In São Paulo, a fashion brand called EZILDINHA sells elegant linen and silk dresses to customers across Brazil. It has no marketing department, no media-buying agency, no in-house copywriters, and no data team. What it has is a small group of people and an operating layer made of AI. The product descriptions, the SEO, the editorial blog, the Google Ads, the Instagram and Facebook campaigns, the email calendar — the work that a competing brand would staff with a dozen specialists — runs through Claude and ChatGPT, reviewed by a human before it ships.
That arrangement would have been impossible three years ago and unremarkable to describe today, which is the whole point. The most consequential shift in small business has nothing to do with chatbots answering trivia. It is that the operational work which used to require either the owner’s nights and weekends or a payroll the business could not afford has become a software layer any disciplined operator can run.
The short version: Small businesses worldwide are adopting AI faster than the coverage suggests, and the smallest operators have the most to gain, because in a one- or two-person business the entire operational burden sits on one or two people. This briefing tells the story through real, sourced companies on five continents — from a Brazilian fashion label to a Polish support-software firm to an Italian payments startup to sellers across India and Indonesia — shows the exact tools behind those stories, names the honest failures, and lays out a 90-day plan an owner can run without hiring anyone.
It is a story about leverage, and it starts with one operator.
The one-operator brand, up close
Return to EZILDINHA, because the brand is the archetype this entire page is about. Strip its operation to the parts that consume a small-business owner’s week and you find the five jobs every small business runs: getting found, making the product look and read well, talking to customers, handling the back office, and deciding what to do next. EZILDINHA runs the first three almost entirely through AI, with a person reviewing the output.
Getting found is the hardest job for a brand with no ad budget to waste, and it is where the leverage shows most. The brand researches keywords, drafts on-page copy, structures its product data, and builds content designed to be cited by AI answer engines, not only ranked by Google — the discipline that decides whether a small brand is visible at all in a market where customers increasingly ask an assistant before they scroll. The paid side, the part most owners assume requires an agency, runs the same way: AI drafts and iterates the Google and Meta campaigns, and a human approves the spend.
None of this makes EZILDINHA unusual in Brazil. The country’s small-business agency, Sebrae, found in a national survey that 44% of Brazilian small businesses already use some form of AI and 51% use generative-text tools (Agência Sebrae). What makes the brand instructive is not that it adopted AI. It is that it built an operating system around it, deliberately, with judgment in the loop — and now competes for attention with brands that spend many times what it does.
The stack behind the story
Stories like EZILDINHA’s are possible because a layer of tools now does, for a few dollars a month, work that used to require staff. Four examples, each a real company with a documented result, show what that layer actually does. These are the tools The Leverage Years teaches operators to use, which is why their results are worth reading closely.
The first is Tidio, a Polish customer-service software company whose “Lyro” agent is built on Claude. Anthropic’s customer account reports that Tidio automated 71% of its own support and that merchants using Lyro resolve up to 90% of inquiries without a human (Anthropic, vendor-reported). For a small store, that is the difference between answering tickets at midnight and answering only the ones that need a person.
The second is Reversia, a European Shopify app, also built on Claude, that translates an entire store — product copy, collections, SEO tags — across more than 110 languages, with the merchant setting a brand prompt so the translations keep the store’s voice (Anthropic, vendor-reported). International expansion, once a five-figure agency project, becomes a setting. It is the precise capability a brand like EZILDINHA needs to sell beyond its home market.
The third sits underneath millions of small stores. Shopify’s Sidekick assistant, which answers a merchant’s questions in plain language and guides setup, SEO, and analytics, runs on Claude; Shopify has reported AI-attributed orders across its platform rising roughly elevenfold over 2025 (Anthropic / Shopify, vendor-reported). The fourth, Triple Whale, a US analytics tool for direct-to-consumer brands, switched from a competing model to Claude and reports one customer reviewing 600 marketing assets in ten minutes against a prior five hours (Anthropic, vendor-reported).
Read those four together and the EZILDINHA story stops looking like a one-off. A small operator now has, available off the shelf, a support agent, a translator, a store co-pilot, and an analyst — the functions a larger competitor pays salaries for.
The same pattern, around the world
The EZILDINHA effect is not Brazilian, and it is not limited to fashion. Across very different markets, small and mid-sized operators are using AI to reach customers and run operations at a scale their headcount could never support.
In India, Flipkart has built AI tools specifically for its small-town sellers, including seller dashboards that deliver demand and pricing insights by voice in Hindi and regional languages — a deliberate effort to bring vernacular-speaking micro-merchants into ecommerce (Business Standard). In Indonesia, Tokopedia gives its merchants AI tools for pricing, descriptions, and product photos, and reported that using Google’s Vertex AI to improve listing data lifted the number of unique products listed by about 5% (Google / Tokopedia, vendor-reported). In Italy, the payments startup Satispay reports its engineers now write roughly 75% of their code with Claude, freeing a small team to move at the pace of a larger one (Anthropic, vendor-reported).
And in the United States, OpenAI’s small-business stories document the same thing at street level: an 86-year-old salvage yard in Reno that built a thousand-item inventory system in an afternoon and troubleshot a failing machine on the spot; a third-generation tamale company in California whose co-founder, who learned English as a second language, drafts staff communications in Spanish and delivers them confidently in English (OpenAI). In a separate program, OpenAI documented a San Francisco skate shop building a catalog-to-Shopify workflow that targeted a task eating about ten hours a week, and a Detroit caterer saving roughly two hours a week on inventory (OpenAI Small Business Jam).
Different continents, different trades, one pattern: the owner uses AI to carry the work that was keeping them from the work only they can do.
How big is the shift, really?
The anecdotes are backed by survey data that has moved fast. The U.S. Chamber of Commerce, surveying nearly 4,000 firms, found 58% of small businesses now use generative AI, up from 40% a year earlier, and that 82% of AI-using small businesses grew their workforce in the past year — adoption that accompanied hiring, not layoffs (U.S. Chamber of Commerce). Salesforce’s survey of 3,350 SMB leaders across four regions found that, among those using AI, 91% say it boosts revenue and 87% say it helps them scale (Salesforce, vendor survey). The Brazilian figure from Sebrae — 44% — shows the wave is global, not Silicon Valley’s alone.
More operators, more places
The pattern holds across trades and continents, and a few more short cases make the range concrete. In Houston, the owner of a power-washing startup built a custom assistant for customer sourcing and email because he could not afford to outsource the writing. In New York, a one-person consulting firm built an assistant that reviews fifty-page RFPs and runs a go/no-go checklist, saving 30 to 45 minutes per proposal. In San Francisco, the owner of Oren’s Hummus, a first-time AI user, now reaches for it almost daily, “created multiple uses in minutes for what would have taken hours” (OpenAI Small Business Jam).
The same leverage scales into mid-sized operators that started small. In Brazil, Magazine Luiza’s long-running “Lu” assistant now runs on generative AI for product suggestions and support over WhatsApp, and the retailer reported Google’s AI delivering a 27% lift in return on ad spend (Google / Magalu, vendor-reported). In Africa, Jumia tied AI chatbots and AI-driven workflows to a roughly 25% drop in site bounce rates and to its push toward profitability (TechCabal). The lesson a small operator should take is not the scale of these firms but the direction: the work that used to require people now scales with software, which is exactly why a brand with five employees can behave like one with fifty.
What it changes, in plain terms, is the shape of the business. An owner who reclaims an hour a day buys back a full working day each week, and that day goes to the buy, the relationships, and the decisions that actually grow the business. The first employee an owner hires can be a salesperson rather than an admin, because the admin already runs as supervised software. The business can stay lean longer, reinvest more, and hire deliberately rather than out of exhaustion. That is the quiet, compounding advantage the disciplined operators are banking while their competitors are still doing it all by hand.
The five functions, and where AI actually helps
Behind the stories sit five kinds of work every small business runs. Here is what AI does well in each, and the line where the human stays in charge.
Marketing, content, and search is where most owners start, because it eats the most nights. The roaster Henry’s House of Coffee in San Francisco uses AI for product descriptions, SEO, and email and calls it “a game-changer” that lets the owner focus on roasting (U.S. Chamber). The trap is shipping the model’s default voice; the fix is a written voice file and a human edit on everything that goes out.
Customer service is the second function owners hand over, and the good ones use it to answer faster, not to hide behind a bot. Tidio’s Lyro, on Claude, shows the ceiling: most routine inquiries resolved automatically, the hard ones routed to a person. The line is simple — constrain the assistant to your real policies and prices, and never let it guess at an answer it does not have.
The back office — invoicing, payroll, month-end, contracts — was the last to benefit, because it requires acting in real systems rather than writing text. That changed in 2026, when Anthropic launched a small-business mode connecting Claude to QuickBooks, PayPal, HubSpot, and Google Workspace with a strict approval gate (Anthropic). The rule there is the right rule everywhere: the assistant prepares the work; the owner approves anything that sends, posts, or pays.
Operations and institutional memory is the quiet winner. The Reno salvage yard turning decades of parts knowledge into a searchable system is the small-business version of capturing what usually walks out the door when a long-time employee retires. The numbers are the fifth: turning the business’s own sales data into a purchasing or staffing decision, as the Detroit caterer does. Here the owner’s judgment is least replaceable — AI surfaces the pattern; the person carrying the risk makes the call.
The honest failures — because the wins are only half the story
A page that only counted successes would be marketing for AI, not a useful account of it. Two cautionary cases belong on every owner’s desk.
The first is Klarna, the Swedish payments company, whose AI customer-service assistant was the most-cited deployment in the world: in early 2024 it reported handling two-thirds of service chats, the work of 700 agents, with resolution times falling from eleven minutes to under two (Klarna). The instructive part came later: in 2025 Klarna walked back its AI-only posture and re-hired people, guaranteeing customers a human option (Customer Experience Dive). The lesson for a small business is to use AI to raise the floor on service, not to remove the human a customer sometimes needs.
The second is Anthropic’s own “Project Vend,” in which it let Claude autonomously run a small office shop. It managed a real operation and also made real mistakes — mispricing, inventing discounts — which is exactly why its small-business product keeps a human approving consequential actions (Anthropic). Treat an AI agent as a capable assistant that needs supervision, not an employee you stop checking. The owners who get this right are the ones who never confused fast with finished.
The operating system an owner should build
The owners who get durable value do not have better prompts than everyone else. They have a small, written operating system — three documents that take an afternoon and turn a powerful tool into a safe, repeatable one.
The first is a voice and standards file: how your business sounds, what it promises, what it never says. This is what keeps AI-drafted marketing and customer messages sounding like you instead of like the model’s default, and it is the single thing that separates leverage from generic noise. The second is a never-upload list: one page naming what never goes near a model — customer financial data, anything that identifies a person, confidential terms. Half of small-business owners told Anthropic that data security was their top hesitation, and this is the answer to it. The third is a review gate: a short checklist run before anything publishes, sends, or pays — is it accurate, does it sound like us, did the model invent anything, would I sign my name to it.
With those three in place the tools become interchangeable and the leverage compounds. Without them, an owner keeps buying subscriptions and wondering why the hours never come back.
The first 90 days
The path from zero to a working system is short and sequential, and the most common way to fail is to try to do everything at once.
Weeks 1–2: pick the single function costing you the most hours or the most sleep — usually marketing or the back office — and start there. Weeks 3–4: write the three files. Weeks 5–9: run that one function on AI every working day, reviewing every output through the gate, and track one number so you know it is working. Expect the first week to feel slower and the third to feel like a different business. Weeks 10–13: add a second function, and write the workflow down so it survives outside your head. A small business with its AI operating system documented has built an asset that makes it more valuable and less dependent on any one person.
The bottom line
The story of small business and AI is not about replacement, and it is not a distant future. It is about an owner in São Paulo, or Reno, or Pacoima, who was doing the work of six people and now does it with a capable assistant carrying the first draft of all of it. The adoption is real and global, the tools are cheap, and the interface is a conversation. The only scarce input left is the operating discipline to use them well — and that, unlike budget or headcount, is within reach of anyone willing to start small and build deliberately. EZILDINHA is not an exception. It is a preview.
Frequently asked questions
How are small businesses around the world using AI?
Across marketing and search, customer service, the back office, operations, and decision-making. Real examples span continents: EZILDINHA (Brazil) runs its full marketing operation on AI; Flipkart (India) gives small-town sellers voice AI in regional languages; Tokopedia (Indonesia) lifted unique product listings ~5% with AI; Satispay (Italy) writes ~75% of its code with Claude; and US firms from a Reno salvage yard to a San Francisco coffee roaster use it for inventory, content, and customer communication. The common pattern is AI drafting and carrying volume while a human keeps judgment.
How many small businesses actually use AI?
More than most people assume, and rising fast. The U.S. Chamber of Commerce found 58% of US small businesses use generative AI, up from 40% a year earlier, and that 82% of AI-using firms grew their workforce. Brazil's Sebrae found 44% of small businesses already use some form of AI. Salesforce found 91% of AI-using SMBs report a revenue boost. Adoption is mainstream and global, not experimental.
Can a one-person business really compete with bigger companies using AI?
Yes, and that is the core shift. AI gives a single operator off-the-shelf access to a support agent (Tidio), a translator (Reversia), a store co-pilot (Shopify Sidekick), and an analyst (Triple Whale) — the functions a larger competitor pays salaries for. EZILDINHA, a small Brazilian fashion brand, runs marketing, SEO, and paid ads on AI and competes for attention with brands spending many times more. The barrier was never the work; it was the cost of the work, which has fallen sharply.
Is it safe to use AI with my business and customer data?
It is safe if you set a boundary. Half of small-business owners cite data security as their top hesitation. The answer is a one-page “never-upload list” naming what never goes near a model — customer financial data, anything identifying a person, confidential terms — plus a human approval step before anything sends or pays, the model used by Anthropic's small-business product. Used with those guardrails, AI is a controllable assistant, not a leak.
What are the risks, and what went wrong for others?
The two best-known cautions: Klarna celebrated an AI assistant doing the work of 700 agents, then walked back its AI-only stance in 2025 and re-hired people to guarantee a human option. And Anthropic's own “Project Vend” let Claude run a shop autonomously and produced real mistakes. The lesson is to use AI to raise the floor and keep a human approving consequential actions — never confuse fast with finished.
How should an owner with no time and no tech background start?
Pick the one function costing you the most hours, write three short files (a voice and standards file, a never-upload list, and a review gate), then use AI for that one function every working day for a month, tracking one number. Add a second function only once the first runs cleanly, and write the workflow down so it survives outside your head. The interface is a conversation in plain language; the skill that matters is knowing your business well enough to brief and judge, which an experienced owner already has.