25 AI Workflow Automation Examples From Real Companies

Founders ask me for examples before they ask me anything else. Fair enough. It's easier to picture AI running a piece of your work when you've seen someone else do it.

So here are 25. Every one is a named company running a specific workflow with AI, and every one comes from a source you can read yourself. I've grouped them by function. For each I say what the work is, what the AI does, who stays in the loop and what the company says came of it.

How I picked these

I didn't work on 23 of these. They come from what the companies published themselves (press releases, engineering blogs, earnings calls) or from major press, read as of October 2026. Where the only source is a vendor's case study, I say so, because a vendor has a reason to make the numbers look good.

The results are the companies' own numbers, not mine. I haven't audited any of them. Where a number is an estimate (time a job "would have" taken) or later got walked back, I say that too.

Two entries are my own anonymised work. They're labelled, and they use only facts I can stand behind.

The 25 at a glance

Customer support

Support is where most companies start, because the work is high volume and the questions repeat. It's also where the most public reversal happened, so read the first entry to the end.

1. Klarna: support chat

The workflow: Handling customer service chats about refunds, returns, payments and disputes.

What the AI does: Klarna launched an AI assistant built with OpenAI in February 2024. It answers chats across 23 markets in more than 35 languages.

Who stays in the loop: Customers can still choose a live agent.

What the company says happened: In the first month it handled 2.3 million conversations, two thirds of all support chats, which Klarna called the equivalent work of 700 full-time agents. Resolution time fell from 11 minutes to under 2.

Worth knowing: In May 2025 the CEO told Bloomberg (as reported by Fortune) that the focus on cost had led to lower quality, and that the company would invest in human support again. Both halves of this story are useful.

2. Octopus Energy: AI-drafted customer emails

The workflow: Replying to customer emails at a UK energy supplier.

What the AI does: Drafts a personalised reply to each email.

Who stays in the loop: A team member reviews the draft and sends it, according to Business Insider's report.

What the company says happened: In 2023 the CEO, Greg Jackson, said AI was answering more than a third of customer emails, doing the work of 250 people, with 80% customer satisfaction against 65% for emails written by staff (City AM).

Worth knowing: These are a CEO's comments relayed by the press, and they're from 2023. The draft-then-review design is the part worth copying.

3. Lyft: rider and driver support

The workflow: Driver onboarding questions, charge disputes and ride problems.

What the AI does: An assistant built on Anthropic's Claude resolves common issues and writes summaries for agents when it hands over (Lyft's announcement).

Who stays in the loop: Safety concerns, complex disputes and cases needing empathy go to human agents.

What the company says happened: Resolution time down by more than 87%, per Anthropic's case study (a vendor case study). Before, customers waited 30 to 40 minutes to reach an agent.

4. Salesforce: its own help site

The workflow: Answering customer questions on help.salesforce.com, around the clock.

What the AI does: An Agentforce agent answers from Salesforce's help content.

Who stays in the loop: Certain topics, renewals among them, are deliberately routed to people.

What the company says happened: By April 2026, Salesforce says it had handled 2.6 million conversations at a 63% resolution rate, and it redeployed hundreds of support engineers.

Worth knowing: A year earlier, at around 500,000 conversations, Salesforce reported an 84% resolution rate. The rate fell as volume grew. Quote these numbers with their dates.

5. Synthesia: support for a fast-growing product

The workflow: Handling inbound support questions while contact volume grew several times over.

What the AI does: Intercom's Fin agent answers support questions.

Who stays in the loop: Human agents take the complex cases.

What the company says happened: In Intercom's case study (a vendor case study), Synthesia says Fin resolves 55% of conversations while monthly contacts rose from about 40,000 to 316,000. Resolution time fell from over five days to under five hours.

Sales

Sales is where a bad automated message costs the most. Notice how many of these keep a person on the send button.

6. Vercel: inbound lead qualification

The workflow: Researching and triaging every inbound sales lead.

What the AI does: Vercel's agent researches the lead and their company, sorts them into a category and drafts a personal follow-up email.

Who stays in the loop: Everything goes to Slack, and a person approves each email before it's sent.

What the company says happened: One person now handles the work of ten, and the other nine moved to more complex sales work.

7. Salesforce: working dormant leads

The workflow: Following up on old leads that sellers never had time for.

What the AI does: An internal SDR agent prospects, sends outreach and qualifies leads.

Who stays in the loop: Human sellers take over for relationships and closing. Note that here the agent does send on its own.

What the company says happened: Salesforce says the agent worked more than 43,000 leads and generated $1.7 million in new pipeline from dormant ones. Early on it answered "I don't know" about 30% of the time. After the team cleaned up the underlying data, that fell below 10%.

Worth knowing: That last detail is the most useful one in the post. The fix was the data, not the model.

8. Morgan Stanley: advisor meeting notes and follow-up

The workflow: Writing up client meetings, logging notes and sending the follow-up email.

What the AI does: With the client's consent, Debrief takes notes, lists action items, drafts a follow-up email and saves a note to Salesforce.

Who stays in the loop: The advisor edits the email and sends it when they choose.

What the company says happened: The launch release quotes one advisor saying it saves him about half an hour per meeting. That's a testimonial, not a firm-wide figure.

9. A software startup: the sales pipeline (my work)

Disclosure: I built this one.

The workflow: Moving every prospect from waitlist to won or lost, with the follow-ups in between.

What the AI does: Three times a day, it picks up new signups, matches replies, notices bookings and no-shows, turns call transcripts into summaries and to-dos, and works out which follow-ups are due.

Who stays in the loop: The founder. It drafts every email and sends none; she reads and presses send. It also asks rather than guesses, and it reports back after every run.

What happened: A non-technical founder runs it herself. Adoption was uneven at first, until we fixed a bug where links in drafted emails expired. I don't have an hours-saved number because we didn't measure one. The full case study has the details.

Finance

10. Microsoft: cash collections

The workflow: Collecting payments and answering customer billing questions, across a team of more than 1,000 collectors.

What the AI does: Predicts late payments, matches payments to invoices, routes emails and drafts replies to customers (Microsoft Inside Track, June 2026).

Who stays in the loop: Collectors focus on the exceptions.

What the company says happened: Payment-matching accuracy rose from 40% to 90%, 98% of payments are applied within 48 hours, and collectors spend 40% less time preparing for calls.

Worth knowing: Microsoft sells the tools it used, so read this as part showcase.

11. U.S. Venture: bank and card reconciliation

The workflow: Monthly reconciliation of credit card and bank statements.

What the AI does: Microsoft 365 Copilot for Finance matches and reconciles the statements.

Who stays in the loop: The finance staff who run the process. The case study doesn't go further.

What the company says happened: In Microsoft's case study (a vendor case study), a finance operations manager says it cuts about 80% off her reconciliation time, and the team saves more than 30 hours a month. Setup took four weeks.

Operations

12. Walmart: product catalog data

The workflow: Creating and cleaning the product data behind a huge online catalog.

What the AI does: Several large language models generate or improve product data fields.

Who stays in the loop: Not stated.

What the company says happened: On its August 2024 earnings call, CEO Doug McMillon said Walmart had created or improved more than 850 million pieces of catalog data, work he said would otherwise have needed nearly 100 times the current headcount.

Worth knowing: The 100x is an estimate of the alternative, not a measured saving.

13. UPS: customs clearance

The workflow: Clearing dutiable packages into the US after the de minimis exemption ended.

What the AI does: Agentic AI is part of the customs brokerage system that clears packages.

Who stays in the loop: People handle the packages the system can't clear.

What the company says happened: On its Q3 2025 earnings call, CEO Carol Tomé said volume needing clearance went from 13,000 packages a day in March to 112,000 in September, with 90% cleared without manual intervention, up from about 21%.

Worth knowing: She credited the result to UPS's technology as a whole, with agentic AI as one part.

14. Amazon: warehouse robot coordination

The workflow: Routing robots around Amazon's fulfilment centres.

What the AI does: DeepFleet, a generative AI model, coordinates how the robot fleet moves.

Who stays in the loop: Robots work alongside employees; Amazon says it has upskilled more than 700,000 people to work with robotics.

What the company says happened: Amazon says it improves robot fleet travel time by 10%. That was announced at launch, so treat it as the company's expectation.

15. A 25-person marketing agency: the sharing loop and client dashboard (my work)

Disclosure: I built the infrastructure here. The team wrote most of their own processes, which was the point.

The workflow: Getting a team's written-down processes shared, checked and into everyone's hands, and giving each client a live view of their work.

What the AI does: When a non-technical teammate says "share this with the team", their process is checked automatically, with hard checks and an AI review. Safe changes merge on their own. The client dashboard syncs hourly from the team's sheets and never overwrites good data with a bad read.

Who stays in the loop: Anything risky goes to a person. Each client sees only their own data.

What happened: 22 teammates set up, 16 shared team processes, at least 5 teammates besides the founder contributing their own, and the first fully automatic share-and-merge about 10 weeks in. The dashboard was live for the first client about 7 weeks after we started. No time-saved number exists. Case study here.

HR

16. IBM: employee HR questions

The workflow: Answering employee HR questions and handling routine requests such as letters, vacation and payroll access.

What the AI does: The AskHR virtual agent automates more than 80 HR tasks.

Who stays in the loop: AI handles routine questions; human advisors handle the complex ones.

What the company says happened: IBM reports a 94% containment rate on common questions, meaning they're resolved without a person.

Worth knowing: That's common questions, not 94% of all HR work. IBM also sells the platform.

17. Chipotle: frontline hiring

The workflow: Screening restaurant job applicants and scheduling interviews.

What the AI does: A hiring assistant built by Paradox chats with candidates, answers questions, collects details, schedules interviews and sends offers.

Who stays in the loop: Hiring managers decide who gets an offer.

What the company says happened: The CEO said in 2025 that the AI cut hiring time by 75% (CNBC, via NBC Washington). Paradox's own case study puts time to hire at 4 days, down from 12.

18. General Motors: interview scheduling

The workflow: Scheduling interviews across one to two million candidates a year.

What the AI does: A Paradox assistant handles candidate outreach and scheduling by text and chat.

Who stays in the loop: Recruiters keep the relationship work.

What the company says happened: In Paradox's case study (a vendor case study, undated), time to schedule an interview fell from more than five days to 29 minutes, saving $2 million a year.

Engineering

Engineering teams publish the most detail, and their examples show the pattern clearly: AI does the bulk, automated checks catch failures, people review.

19. Airbnb: migrating 3,500 test files

The workflow: Moving thousands of front-end tests from one testing library to another.

What the AI does: Airbnb's pipeline had an LLM rewrite each file, then run the tests, lint and type checks, retrying whenever a step failed.

Who stays in the loop: Engineers built the pipeline, tuned the prompts and finished the last 3% of files by hand.

What the company says happened: 75% of files migrated in four hours, and the whole job finished in six weeks. The original manual estimate was a year and a half.

20. Amazon: Java upgrades

The workflow: Upgrading internal applications to a newer Java version.

What the AI does: Amazon Q Developer generates the code changes.

Who stays in the loop: Developers review them; Andy Jassy said 79% of the AI-generated changes shipped with no further edits.

What the company says happened: In an August 2024 post, Jassy said the average upgrade went from about 50 developer-days to a few hours, saving an estimated 4,500 developer-years and about $260 million a year.

Worth knowing: Amazon sells this tool, and the savings are estimates.

21. Uber: automated code review

The workflow: Reviewing every code change for bugs, security issues and standards.

What the AI does: uReview comments on code changes. It covers more than 90% of the roughly 65,000 changes Uber's engineers make each week.

Who stays in the loop: Engineers rate each comment and decide whether to act on it.

What the company says happened: Engineers rate 75% of comments useful. Uber estimates it saves about 1,500 hours a week.

22. Stripe: coding agents that open pull requests

The workflow: Small and medium code changes requested by engineers.

What the AI does: An engineer asks in Slack or a web app, and a Stripe "minion" writes the code in an isolated environment, runs the checks and opens a pull request.

Who stays in the loop: Humans review the code before it merges.

What the company says happened: More than a thousand minion-written pull requests merged each week, as of February 2026.

Marketing

23. Klarna: images and copy

The workflow: Producing campaign images and marketing copy that agencies used to make.

What the AI does: Image tools generate visuals, and AI writes most of the copy.

Who stays in the loop: The image process still includes checks for brand consistency, quality and legal compliance.

What the company says happened: In a May 2024 release, Klarna said AI accounted for about $10 million a year in marketing savings, and the image production cycle went from six weeks to seven days.

24. Mondelez: social and retail content

The workflow: Making social media and product-page content for brands including Chips Ahoy, Milka and Oreo.

What the AI does: A generative AI tool built with Accenture and Publicis produces the content.

Who stays in the loop: A Mondelez executive told Reuters people will always check what the tool produces.

What the company says happened: Mondelez told Reuters in October 2025 that the tool cuts content production costs by 30% to 50%.

25. Unilever: product imagery

The workflow: Creating product images for each sales channel without new photo shoots.

What the AI does: Teams generate imagery from 3D digital copies of products, starting with brands like Dove and Vaseline.

Who stays in the loop: Not stated.

What the company says happened: Unilever's CMO told Marketing Dive in March 2025 that some content is now made twice as fast at half the cost.

What the good ones have in common

Read these side by side and the tool barely matters. Klarna and Lyft both run support, on different models. What separates the strong examples is how the work was set up.

The process was defined before the AI touched it. Airbnb's pipeline had an order: rewrite, test, lint, retry. Salesforce's sales agent got better when the data underneath got cleaner, not when the model changed. AI can follow a workflow. It can't follow a habit. I wrote about this in AI can only do work that's been defined.

A person approves where a mistake would be public. Vercel puts every email in Slack for approval. Morgan Stanley's advisors send their own follow-ups. Octopus staff review drafts. Klarna's walk-back shows what happens when the measure is cost and nobody owns quality. My rule: let AI act alone where a mistake is cheap, easy to spot and easy to undo. Customer conversations usually fail that test, which is why I build drafts, never sends.

Someone brought the people along. Most of these write-ups are about the tool. The quiet part is the team that had to change how it works. Several companies here note that people moved to more complex work rather than being replaced. In my experience the workflows that last are the ones where the worried people were heard out early. That's part of what I do, as a consultant and a mindset coach.

It reports back. Uber's engineers rate every comment. The startup founder gets a report after every run. When people can see what the AI did, they keep using it. When they can't, they quietly stop.

If you want help picking your first workflow, mapping it step by step and getting your whole team behind it, that's what I do as an AI workflow consultant. Start with one piece of work that happens every week and that one person does well.