A customer is on one of our client’s stores at 9pm, looking at a Miele oven, and they ask a specific question: does this model have the pyrolytic cleaning function, or is that the tier above? It’s a good question — the kind that decides whether they buy tonight or close the tab. And on a catalogue of nearly seven hundred products, the honest truth is that the precise answer might not be written into the store’s own data. So what happens next?
For most website chatbots, one of two things — and both lose the sale.
The two bad answers
The first is that it invents one. It sounds confident, it’s fluent, and it’s wrong — because the model would rather produce a plausible answer than admit it hasn’t got one. On a casual chatbot that’s just annoying. On a store where a customer is about to spend four figures on an appliance, a confident wrong answer is worse than no answer at all: they buy the wrong thing, and it comes back to the retailer.
The second is the dead end: “Sorry, I don’t have that information.” Honest, but useless. The customer came with a real question, got nothing, and left — after the retailer had already paid, in SEO or ad spend, to bring them to the store in the first place.
Neither is good enough for the stores we manage. So we do something about it.
A real example: C.R. Wood
C.R. Wood Electrical, family-owned since 1959, is one of the UK’s largest independent Miele dealers. We manage their Magento stores — 677 Miele products across a domestic site and a commercial one — and every product’s information is grounded in Miele’s own terminology, because that’s how we built the catalogue in the first place. Accurate, consistent, Miele’s language, not ours.
As part of managing that store, we deploy an AI assistant called Concierge on the site. And Concierge is built on a principle we won’t move off: it answers from real sources, or it doesn’t answer.
It’s grounded first in the store’s own content — products, delivery, returns, policies. Where that content covers a question, it answers from it. No invention. That alone is a solid assistant, but on its own it would still hit the dead end the moment a customer’s question ran deeper than the store’s own data. Which, on a technical catalogue like Miele’s, happens often.
When the store’s data runs out, it doesn’t stop — and it doesn’t guess
Here’s the part we’re pleased with. When a question isn’t covered by the store’s own information, Concierge doesn’t give up and it doesn’t fabricate. It retrieves the answer from Miele’s own authoritative source — and it tells the customer that’s where it came from.
So that 9pm oven question gets answered from Miele’s official product information, attributed to Miele. The customer gets an accurate answer to a spec that was never written into the store, sees plainly where it came from, and carries on towards checkout. C.R. Wood don’t have to document every nuance of seven hundred products by hand, and they never have to put an unverifiable claim in their own voice.
The principle didn’t change — still grounded, still attributed, still engineered not to invent. We just widened the sources it’s allowed to draw on, from the store’s data to the store’s data plus the manufacturer’s official source. It’s the grounded version of “it can answer anything”: not an assistant that knows everything, but one that knows where to look and always says so.
The goal was never a chatbot that always has an answer. It’s one that only gives answers it can stand behind — and shows the customer the source.
It gets better every week — with us in the loop
“Self-improving AI” is one of the most over-sold phrases going, so let’s be precise, because the honest version is the reassuring one.
After each conversation, Concierge reflects on how it went and, where it fell short, proposes a specific improvement — a gap in the store’s content to fill, a better way to handle a recurring question, a source worth adding. Those proposals go into a queue that we review and approve before anything changes on a live store. So it compounds: the assistant gets measurably sharper over time, shaped by real customer questions rather than guesswork — but it never quietly rewrites itself unsupervised on a client’s site. For anything customer-facing, that should reassure you, not worry you. It learns, and it asks. The human in the loop is us.
And when a question genuinely needs a person — a complex order, a judgement call — it hands off to a human in real time rather than talking in circles.
Our own software, deployed and managed
Concierge is built by our sister company, Momentum Technology Solutions, and we deploy and manage it as part of the store relationship. That’s the difference between us and an agency that installs an off-the-shelf plugin and hopes: when the tool needs to do something specific — like reach into a manufacturer’s product data and answer accurately — we’re not waiting on a third party. We build it, we run it, and we tune it around how each store actually works.
That’s what managing an ecommerce store means to us — not just keeping the lights on, but bringing better tools to the problems our clients’ customers actually hit. If you run a store, especially one carrying products you didn’t write the specs for, it’s worth a conversation.