The truth about AI chatbots in 2024
Most people close a chat window the moment they see a bot icon. It is usually because the software tries to act like a person but cannot answer a basic question about a delivery. We focus on building tools that do one job well instead of ten jobs poorly.
Why 'smart' bots usually fail small businesses
In January 2024, our team of 5 in Cambridge audited 47 different retail bots used by UK-based online shops. We found that 38 of them used complex language models that tried to be friendly but failed to provide factual data. If a customer in Ely asks if a specific plumbing part is in stock, they do not want a poem about pipes. They want a yes or no. These large models often guess the answer, which leads to angry phone calls and extra work for your staff. (Honestly, most of this high-end AI is just a fancy spreadsheet that talks back, and it usually gets the numbers wrong.)
We noticed that businesses often spend over £2,400 on setup for these complex systems. Within 3 months, they usually turn off the 'chat' feature because it creates more mess than it clears. A local shop does not have the time to train a machine on 10,000 pages of manual text. You need a system that knows your 43 core products and your 3 shipping zones. When the software tries to be too clever, it misses the simple stuff, like your actual opening hours on a Bank Holiday Monday. We prefer logic that follows a straight line.
A local shop does not have the time to train a machine on 10,000 pages of manual text. Logic should follow a straight line.
The 83% rule for customer queries
Between March and July 2024, we tracked 156 projects where we replaced 'chatty' bots with simple rule-based tools. We discovered that 83% of all customer questions in retail and back-office settings are about the same four things: order status, invoice copies, opening times, and returns policy. You do not need a brain the size of a planet to handle these. You need a tool that can look at your database and pull out a tracking number in 3.2 seconds. This saves your office team roughly 14 hours a week on repetitive emails.
By narrowing the focus, you eliminate the risk of the bot saying something stupid. If the bot only has buttons for 'Where is my order?' and 'I need an invoice', the customer cannot get it lost in a conversation about the meaning of life. One client, Fenland Trade Parts, saw their email volume drop from 62 messages a day to 19 after we installed a simple data-pull bot. This allowed their head of admin, Sarah, to actually finish her Friday reports by 15:00 for once. It is about reclaiming time for boring but necessary tasks.
Comparing costs: complex vs. simple
Let's be upfront about the money. A complex bot often carries a monthly 'platform fee' of £150 or more, plus the cost of the tokens used for every reply. Over a year, a small business might spend £3,100 just to keep the lights on for a bot that half their customers hate. Our approach at TTM Bot is different. We usually charge a one-time setup fee of roughly £940 for a standard rule-based bot that connects to your existing software. There are no hidden fees that grow as you get more customers. It stays the same price whether you have 10 enquiries or 470.
Maintenance is another hidden drain. When a complex AI updates its 'logic', it can change how it talks to your customers overnight without warning. A rule-based bot only changes when you tell it to. If you change your Saturday hours from 10:00 to 09:00, you update one line of code and it is done. We think tech should behave like a reliable hammer—it does exactly what you expect every time you pick it up. We have seen too many Cambridge firms get burnt by 'next-gen' promises that end up costing more than a part-time employee.
Tech should behave like a reliable hammer—it does exactly what you expect every time you pick it up.
How we built the Fenland Trade bot in 11 days
In September 2024, Fenland Trade Parts came to us with a problem. They had 83 ongoing accounts and only 2 people in the office. They were spending 4 hours every morning just re-sending PDFs of invoices to customers who had lost them. We didn't give them a bot that could talk; we gave them a bot that could search. We connected their accounting software to a simple interface. Now, a customer types in their account number and postcode, and the bot shows them their last 5 invoices with a 'Download' button. It took us exactly 11 business days from the first call to the launch.
The result wasn't a 'transformation'—it was just a better Tuesday. The office staff stopped dreading the 09:00 login. They recovered about 22% of their working day. This is what we mean by practical automation. We don't want to change how you do business; we just want to remove the bits that make your eyes glaze over. By the way, we did this without using any 'cloud-learning' or expensive subscriptions. It is a local tool for a local firm, built right here in Cambridge at our Innovation Way office.
What to ask before buying any bot
Before you sign a contract with any tech firm, ask them three specific questions. First, ask if the bot can actually pull data from your specific CRM or accounting tool. If they say 'we can build a connector', that means it will take months and cost a fortune. Second, ask what happens if the bot gives a wrong price to a customer. Who is liable? Third, ask for the total cost over 2 years, including all hidden API fees. If the numbers are round like '£5,000', they are probably guessing. Look for specific quotes like '£1,140 plus £20 monthly hosting'.
At TTM Bot, we are happy to tell you if a bot is a bad idea for your shop. If you only get 3 messages a week, you don't need automation; you just need to check your phone more often. We only work with firms that have at least 15 repetitive tasks a day. That is where the math makes sense. If you want to see if your admin load is high enough to bother with this, you can book a 20-minute intro call with Marcus Thorne. We will look at your current email volume and give you a straight answer on whether a simple bot will actually save you money.
