Published · Updated
7 MIN READ
Where to use AI in your business: a map by department and task
Where to use AI in your company is something you find by looking at tasks, not job titles or tools. The best entry points are frequent pieces of work that eat up hours, follow a pattern, involve text or digital data and have errors that are easy to check. Customer service, sales — including AI lead qualification — and administrative routines tend to concentrate these tasks, and that's where the map starts.
The problem is rarely a lack of opportunity. In a Gartner survey of 2,986 employees, conducted in July 2025, only 42% said they knew how to identify where AI could improve their own work. The tool arrives before the method, and usage ends up scattered across loose conversations.
At Wama, this question opens the AI Assessment, the first stage of the journey: it shows where AI speeds up slow, time-consuming and error-prone processes before any execution. What that work delivers is detailed in the article on the AI diagnostic for companies.
Where should a company use AI first?
Use AI first where the same task repeats many times a week, adds up to significant time and produces a result someone can check in a few minutes. Reading, classifying, summarizing, drafting and finding information are the verbs where generative AI pays off early.
OpenAI's guide to identifying use cases, published in October 2026, points to three signs of opportunity: repetitive, low-value tasks; skill bottlenecks (when the team waits on another department to move forward); and work stalled by ambiguity, where there's no starting point.
Brazilian data confirms where adoption starts. In the IBGE's semiannual PINTEC survey (IBGE is Brazil's national statistics institute), released in September 2025, industrial companies with 100 or more people that used AI in 2024 applied it mainly in administration (87.9%), sales (75.2%) and product and process development (73.1%), ahead of production (52.0%). If you want to go from the map to the workflow, the next step is in which process to automate with AI first.
A map by department: customer service, sales, finance and operations
Each department has a type of task where AI usually comes in with little risk. The map below is meant to start the conversation with each team, not to decide on its own.
Department | Typical task | What AI does |
|---|---|---|
Customer service | Questions that repeat every day | Classifies the request and drafts the reply based on company documents |
Sales | Logging conversations and proposals | Summarizes meetings, fills in the CRM and prepares the proposal draft |
Finance | Checking documents | Reads invoices and payment slips, flags discrepancies and prepares the report for review |
Operations | Looking up internal procedures | Finds the right rule in company material and cites where it came from |
Marketing | First drafts of copy | Writes drafts that a person reviews before publishing |
In customer service, the channel changes the math: anyone serving customers on WhatsApp needs to look at the rules and costs of an AI agent on WhatsApp before choosing the tool.
In sales, the gain depends on what the system lets the AI read and write. CRM integration with AI sets that limit before the agent touches the pipeline.
How do you build a task inventory in one week?
For five business days, each person writes down what they did, how many times and how long it took. "I'm a financial analyst" doesn't help; "I reconcile bank statements every morning" does. The job description hides the small tasks, and they're the ones that add up to hours by the end of the month.
A simple spreadsheet does the job, with five columns per row:
Task: the verb and the object, such as "answer delivery time questions."
Times per week: counted, not estimated from memory.
Minutes each time: the time of whoever does it, including looking for information.
Where the information is: a system, a spreadsheet, email or someone's head.
Who checks it: the person who notices if the result came out wrong.
The last column is the one that best separates good candidates from bad ones. A task with no one able to check it isn't a starting point, however many hours it takes. And when the information lives in scattered documents, the project comes to depend on a well-built AI agent with a knowledge base.
Where AI shouldn't act on its own
Some repetitive tasks aren't good candidates, or only come in with a person in the loop. The yardstick is the cost of the error combined with how hard it is to notice.
Decisions about people: credit, hiring, firing and individual pricing affect rights and call for recorded human review.
Sensitive personal data: health, biometrics and data on minors require a legal basis and care with what goes to the model, as the article on LGPD (Brazil's data protection law) and artificial intelligence explains.
Numbers that go straight to the customer: calculations, billing and contract deadlines need checking before they're sent.
Actions that can't be undone: paying, canceling or deleting stays with a person.
Tasks that shouldn't even exist: a report nobody reads gets eliminated, not automated.
At Wama, every AI result goes through human review before reaching the client, and that's the rule we recommend at the start of any new use. Writing it down in a company AI usage policy keeps each team from inventing its own limit.
How do you turn the map into an execution plan?
Pick the two or three highest-scoring tasks, measure today's time and test for a few weeks before buying anything big. The before number is what lets you say afterward whether AI helped.
There's evidence the gain is real when the task is well chosen. In an MIT study of 453 professionals, released in July 2023, access to ChatGPT on writing tasks cut time by 40% and raised quality by 18%. Writing is one type of task; the result doesn't transfer to every process.
At Wama, the map of where AI comes in becomes the custom Implementation Plan, with the baseline to measure before and after. There's no boxed solution: each company gets its own plan. The stages of AI implementation in companies show what comes after the pilot, and the cost of each case appears in how much it costs to implement AI.
Common mistakes
Starting with the tool: subscribing to a platform and then looking for where to use it reverses the order and leaves licenses idle.
Mapping job titles, not tasks: the job description hides the small tasks that add up to hours.
Picking the flashiest case: complex cases impress in the presentation and delay the first result.
Not measuring beforehand: without today's time, nobody can prove a gain later.
Automating the exception: AI pays off on the pattern; the rare case stays with a person.
Hiring without criteria: whoever will execute needs to show a method, and the signs are in how to choose an AI implementation company.
Frequently asked questions
Where do you start using AI in a company?
Start with a one-week task inventory, with the frequency and time of each task. Choose the ones that repeat, add up to hours, follow a pattern and have errors that are easy to check. Test two or three for a few weeks, measuring time before and after, and only then expand to other departments.
Which departments use AI the most?
In Brazilian industry, among companies with 100 or more people, the IBGE's 2024 semiannual PINTEC showed usage concentrated in administration, sales and product and process development, ahead of production. In service companies, customer service and sales usually show up early on the map, because of the volume of text that flows through them.
Do you need a lot of data to use AI?
For reading, summarizing and drafting tasks, no. What you need is for the information to be accessible and organized, such as documents, customer service history and written procedures. Projects that train a proprietary model need more volume, but they're rarely the first step for a company still mapping where to use AI.
Should every repetitive task use AI?
No. A task that's always the same, with no interpretation, is solved with a simple rule and comes out cheaper. A task with an expensive, hard-to-notice error, or that handles sensitive data, calls for a person in the loop. And a task nobody uses should be eliminated before any automation, with or without artificial intelligence.
How does Wama help find where to use AI?
Through the AI Assessment, the small, paid first step of Wama's journey. It shows where AI fits into the company and produces the custom Implementation Plan. When the client moves on to the plan, the diagnostic fee is deducted from the first installment, so the mapping doesn't become a separate cost.
If you want to know which tasks in your company are worth using AI on first, talk to Wama and start with the map.
