AI Readiness Checklist for Small Business: What to Answer Before You Start

Last week a vendor demoed a tool that answers customer chats, builds sales reports and writes product captions. Your team loved it, and the price seemed fair. One doubt remains: what exactly the tool will do tomorrow morning, and who answers for its mistakes.

When it comes to AI for small business, readiness rarely depends on how tidy your data is. It depends on four things you can check in one afternoon: someone owns the process and knows its current numbers, the source documents are kept, a wrong figure can be traced, and every irreversible step has a human sign-off. If all four hold, you're ready to choose a tool. If not, fix those before spending.

Order matters, because buying AI and getting results from it are different things. BCG's global study found only 5% of companies capture AI value at scale, while 60% report little or no value despite investing (Source: BCG, 2025). In IBM's survey of 2,000 CEOs, 64% admitted buying some technologies out of fear of falling behind, before clearly understanding their value (Source: IBM Institute for Business Value, 2025).

In Indonesia, the interest is there; readiness hasn't caught up yet. Lark's research found nine in ten companies in Indonesia have worked to build a work culture that is open to AI technology, but only 19% rate their digital maturity as high (Source: Lark via Koran Jakarta, 2026). Readiness lags elsewhere too: in a survey of SMEs across 10 OECD countries, 67% of those not yet using generative AI are unsure how to use it or assess its risks (Source: OECD, 2025).

In our experience building systems for a finance team and a marketplace seller, the question that stalls a project is rarely “is the data clean?” Far more often it's “whose job is this step?” So each checklist item below is a question with a clear verdict: ready, fix first, or not a blocker.

Who owns the process, and what are its numbers today?

Pick one job you want AI to help with, say the daily order recap or bank reconciliation. Name one person accountable for the result, and note how long the job takes now and how often it goes wrong.

Without an owner, every problem bounces between staff, vendor and tool. Without a baseline, you'll never know whether AI is helping or just creating new busywork. Rough numbers are enough. Payback maths can wait.

Verdict: ready if you can write the name and numbers down now. Fix first if the answer is “everyone handles it” or “it feels slow”.

Are the source documents kept, even if messy?

Many owners think they need months of data clean-up before touching AI. Statements from several banks, card bills, supplier invoices, marketplace settlements and e-Faktur tax invoices rarely share a format. That's normal, and no reason to wait.

Our Finance AI Automation build doesn't wait for uniform formats. It reads nine bank-statement formats and seven card issuers, image-only scans included, checks every figure against its source document, and holds back unmatched amounts rather than rounding them (Source: Finance AI Automation case study, Agentiq). Built by Agentiq for an anonymised client; all screens shown use fictional sample data.

The real trouble is evidence that lives only in someone's memory or across three staff phones. Try a simple test. Imagine you must produce a week of orders and customer complaints tomorrow, including the transfer screenshots and QRIS receipts buried in WhatsApp groups. Pull it together within an hour and you're fine.

Verdict: not a blocker if the documents exist, however messy. Fix first if most of the evidence lives only in people's heads.

Can you trace a wrong number back to its source?

Online sellers feel this one most. Orders from Shopee, TikTok Shop, Lazada and Blibli often pass through a rented order platform and several connectors before being booked as revenue. When a figure is off, there's nowhere to look.

We found exactly that when building the Warehouse Management System. The client's rented platform dictated how the warehouse worked, couldn't be extended, billed monthly and was opaque. Its replacement is a system the client owns: order sync, a warehouse-floor app, shipping labels and finance reconciliation on one server. Built by Agentiq for an anonymised Indonesian marketplace seller: brand, order and tracking numbers, product and staff names and addresses are fictional, and every screen shown uses synthetic sample data.

Also ask whether a tool replaces a step or just adds another app. Lark found 49% of Indonesian respondents feel overwhelmed by the number of apps they manage, and 48% must check several platforms every hour to track work (Source: Lark via Koran Jakarta, 2026). Leaders see it too: 50% of CEOs in IBM's survey say their pace of investment has left them with disconnected, piecemeal technology (Source: IBM Institute for Business Value, 2025).

Verdict: ready if every important figure traces back to a document or system you can open yourself. When every answer comes back as “ask the vendor”, put this one in the fix-first column.

Which steps must never happen without a human sign-off?

A chat reply with the wrong price can be corrected. A supplier transfer, a deleted order or a posted journal entry is much harder to reverse. Before choosing a tool, mark the steps that can't be undone and decide who presses the button. In the finance system above, nothing becomes permanent until someone presses a button on a screen showing exactly what will be written.

Know which customer data the tool will touch and who approves it, since Indonesia's Personal Data Protection Law (UU PDP) already applies. National AI rules are still taking shape: the Presidential Regulations on the National AI Roadmap and AI Ethics await the President's signature (Source: Infobank News, 2026). Once signed, each ministry is expected to write its own sector rules (Source: ANTARA News, 2026).

Verdict: ready if the critical steps and who owns each are written down. Unfinished regulation isn't a blocker; not knowing what data will be processed is what to fix first.

AI readiness for small business: what do your answers mean?

If every answer is “ready”, pick one process and measure it. Where an answer says “fix first”, the fix rarely needs a technology budget: naming a process owner, logging hours for a week, or creating a shared folder for transfer receipts.

For an established business, the bigger risk isn't starting late. It's paying for a tool whose results can't be judged, and joining BCG's 60%. The upside is real: in the same 10-country OECD survey, 91% of SMEs using generative AI report efficiency gains (Source: OECD, 2025). Preparation makes the difference, including preparing people. Some 86% of Indonesian employees say they need AI and cybersecurity training, yet only 36% say they were properly prepared (Source: Lark via Koran Jakarta, 2026).

• Write down your most time-consuming job, who owns it, and the hours it takes in a week.
• Gather a week of evidence, from WhatsApp chats and bank statements to marketplace exports, in one shared folder.
• Mark the steps that can't be undone and who is authorised to approve them.
• Ask each vendor which step their tool replaces and how wrong figures get traced.

If you'd like to see what this checklist looks like for your business, tell us about your most troublesome process through our contact page. The Agentiq team will help you sort what's ready from what needs fixing first, with concrete recommendations, not a template.