AI for Small Businesses: How to Use AI Without the Hype (and Without Breaking Trust)

There’s a weird gap in the AI conversation.

On one side, you’ve got big promises: automation, transformation, “work smarter”.
On the other, most small business owners are staring at their day-to-day thinking: That sounds impressive… but what does it actually do for me?

In this episode of What One Thing, we spoke with Ian Lockwood about a very specific use case: using an AI chatbot to reduce inbound call pressure during peak season, without hiring extra staff, and without turning the whole business upside down.

The value was in the intent behind this project.

Here are the key lessons, and how to apply them in a way that stays practical, safe, and genuinely useful.

Lesson 1: Start with the pressure point, not the tool

AI becomes a distraction when it starts with curiosity about a tool:
“Ooh, I wonder what this can do?” is a recipe for a rabbit hole with zero ROI.

It becomes useful when it starts with a bottleneck in your business:
“What’s currently costing us time, energy, or momentum and how can we improve that?”

In Ian’s case, the trigger was simple: peak season was approaching and inbound calls were likely to overwhelm the sales team. The goal wasn’t “do AI”. The goal was to take load off the phones and improve responsiveness.

Practical tips

Before you start: pick one operational problem that creates friction every week, write a one-sentence outcome ("we want customers to get answers faster without tying up the team"), then use that to filter your options. If it doesn't solve the problem, ignore it.

Lesson 2: The first version can be simple, and still valuable

One of the biggest misconceptions is that AI requires a major project.

Ian described a spectrum: from tools that crawl your website and produce a basic chatbot, all the way to complex, flowchart-heavy systems that cater for every scenario.

For most small businesses, the “basic but solid” option is the right start, especially if your goal is to answer FAQs and reduce low-value contact.

Practical tips

Choose a tool that can learn from your website and a small set of uploaded documents.

Treat the first launch as a test, not a forever system.

Don’t try to do everything at the start, aim for "answers common questions well" before you try "integrates with everything."

Lesson 3: Guardrails aren’t optional, they’re a MUST

AI can hallucinate. Worse, it can hallucinate confidently and convince others it’s true.

Ian gave a real example about a customer who complained at 8pm that a delivery hadn’t arrived, and the chatbot promised it would be rebooked for 10am the next day, despite having no access to delivery systems at all.

That’s the risk if you don’t set clear boundaries to AI that’s speaking on your behalf; it will do whatever it takes to please the customer, and it might not always be possible.

Practical tips

Before you go live, list your high-risk topics: delivery promises, pricing specifics, anything you can't afford to get wrong. Add clear boundaries to the prompt: what it can and cannot claim, and when to hand off to a human. And check those chat logs regularly so you can nip any problems in the bud.

Lesson 4: Integration is where complexity (and value) ramps up

A chatbot that answers questions is useful. A chatbot that checks stock, supports quotes, or interacts with other systems can be transformative, but that’s where you move into deeper implementation.

Ian explained that live, dynamic information (like stock status) isn’t reliable if the chatbot is only trained on static website content. At that point, you need a live link to your systems and often development support.

Ian used Zapier to trigger emails to the sales team when a quote request had everything it needed: name, email, phone number, and the details of what the customer wanted. There was no phone call required or any chasing for basic information.

Practical tips

Separate phase one (answers questions) from phase two (does things).

If you're integrating, define the trigger, the data needed and the outcome before you build anything.

And budget time for testing, because integrations are where mistakes get expensive.

Lesson 5: Your knowledge base is the hidden bottleneck

The most underestimated part of AI implementation isn’t the widget on the website. It’s the knowledge base: what you feed it, how you structure it, and whether it can retain context when working with large volumes of data.

Ian ran into size limits on several platforms and learned something important about how AI handles big structured data sets: it chunks information up due to token limits, and when it does, it can lose track of what the column headings were. A spreadsheet with 8,000 rows becomes unreliable fast if each chunk doesn't know what it's looking at.

Practical tips

Before you build anything, ask: “Do we already have the information in a usable format?”
Prioritise the pages/documents that answer real customer questions (not everything you’ve ever written).
If you have large structured data, ensure it keeps context (e.g., include labels so meaning doesn’t disappear).

The bottom line: AI works best when it protects your people

This conversation wasn’t about replacing a sales team. It was about making them more effective, reducing queues, removing repetitive “it’s on the website” questions, and freeing humans up for the real work: judgment, nuance, relationships, and the out-of-the-ordinary cases that need a human.

Ian's business is up 20% year on year, and call volume is down roughly 20%, with the same team and no additional headcount.

The best mindset shift you can adopt is this: AI isn't a strategy, it's a tool. And the best tools don't change who you are, they just remove enough friction that you can do more of what matters.

Brought to you by Corbar Accounting and Affirm IT Services Ltd.