Falling out with an AI agent — and what it tells us about using AI in sales
I’ve fallen out with an AI agent. How can that happen and what does it tell us about the risk of taking shortcuts when using AI in sales?
AI agents don’t have emotions.This means my “falling out” is a one way thing. If I go back to that same agent now it will treat me in the same even-handed, unemotional way it has done for the last four weeks.
This is probably a good thing.
But it also means that agent doesn’t understand the emotions it generates in a customer who’s been trying to solve an invoicing problem for the last month.
At a practical level it means that a lengthy conversation becomes repetitive, because AI doesn’t have the emotional bandwidth to vary its approach. Unlike a human, it can’t for example, offer to sort things out over a coffee.
All of this means it’s unsatisfactory to say I’ve fallen out with an AI agent. But I’m hacked off with the company involved and that’s the business-critical summary.
How can you fall out with an AI agent?
Your natural approach when dealing with something dispassionate and logical is to do the same. Meet cold, hard facts with better facts.
If you’re the company director who’s decided to use AI agents to handle first stage customer enquiries then that agent needs to get its facts right.
The problem is it’s dealing with the public.
You might reasonably assume that 90% of enquiries are straightforward — and maybe that’s true. But if your responses to the other 10% of customers entirely defy logic then your service reputation can be shredded in a very short time indeed.
The point of this post isn’t to moan about a specific company, it’s to highlight how important it is that AI agents are trained well. However…the story has better context if I mention the “fallout” is with an AI agent at a utility company.
Ironically this agent works for a company that’s named after a very intelligent eight-legged sea creature. According to the Natural History Museum “they can solve complex puzzles, use tools and recognize individual humans”.
The AI agent I’m dealing with struggles with all three of these things.
What was sales training like in the old days?
My first sales management job had two targets. One, naturally, was a £ sales target. The second target was what proportion of my time was spent on the road training my team.
That goal was 80%. That’s 4 days out of every week spent coaching sales people face to face. That investment in people was extraordinary.
Not surprisingly, that commitment influenced whether people chose to join the organisation and it helped build a reliable pipeline of talent for the future.
It also meant that customers received a consistently high standard of service. That built loyalty and it built repeat.
It’s long been a joke in sales teams that new starters are handed a car and a satnav and expected to deliver results.
But in reality this is rarely what happens. Businesses take time to
* induct people into company functions
* ensure they understand the product they are selling
* support them with FAQs
* help them reflect the company culture — “the way we treat customers here”
* present themselves externally in a professional manner
They do these things because every customer interaction changes reputation, for better or worse.
Throughout all our years in selling this has been best practice. So it’s extraordinary (and sort of comical) that some company owners think they can set up an AI agent and customers will think it’s not noticeable.
How can I tell I’m emailing an AI agent?
AI agents can be really valuable at the right point of your customer experience. For one thing, AI agents are fast and they don’t sleep. If someone has a simple query for you at 3am your AI agent will answer it.
If AI agents can filter out simple queries then you’re left with more time for your sales team to focus on bigger clients with bigger deals.
The art is in making that customer experience seamless.
For a couple of weeks or so my recent experience seemed normal. Responses were slow, but I attributed that to it being a holiday period.
Things only started going awry after I clicked the optional “Escalate this case” button. A few giveaways raised my suspicions I was dealing with an AI agent:
* The manager it was escalated to just had a first name, Angeline. No surname
* Angeline’s emails sign off as “management”. No job title, department or contact details
* Twice I’ve had emails from Angeline that jump the thread. It’s like she’s just been introduced to a customer with a problem rather than referring back to the earlier conversation
* Automated Customer Review emails pop up after every interaction. If I click the frowning face (and I always click the frowning face) I get an apologetic email from — you’ve guessed it — Angeline! Angeline in management
* Each solution email from Angeline only addresses a limited number of my account problems. She can’t identify four separate problems and fix all of them
* The final giveaway: She emailed, telling me the billing problems were fixed and my balance was zero. This for a property we sold three years ago. I’d expect a human to spot that
Artificial Intelligence or Superficial Intelligence?
With my customer service antennae raised I used AI to see if Angeline is real. The surprise is that she actually does exist. So why does it look like the responses I’m getting are AI generated and what does that do to trust?
It turns out this large organisation is very proud of its AI agent, called Arlo. So proud they post about it.
Behind the scenes they are using Magic Ink to create customer service responses. They claim this is doing exactly what we outlined above — freeing up service teams to focus on more complicated cases.
Perhaps the critical line in this article is this:
“Team members are also trained to review everything written by Magic Ink with a fine-tooth comb”
My experience is more yard brush than fine-tooth comb.
We’ll probably never find out why escalating our problem made communication worse. Maybe it’s the messages generated by the AI agent. Maybe the team are so busy they default to copy and paste.
It doesn’t matter why. What matters is that a large organisation with a very good reputation for customer service ensures that a sentient human trains and assesses the output from its AI agents.
AI can drive efficiency for the 80% of simple queries. But it won’t drive effectiveness if it’s expected to — but doesn’t resolve the more complex 20%. In fact more work will be needed to unravel those issues and patch up customer relationships.
How can you use AI agents without impacting customer service?
The answer to this is found by going back to the old days. Though it’s actually not that far back.
When you recruit sales people you give them simple problems with small customers first. You train them and scrutinize the outcomes. Only then do you let them loose at scale.
AI is changing how tasks are delivered. But it doesn’t change why they are done.
Better to limit to a small number the queries your AI agent is allowed to deal with until you’re truly happy it behaves the way you would want a human to.
Some questions to consider might be:
* What customers and issues is AI allowed to handle?
* At what stage should a human be involved?
* How do we assess customer response?
* How can we tell if there are repeated messaging problems?
* Can customers clearly identify when we are and are not using AI?
Your sales process might have changed but that doesn’t mean your customer expectations have.
If you’re rethinking how your sales process works, our fractional sales director service can help you build one that holds up — with or without AI in the mix.]
Frequently asked questions about using AI in sales
Can AI agents replace salespeople?
Not in any meaningful sense, no. AI agents can handle high-volume, low-complexity queries — and handle them quickly. But selling involves reading context, building trust, and navigating ambiguity. Those are human skills. The businesses getting the best results from AI use it to free up their sales teams so they spend more time with customers, not replace them.
How do I know if my AI agent is damaging customer relationships?
Watch for rising complaint volumes, lower review scores after AI interactions, and queries that escalate to human teams at a higher rate than expected. If customers appear to be losing confidence in you after interacting with AI then that’s the signal to review how it’s deployed and what it’s allowed to handle.
What’s the right way to introduce AI into a sales process?
Start narrow. Identify the query types where AI performs reliably, like simple FAQs, booking confirmations or technical specifications, and limit it to those. Build in human review of AI outputs before they reach customers. Expand the scope only when you’re genuinely confident in the quality. It’s the same approach you’d take with a new recruit.
Not sure your sales process is working as smoothly as it should? Our Untangle Your Sales scorecard gives you a clear picture in under three minutes.
*Angeline is not the real name of the person handling our account