AI for Small Businesses: Which Department Should You Start With?
The short answer
A small business should start with sales and customer communication — because that's where the gain shows up directly in revenue and results appear in week one. The order: (1) sales/customer communication (quote drafts, enquiry triage, fast replies), (2) marketing and content (social, product copy, visuals), (3) operations and admin (document reading, reporting, data entry), (4) accounting and finance (only with human review), (5) HR last — the discrimination and personal-data risk is highest there. All of it is doable without hiring anyone, without a separate budget, in the $100–300 a month range.
A small business's real constraint isn't budget — it's attention
In a corporate, an AI project belongs to a department; someone gets hired for it, a budget line opens. In a small business none of that exists. The owner is also the salesperson, the bookkeeper and occasionally the delivery driver. That's why every approach that begins with "let's build a strategy first" dies here.
The right question in a small business is: "Over the next two weeks, without hiring anyone, which task can I do less of and still get the same result?" The answer is almost always in sales and customer communication.
We work with Hedef Grup — a sales and distribution company with 60+ years in food, cleaning and personal care. Walk into a company like that talking about "digital transformation" and they'll show you the door, rightly. They have a system that works and has paid the bills for six decades. What worked there was taking one repetitive task nobody enjoyed and making it invisible. The word transformation never came up; the result spoke.
Department by department: what do you gain where?
| Department | First application | Time to result | Order |
|---|---|---|---|
| Sales | Quote drafts, follow-up emails, enquiry triage | Week 1 | Start here |
| Marketing | Content calendar, product copy, ad creative | Weeks 2–3 | Second |
| Operations | Reading documents, normalising order data, reporting | Months 1–2 | Third |
| Accounting & finance | Spotting cost anomalies, reading cash flow | Months 2–3 | Fourth — human review required |
| HR | Job ads, CV pre-screening, internal docs | Month 3+ | Last — highest risk |
Why did I put HR last? Because a wrong decision there can't be fixed by editing a paragraph: it can mean a person not getting hired, or personal data processed without consent. I studied HR, and that's exactly why I'm most cautious here.
The first 14 days: a concrete starting plan
Days 1–2: write down your own week
Don't hire a consultant, don't run a survey. Log the work you personally do for a week. If you're the owner, the three most annoying items on that list are the company's three most expensive tasks — because your hour is the most expensive hour.
Day 3: get one assistant subscription and give it real work
Don't evaluate on the free tier — paid business plans commit to not training on your data; free ones usually don't. Give it a real quote you wrote yesterday and ask it to rewrite it. You need to judge quality on your own work, not on sample prompts.
Days 4–7: hand over one task completely
Quote drafts, or triaging inbound enquiries. Just one. Write down how many minutes you saved each day — yes, a notebook is fine. That measurement becomes the basis of every decision that follows.
Days 8–11: bring in one more person
Pick the most curious person on the team — not the most senior. Show them your setup and ask them to adapt it to their own work. This step decides whether the thing survives while you're on holiday.
Days 12–14: write the rules — two pages is enough
What data never gets pasted (customer identity/financial data, signed contracts), which tasks can be delegated, who approves outgoing output. Writing it takes an hour and prevents the biggest accident you'd otherwise have.
We worked on the restaurant side with Pizza Station and Cookzania; Cookzania has branches in Istanbul, Sakarya and Izmir. Restaurants are one of the least theory-friendly sectors — once evening service starts, nobody cares about any system. What I learned there: in a small business a new tool has to ride on top of the existing habit, not try to replace it. If the team already lives in WhatsApp, your solution has to run through WhatsApp.
The four traps small businesses fall into
- Buying software without designing a process. A dashboard is purchased, three seats are opened, two months later nobody logs in. Without a process, software is just a cost line.
- The owner doing everything. You can build the first setup, but if it doesn't reach a second person it stays hostage to your calendar and dies in the first busy week.
- Pasting customer data around carelessly. Where personal data is involved this is a data protection matter, and the fine dwarfs the efficiency you'd gain.
- Carrying on without measuring. Six months pass on "it feels better". Track three numbers: time, correction rate, hours saved per month.
Should you hire a consultant?
Honest answer: not for the first 90 days. None of the steps in this article need one, and nobody from outside will understand your business better than you do. Consulting is genuinely worth it in two cases: when you're building a system over your own data (a security and architecture job), and when you're wiring multiple systems together (an integration job).
I'm saying this as an agency owner, and yes, this sentence costs me work. But a system built for a business that hasn't lived through the first stage itself doesn't stick — I've watched it fail too many times.
Key takeaways
- Start with sales: the gain shows up in revenue and the result arrives in week one.
- Leave HR for last — mistakes there aren't recoverable and personal-data risk peaks.
- A new tool must ride on the existing habit; one that tries to replace it goes unused.
- A setup that never reaches a second person is hostage to the owner's calendar — it won't survive.
- You don't need a consultant for the first 90 days; start by writing down your own week.
Frequently asked
Does AI make sense for a five-person business?
Yes — and the impact is bigger at five people than at five hundred, because the repetitive load per person is far higher. In a five-person team, four hours a week each adds up to 80 hours a month: half a headcount, without hiring.
My staff aren't technical — can this still work?
It can, but the setup differs: instead of putting the tool in front of the team, you hide it behind the existing flow. The team still messages on WhatsApp while classification and summarising happen behind it. "Everyone learn this new program" has the highest failure rate of any sentence in a small business.
What will it cost monthly?
For a typical 5–15 person business, $100–300 a month is realistic: an assistant seat per person and one automation tool. The only thing that pushes you above that band is a custom setup over your own data. I break the line items down in the budget article.
Are there government grants or support programmes?
In Turkey, KOSGEB and TÜBİTAK programmes support technology and R&D work — but they're aimed at the product or technology you're developing, not at buying off-the-shelf tools. Postuby was supported under TÜBİTAK's 1507 R&D Starter programme, and from that experience I'd say: these grants are for people building products. Don't apply just to buy tools; you'll spend the time for nothing.
My competitors aren't using it — should I rush?
Don't rush, but don't defer either. The advantage here isn't the kind that closes in six months; the real gap comes from whoever builds the habit first moving faster on the second and third application. Even if your competitor buys the same tool a year from now, by then you'll be three processes ahead.