Decisions4 min read
Five practical AI automations for a small business, and the ones I would not automate yet
By Sannan MalikPublished
Short answerFive AI automations suit most small businesses: sorting enquiries and drafting replies for staff to approve, pulling fields from invoices into the accounting system, turning call and meeting notes into CRM entries, qualifying and routing leads, and answering staff questions from internal documents. Each one keeps a person checking the output. I would not yet let AI decide on money, give legal or medical judgement, or message customers unreviewed.
Five automations where a person still signs off
Every one of these takes messy input, such as an email, a PDF or a transcript, and turns it into something a member of staff can check in seconds. That is the shape I look for. The AI does the reading and the first draft. A person does the deciding.
The third column matters most. If nobody can say who checks the output, the automation is not ready to build.
| Automation | What it does | What a person still checks | What it needs connected |
|---|---|---|---|
| Enquiry sorting and draft replies | Reads each incoming email, form or chat message, tags the topic and urgency, and drafts a reply in the customer's language | Reads and edits every draft, then presses send | Your shared inbox or chat channel, plus a short library of your approved answers |
| Invoice and document extraction | Pulls the supplier, date, amounts, line items and reference numbers out of PDFs and scans | Fields the system marks as unsure, and any total that does not add up, before posting | The mailbox or folder invoices arrive in, and your accounting or ERP system |
| Call and meeting notes into the CRM | Turns a transcript into a short summary, agreed actions and follow-up dates on the right contact | The summary and the actions, especially names, prices and dates | Your call recordings or meeting transcripts, and the CRM |
| Lead qualification and routing | Checks new leads against your criteria, removes duplicates and assigns each one to the right salesperson | Leads near the cut-off, and any lead the rules reject | Your web forms, the CRM and a list of who covers which region or service |
| Staff questions from internal documents | Answers questions such as leave rules or how to raise a purchase order, and shows the document it used | Staff open the linked source before acting on anything that matters | Your policy documents, manuals and price lists, kept in one maintained place |
What my two AI projects show about these five
Lead routing looks simple. It is mostly data hygiene.
On AI-Powered Lead Discovery, where my role was system architecture and engineering, the platform does source monitoring and research workflows, and next to them sit data validation, deduplication and permission and suppression controls. Those controls keep outreach consent-aware. A small business routing its own web leads needs a lighter version: no duplicates, and a record of who asked not to be contacted.
AI Voice Operations, where I led the AI system architecture, handles inbound and outbound call workflows in Urdu and English, with speech-to-text and conversation logs. The logs are the part I would copy. Whatever you automate, store the input, the AI output and what the person changed. That is how you find out whether it works.
Decisions I would not hand to AI yet
Some work should stay with people for now. The test I apply: if a wrong answer costs money, health or legal standing, or reaches a customer before anyone has read it, the AI drafts and a person decides.
- Approving payments, refunds, discounts or credit. AI prepares the case; a person approves it.
- Pricing a quote or changing a contract term.
- Legal judgement, such as whether a clause is acceptable or a dispute is worth pursuing.
- Medical or clinical judgement of any kind, including triage advice to patients.
- Hiring, firing or disciplinary decisions about staff.
- Any message, email or post that goes to a customer without someone reviewing it first.
- Deleting or overwriting records in your accounting system or CRM.
My rule for picking the first one
High volume, low risk, easy to check. All three.
Volume means the task happens many times a week, so a saving shows up quickly. Low risk means a mistake is caught before it does harm. Easy to check means a person can confirm the output faster than producing it from scratch. Invoice extraction usually passes all three: the totals either match or they do not. A staff question bot is harder to check: a confident wrong answer about a policy travels before anyone opens the source.
So for most small businesses I would start with enquiry drafting or invoice extraction. Run it alongside your staff first, with every output reviewed, and compare its work with theirs on your own real emails or invoices. Only after that would I let any step run without approval, and only the low-risk ones.
How an automation project is priced
My app packages have fixed prices: Launch at $999, Growth at $1,699 and Scale at $2,500. Custom AI systems are not part of them. They are quoted separately, because the price depends on how many workflows you want, which systems they connect to, how much data passes through and how accurate the result has to be. Hosting, model usage and other third-party fees are not included either.
Send me the one task you would automate first, with a few real examples of the input. I will tell you whether it passes the three-part test, and give you a written scope and a fixed quote.
Questions
- Do I need clean data before starting an AI automation?
- Not perfectly clean, but you need one place where the source lives. If your approved answers or policies sit in five versions across inboxes, sort that first. The AI can only be as current as the documents it reads.
- Can these automations run on our own servers?
- For many tasks, yes. Some language and speech models can be self-hosted when data must stay in-house. It costs more to set up, so I decide this task by task during scoping.
Sannan Malik is a software architect and product builder with 8+ years in software, CEO of Alpha Notic FZC and a Certified Scrum Product Owner. He builds mobile apps and business software for companies in the UAE, UK and US.
