AugmaAI to grow
Same tools, three different starting lines.
A shop owner, a four-person startup and a company of thousands can open the same AI app on the same morning. What they should do with it is completely different. Here is what changes for each, where it helps, what the research measured and how to spend the first month. Every figure links to the page it came from.
The owner does five jobs. AI takes the paperwork off three of them.
In a small business the bottleneck is the owner's time. AI rarely replaces anyone here; it gives back the evenings spent answering the same questions, writing posts and chasing invoices.
Where it helps
- Replies on LINE and chatDraft answers to price, stock and opening-hour questions; a person sends.
- Marketing contentCaptions, product descriptions and a week of posts from one photo shoot.
- BookkeepingRead receipts and bank lines into categories for the accountant to check.
- Quotes and documentsTurn a chat request into a quotation or contract draft in your own template.
- InventoryFlag slow movers and suggest reorder points from past sales.
- TranslationAnswer tourists and overseas buyers in their language.
What the numbers say
- 31%of SMEs surveyed across OECD countries have generative AI in use, from 24% in Japan to 39% in GermanyOECD, Generative AI and the SME Workforce, 2025 ↗
- 57%of non-adopters say it doesn't suit their work; 50% cite a lack of staff skillsOECD, Generative AI and the SME Workforce, 2025 ↗
- 54%of SMEs using AI report at least moderate value; 6% call it transformationalOECD D4SME Survey, 2026 ↗
Realistic savings are modest: studies cited by the OECD put time saved at 2.8% to 5.4% of work hours among users. Small, but for an owner that is an hour or two a week.
The first 30 days
- 1
Week 1
List the ten questions customers ask most, and time how long replies take.
- 2
Week 2
Write the true answers once: prices, policies, hours. That is the assistant's source.
- 3
Week 3
Let it draft replies and posts; you approve every one before it goes out.
- 4
Week 4
Compare reply time and errors with week 1. Keep it, fix it or drop it.
The common mistake
Letting it talk to customers unchecked
One confident wrong price on LINE costs more than a month of saved time. Keep a person on send until the answers prove themselves.
Small teams, more ideas tested before the runway ends.
For a startup the scarce thing is time to learn whether anyone wants the product. AI writes much of the first code, the prototypes and the paperwork, so a few builders can test more ideas before the money runs out.
Where it helps
- Coding assistantsAgents write and refactor code; builders review, test and decide.
- Prototypes in daysClickable versions to put in front of users before a full build.
- User researchSummarise interviews and support tickets into themes.
- Testing and QAGenerate tests and catch regressions on every change.
- Fewer early hiresSupport, docs and ops drafts covered before the first specialist hire.
What the numbers say
- 25%of Y Combinator's Winter 2025 startups have codebases about 95% AI-generated, by highly technical foundersYC's Jared Friedman, via TechCrunch, Mar 2025 ↗
Speed is not the same as quality. The YC founders quoted could have written the code themselves; knowing what good code looks like is what lets them review it.
The first 30 days
- 1
Week 1
Pick one coding assistant for the whole team and agree how changes get reviewed.
- 2
Week 2
Add tests and checks the agent must pass, so speed doesn't become debt.
- 3
Week 3
Build one throwaway prototype and put it in front of five real users.
- 4
Week 4
Count what shipped and what broke. Decide what the next hire really needs to be.
The common mistake
Shipping code nobody understands
Generated code is cheap to write and expensive to own. If no builder can explain a change, it isn't finished.
Pilots are easy. Scaling is the work.
A corporate already has the data, the processes and the people. What AI changes is how knowledge moves: answers found in seconds, routine steps run by agents, and decisions made on numbers that used to take a week to gather.
Where it helps
- Knowledge searchAsk policies, manuals and past work in plain language, with sources.
- Agents for routine stepsTriage tickets, reconcile records, route approvals, with a person signing off.
- Customer serviceResolve common requests at any hour, hand the rest to people.
- AnalyticsBriefings drawn from many feeds, read before the meeting starts.
- GovernanceAccess rules, audit logs and a register of every model in use.
- Training at scaleTeach every team the same safe way of working.
What the numbers say
- 25%of AI initiatives delivered the expected return, say 2,000 CEOs in 33 countriesIBM CEO Study, May 2025 ↗
- 16%of AI initiatives have scaled enterprise-wideIBM CEO Study, May 2025 ↗
- 50%of CEOs say fast investment has left them with disconnected, piecemeal technologyIBM CEO Study, May 2025 ↗
Klarna's own figures are a company's claims about its first month, not an independent study; they show what a well-scoped service job can look like, not a guaranteed result.
The first 30 days
- 1
Week 1
Name an owner and one process with a measurable baseline and a clear cost.
- 2
Week 2
Agree the rules with security and legal: data, access, logging, disclosure.
- 3
Week 3
Run it with the team who does the work, people approving every action.
- 4
Week 4
Report against the baseline and write the plan to scale, or to stop.
The common mistake
A hundred pilots, no platform
Half of the CEOs IBM asked already describe piecemeal technology. Shared data, shared rules and fewer, deeper projects scale; scattered experiments don't.
Side by side
Three organisations, three different questions.
From our own work
On the Bangkok flood map, K.A.R.A, the AI we built, writes briefings from 13 feeds. Builders still decide what it may say.
Wherever your starting line is, the first step is small.
We're builders who use AI every day. If you'd like to think out loud about which job to start with, we're glad to listen.