KIWABI AI Program · Prepared for Naohisa Tamura
Twenty-eight days into the ninety day trial. Two separate systems are running. Here is what each one does, what it is already doing for KIWABI, and the two decisions that move it forward.
The Concierge sits in your navigation and in the pop-up Emi launched on 25 August. It knows the Root Vanish and Root Beauté range, answers in English, and sends every shopper to the right product page on your store.
An advisor that talks to your customers is your brand talking to your customers, so this month we checked every product statement it makes against what KIWABI publishes on its own pages, and rewrote anything that did not match. Fifty automated checks now run against it every day. Forty-eight passed on today’s run, and the two that did not are wording and product-detail cases we are correcting.
What it tells your customers about your products comes from your own pages.
Accuracy came first because nothing else is worth building on top of a system that invents. With that in place, the next piece of work is the one Emi has asked for: an advisor that acknowledges what a customer is actually worried about, rather than answering only the product question in front of it.
Shopify revenue, Google spend and Meta spend now sit in a single data warehouse covering January to August. That one table is the foundation of the cross-channel reporting this programme was set up to deliver, and a dashboard runs on top of it, reading from the warehouse rather than from a spreadsheet anyone can edit.
Two things make this harder than a spreadsheet, and both are the reason the figures can be trusted.
Underneath both: every number is computed from your data tables. The language model explains figures, it does not calculate them. That is why this can be put in front of a board.
Google bills KIWABI in yen. Your store reports in dollars. The engine will not divide one by the other on a rate we invented, so it holds the blended return-on-ad-spend figure back and says why.
The rate moves, so a single fixed number would go stale. Our proposal is to take the rate each Monday and apply it to that week, recorded alongside every figure so any number can be traced back to the rate it used. Confirm that convention and we build it into the engine, and the blended figure follows.
Three sources are in the warehouse. The rest of the picture is still outside it, and each one needs an access step on the KIWABI side before we can read it.
The Monday rate, applied to that week, recorded with every figure. Say yes and we build it into the engine, which releases the blended return-on-ad-spend figure and the campaign comparisons underneath it.
Today it is linked from the storefront. Embedded, it meets customers inside the shopping experience, and we can report how many conversations became orders and what those orders were worth. That measurement is the point of the programme, and placement is what unlocks it.
Build the agreed exchange rate into the engine so the blended figure can go live. Connect Amazon and Klaviyo so the reporting covers the whole business rather than three sources of it. Embed the Concierge if you approve it, and begin reporting conversations against orders. The ninety day review lands at the end of October, and the aim is that you judge it on numbers you can defend.