Generative AI for business: real uses beyond chat

Ask a manager whether their company already uses generative AI and the answer is almost always yes. Ask them to show you where, and they will often open a chat tab: someone in marketing asking for caption ideas, someone in sales rewriting an email, someone in finance asking for a spreadsheet formula. All of that helps. None of it shows up in the quarterly results.

The numbers confirm the gap between using and gaining. In McKinsey’s The state of AI in 2026 (published in August 2026, with 1,719 respondents), nearly nine in ten organizations report regular AI use in at least one business function. Only 37% attribute any impact on operating profit (EBIT) to it, the same share as the year before.

The problem is rarely the tool. It is where the tool was placed.

Generative AI for business is the use of models that produce text, images, code and analysis to carry out business tasks: drafting proposals, summarizing support conversations, generating ad variations, triaging tickets and explaining numbers. The value shows up when it leaves the one-off chat and enters a process with data, an owner and a before-and-after metric.

Why doesn’t chat on its own show up in results?

One-off chat creates individual productivity, and individual productivity gets diluted. If an analyst saves 40 minutes a day writing emails, that time turns into more meetings, more breaks or more small tasks. Nobody measured how long the email took before, so nobody knows how much was gained after.

Then there is variation. Each person asks in their own way, with different context, and gets a different answer. There is no standard to review and nothing to improve. It is the same difference between an AI agent and a chatbot: having a conversation is not the same as running a step of a process.

In practice, companies that get a return from generative AI climb three steps, and almost all of them stop at the first.

Real uses of generative AI in business, by function

The test for picking a use case is simple: a repetitive, high-volume task that eats someone’s hours today and has an output you can check. These are the ones that work best for small and mid-sized companies.

Marketing

Ad variations from an approved brief, product descriptions at scale, adapting one piece of copy to each channel, and a weekly summary of campaign performance. The biggest gain is not writing the first ad. It is producing the 15 variations an A/B test needs. We cover nine applications in how to use AI in digital marketing.

Sales

Meeting summaries that go straight into the CRM, proposal drafts built from the client’s history, follow-up emails that carry the context of the last conversation. A rep who spends 20 minutes per meeting logging notes and runs 15 meetings a week loses five hours a week to typing. More examples in AI for sales.

Customer service and operations

Ticket triage by reason and urgency, draft replies for the agent to review, summaries of recurring complaints for the product team. Here generative AI works better as a sorting layer than as the final answer.

Finance and back office

Reading invoices and contracts to extract fields, plain-language explanations of cash flow swings, collection email drafts. It is the least discussed area and one of the most manual. Tellingly, in Brazil’s IBGE Pintec survey (September 2025), among industrial companies with 100+ employees that use AI, 87.9% apply it to administration, ahead of sales and commercial work (75.2%).

Function	Generative AI use	What to measure
Marketing	Ad variations and product descriptions	Hours per asset and variation performance
Sales	CRM meeting summaries and proposal drafts	Logging time and time to proposal
Customer service	Triage and draft replies	First response time and rework
Finance	Data extraction from documents	Documents per hour and errors
Management	Explaining numbers and reports	Reporting hours and decisions made

A concrete example: an online cosmetics store with 600 products and no decent descriptions. Written by hand at 20 minutes each, that is 200 hours. With AI drafting from the spec sheet and a person reviewing in 6 minutes, the total drops to 60 hours. That is 140 hours saved on a project that had been stuck for a year.

How do you measure the return on generative AI?

The math is the same as for any investment, with one extra point: the cost includes human review time, which many people forget.

Here is a simulation in Brazilian reais. An agency with four analysts builds 30 monthly client reports, each taking 3 hours: 90 hours a month. With AI drafting from campaign data and the analyst reviewing, each report takes 1 hour. That frees 60 hours. At an hourly cost of R$70, it is worth R$4,200.

On the other side: R$600 in tools and 8 hours a month tuning instructions and fixing outputs (R$560). Total cost of R$1,160. Net return of R$3,040, or 262% on what was spent.

This math breaks at a point few people admit: if those 60 hours are not used for anything, the money does not exist. It only appears if the agency serves more clients with the same team or avoids a hire. That is why many companies prefer to measure capacity instead of “savings”: clients per analyst, before and after.

Beyond money, track three simple signals:

• Time per task, measured before and after, on the same task.
• Quality, as the share of outputs approved without rework.
• Adoption, as the share of that task type that actually goes through AI.

Where to start without losing six months

The common mistake is starting with the tool. Start with the task.

1. Pick one task that is repetitive, happens every week and has a checkable output. Reports, product descriptions, meeting summaries.
2. Measure the current state for two weeks: how often it happens and how long it takes.
3. Put AI inside the workflow, with a standard instruction, defined inputs and one person responsible for approval.
4. Compare after 30 days and decide: scale, adjust or drop it. Dropping it is also a result.

Two cautions. The first is data: do not paste customers’ personal data into tools without a business contract, especially under privacy laws such as Brazil’s LGPD or GDPR. The second is cost and people. In the same IBGE survey, among industrial companies already using advanced digital technologies, 78.6% cite high costs and 54.2% a lack of qualified staff as obstacles. Starting small addresses both.

Where generative AI still gets it wrong

It is great at drafting, summarizing and reorganizing. It is weak at calculating. A language model can add up a revenue column wrong with total confidence, and the error slips through because the surrounding text sounds right.

The rule that works: the number comes from a query against the data (SQL, spreadsheet, API), and generative AI explains the number. Never the other way around. If you use AI for analysis, our prompt templates for data analysis follow this logic and ask the AI to show where each value came from.

The other limit is context. The AI does not know that your September campaign had a tracking issue, or that client X pays in 90 days. Without that context, the answer is generic. The higher you climb the steps in the diagram, the smaller this problem gets, because the AI starts reading your own company’s data.

Frequently asked questions

What is generative AI for business?

It is the use of models that create text, images, code or analysis to carry out business tasks such as proposals, reports, ads and support triage. The difference from personal use is that it sits inside a defined process, with company data and someone accountable for the result.

What are examples of generative AI in companies?

The most common are ad variations and product descriptions in marketing, CRM meeting summaries in sales, ticket triage in customer service and data extraction from invoices and contracts in finance. All of them have volume, repetition and an output that is easy to check.

How do you measure generative AI ROI?

Add hours saved times hourly cost plus any incremental revenue, then subtract the cost of tools and human review hours. Measure the same task before and after. Remember that saved hours only turn into money if they are used for something else.

Is generative AI safe for company data?

It depends on the tool and the contract. Business plans usually guarantee your data is not used to train the model, but read the terms. Because of privacy laws, avoid pasting customers’ personal data into free or personal accounts.

What is the difference between generative AI and AI agents?

Generative AI produces content when someone asks. An agent uses the same kind of model to run a sequence of actions on its own, such as querying data, deciding the next step and delivering a result. Agents are one way to take generative AI to the third step. See AI agent examples by function.

Generative AI belongs close to the data

The real gain from generative AI is not better writing. It is shortening the distance between data and decision. When it reads the company’s numbers and returns an explanation that someone approves, the time saved becomes visible and the return stops being a hunch. That is what Sherlok is built for: connect Meta Ads, Google Ads, GA4, your CRM and spreadsheets, ask in plain language and get the analysis ready, with every number calculated from the data, not made up by the text.