This is the AI-powered, human-led support framework ContentWorks has built for Vidalux. It answers customer questions from one Product Brain … so every reply traces back to real Vidalux data, never a guess. Two channels … an email assistant and a website chat … a safety net that strips anything it can't verify, and a Brain that learns from its own gaps and keeps itself current. Below: what each part does, the honest status of each, and exactly how your team runs it.
Both the email assistant and the website chat read from the same Product Brain, so a customer gets the same grounded, accurate answer whichever way they reach you. Here's the honest status of each.
The reason this is safe to put in front of customers isn't the AI … it's what sits around it. Every answer passes these four guards before a human ever sees it.
The assistant searches the Product Brain for every question and answers only from what it finds. It can't pull facts from the open internet … if the answer isn't in your Brain, it doesn't make one up.
Before an answer is allowed through, every number and spec is checked back against your real product data. Anything that doesn't trace to source is stripped out and replaced with "confirm on the product page" … so a made-up figure is caught before it can reach a customer.
Prices are deliberately kept out of the Brain so they can never go stale. Instead they're pulled fresh from your live website and refreshed every 30 minutes, so a quoted price reflects your live pricing, not a frozen snapshot.
Every reply lands as a draft for a person to approve … the assistant never sends on its own. If an answer is uncertain or off-catalogue, it's flagged and handed to a human rather than guessed.
It's built so the worst it can do is say "I'm not sure, check the page" … never confidently give a customer a wrong spec or a wrong price. That's the whole design: helpful when it's certain, honest when it isn't, and a human on every send.
Most support bots go stale the day after they're built. This one watches where it falls short and feeds that back … so the Brain gets better over time instead of drifting out of date. The loop runs automatically up to the point a human signs off … nothing reaches the live Brain without your approval.
The watching half is live … gap-logging and the daily new-product check both run on their own. Turning what they surface into a Brain entry is a quick human step … the assistant flags what needs adding, a person writes it and approves it. Less hunting, same human sign-off.
Two things to know, and both are simple. First, the daily rhythm on email. Second … and this is the important change … your team no longer hand-logs questions into a spreadsheet. The assistant captures them itself. Your job is to approve, not to feed.
When a support email comes in, the assistant drafts a grounded reply inside Thunderbird. Your team member opens it like any other draft.
Skim the draft … it's already on-brand and fact-checked, so usually a quick read. Want it different? Just tell it … "make it warmer", "mention free delivery", "shorten it" … and it rewrites the reply, keeping every fact. Or edit by hand … it's yours.
If the assistant marks a reply as uncertain or off-catalogue, that's your cue to answer it personally. The assistant is telling you it isn't sure … trust that.
Happy with it? Send, exactly as you would any email. Nothing goes to a customer until a person hits send.
Remember the spreadsheet where the team logged every tricky question by hand? That manual job is gone. Here's exactly what replaces it … and the only two times you still log something yourself.
You're not feeding the list any more … you're approving it. If you catch yourself typing out a question the assistant already handled, stop … it's already on the list, waiting for your tick.
No spin. Here's exactly what's working today, and what's still in trial.