on how the work is explained. Teams that can see which step of their own job gets faster adopt it; a general "AI is coming" narrative produces resistance.

Sources

• McKinsey & Company, The State of AI: Agents, innovation, and transformation (2025) — 88% of organizations use AI in at least one function, only 39% can show an impact on earnings; only 21% of generative AI users have redesigned a workflow.
• Gartner, Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027 (June 2025) — unclear business value is named among the reasons for cancellation.

Closing

A scenario list is not a menu but a filter. The right scenario is not the interesting one but the high-volume one with a definable correct answer.

See our process automation service or read the article on process mapping.