Three years ago, a car dealership owner I was advising installed a chatbot. He was proud of it. It was well-reviewed, cost a reasonable monthly fee, and his web developer told him it would capture more leads.
Six weeks later, he pulled it.
Not because it didn't work technically. It worked fine. It replied. It was online. It did the thing chatbots do.
He pulled it because a customer had come in for a test drive and opened the conversation by saying: "I asked your chatbot whether the car came in blue and it told me it didn't know and to call during business hours. I nearly didn't come."
Nearly didn't come. A hot buyer, referral from a trusted friend, £38,000 transaction. Nearly lost because the chatbot's answer to a specific product question was to redirect to a phone call that nobody was available to take.
The Difference Between a Chatbot and a Concierge
A chatbot is a router. It takes an input, matches it to a pattern, and returns a pre-built response. It's essentially a very fast FAQ with a conversational wrapper.
When the input matches the pattern: great. When it doesn't — which is constantly, because customers are human and don't ask questions in predictable patterns — the chatbot defaults. It gives a non-answer. It asks the person to rephrase. It says "I'm not sure about that." It tells them to call during business hours.
Every one of those default responses carries a message the business never intended to send: we're not really equipped to handle this right now.
A concierge is different in a foundational way. It doesn't match patterns. It understands context. It holds a conversation the way a knowledgeable, calm front-desk operator holds a conversation — reading what's actually being asked, drawing on what it knows about the business, moving things forward instead of redirecting.
What "Trained on Your Business" Actually Means
This phrase gets used a lot in AI marketing. Almost every chatbot vendor will tell you their tool can be trained on your business. What they mean, usually, is that you can upload an FAQ document and their tool will search it.
That's not training. That's retrieval.
When I say Velaeva is trained on a business, I mean something specific: the concierge knows the actual inventory, pricing tiers, seasonal offers, and availability windows. It knows the objections that come up most often for this business specifically, and how the owner would want them handled. It knows the brand tone — not just "professional" or "friendly" but the actual way this business communicates.
It knows the edge cases. The awkward questions. The situations where a human needs to step in — and exactly what information to capture and pass on when that happens.
The Moment That Always Reveals the Difference
There's a test I run with every business owner who's skeptical about whether this matters.
I show them two responses to the same message — something real, a question a buyer actually sent at 10:47 PM on a Thursday.
Response A (standard chatbot):
"Thank you for reaching out! For personalised medical advice, please contact us during business hours or book a consultation. Our team looks forward to hearing from you!"
Response B (Velaeva concierge):
"Bruising varies quite a bit depending on where we're treating and how your skin responds — for lips and fine lines around the eyes, mild bruising for 2–4 days is common, though plenty of clients have none at all. We always recommend leaving a week before any big event. The best way to get a sense of what to expect for you specifically is a quick consultation — would you like me to flag your details for the team to reach out tomorrow morning?"
The business owners I show this to never need more convincing.
The Calibration Problem Nobody Talks About
Here's what happens when you deploy a standard chatbot without the right calibration work:
Month one: it handles the easy questions. Basic FAQs. Booking links. Hours of operation. Fine.
Month two: the edge cases start showing up. The buyer who wants to know if you can accommodate a very specific requirement. The customer with a concern about a previous bad experience. The enquiry that's half sales question and half complaint.
The chatbot hits its limits. It gives generic answers. It redirects. It frustrates.
By month three, the business owner is either turning it off or ignoring the conversations it's supposed to be handling — because they've learned that the quality is inconsistent and they can't trust it.
This is not a failure of AI. It's a failure of calibration. The tool was deployed without doing the actual work of mapping how the business communicates, what it needs to know, and how it should handle every category of input it will receive.
What Good Looks Like
The gold standard for an after-hours AI concierge is this: a buyer sends a message at 11 PM, receives a response that makes them feel genuinely handled, and either wakes up the next morning ready to move forward — or, in many cases, has already moved forward because the concierge completed the next step for them.
For the dealership owner from the start of this piece — after Velaeva went live — a buyer messaged at 11:42 PM asking specifically about a blue SUV with third-row seating available for test drive that weekend.
The concierge checked live inventory, confirmed availability, asked for a preferred time window, collected contact details, and sent a calendar confirmation. The buyer woke up with a booked test drive. The sales team woke up with a warm lead and full context.
No call during business hours required. No form. No friction. That's what fixing the calibration problem actually looks like.