Everyone loves a shortcut until it leads to an angry customer, a messy refund, or a public mistake that one extra human check would have caught.
AI can be brilliant for speed, but many businesses are using it in the wrong jobs and creating avoidable problems.
Simon Childs, director of the Cheshire-based marketing agency Childsey, has spent years inside high-pressure retail and e-commerce environments, from senior roles across the Theo Paphitis Retail Group to running Dragons’ Den investments commercially, and now helps SMEs build brands and marketing that actually perform.
“AI allows small businesses to compete with larger companies without the same budgets, automating tasks, improving efficiency, and freeing up time. But the wrong investment can be catastrophic, and spending on tools that don’t fit your business or overcomplicating processes can just drain resources and hurt growth. The key is knowing which AI tools genuinely add value and which are more likely to do damage.”
Generic chatbots used for decision-making
Generic chatbots can look like the quickest way to get an answer, but they can also deliver confident nonsense that sounds like fact.
These tools can fabricate facts, policies or data, which becomes dangerous when people treat the output like a verified source rather than a starting point.
In day-to-day marketing and e-commerce, that can mean publishing incorrect claims, misunderstanding a platform rule, or making a pricing or promo decision based on something that simply isn’t real.
Used well, these tools can get work to a draft stage, but relying on them for final decisions is where the damage starts.
Top tip: Use chatbots for first drafts and idea generation, then verify anything factual against primary sources before it influences a decision.
Autonomous AI customer service bots handling complaints or policies
Customer service is where brand trust is won or lost, and giving an AI bot too much freedom can turn “efficient” into “expensive” very quickly.
There is a huge risk of poorly configured systems that mishandle refunds, misinterpret policies, or provide incorrect information – especially around money, orders and customer rights.
Even when the mistake starts with an AI hallucination, the business still owns the outcome in the customer’s eyes.
A hybrid approach keeps the benefits (speed, triage, routing) without letting a bot make decisions on sensitive issues.
Top tip: Let AI handle simple queries and triage, but force human approval for refunds, complaints, and anything involving policy or legal wording.
AI transcription and meeting note tools in high-stakes settings
Meeting notes feel harmless until the notes become the only record – and the record is wrong.
Some transcription tools can hallucinate content that was never said, which can create serious misunderstandings in high-stakes meetings, from commercial negotiations to performance discussions.
In practical terms, that can mean a client swears something was agreed because “it’s in the notes”, or an internal action gets assigned that nobody actually approved.
That’s not just awkward; it can become a compliance and accountability headache if those notes get shared, stored, or relied on later.
Top tip: For important meetings, treat AI notes as a rough draft and confirm key decisions in a human-written summary email straight after.
Simon Childs, Managing Director at Childsey in Cheshire, UK
“AI-powered” tools that are just thin wrappers
Plenty of software now claims to be “AI-powered”, but many of these features are immature, poorly integrated, or add little real value beyond what a team could already do.
The risk isn’t only wasted subscription spend; it’s duplicated work, generic outputs, and teams losing trust in AI altogether because the results feel bland or unhelpful.
This shows up often in marketing: automated social posts that don’t sound like the brand, or SEO content generated at scale that lacks creativity and nuance – and increasingly gets spotted as low-quality by audiences and search engines.
In real use, the best tools are the ones that give control, allow refinement, and prove measurable outcomes rather than vague promises.
Top tip: Run a short trial with clear success criteria (time saved, quality improved, errors reduced) before paying for any “AI upgrade”.
Unsanctioned “shadow AI” used by employees without oversight
Shadow AI is one of the quietest risks because it happens without leadership ever choosing it – an employee just pastes something into a tool to get the job done faster.
This becomes serious when sensitive company or customer information is used without safeguards, creating potential privacy, security and GDPR problems.
It’s especially risky because the business may not even know what’s being shared, by whom, and where it ends up.
29% of UK employees say they deliberately share private company data with ChatGPT at work, according to a study by Kaspersky UK.
Clear policies, training, and practical steps like anonymising data before use are what stop “helpful” behaviour turning into a breach.
Top tip: Publish a simple AI use policy and train teams to anonymise data – if it identifies a person, a client, or a deal, it shouldn’t go into a public tool.
