Overview
This update covers the last few weeks of work on AI Analyst and the Daasity MCP Server. Rates such as ROAS and click-through rate are now calculated from totals, so they hold up when you group by week or campaign. Large results and large views now reach your AI complete, and its errors say what to fix. Your views carry more of your Looker model, including Retailer Inventory, and one measure with an error no longer keeps the rest of its view from building.
What's new
- Rates are calculated from totals.Some measures are written as a rate on each row, such as sales divided by spend, and then added up. Across a week or a campaign, that doesn't give the rate for the group. For these measures (ROAS, click-through rate and cost per click are the common ones), your views now divide the total by the total. Measures that convert a currency or a unit, such as session duration in minutes, are unchanged.
- Fields from joins with a few repeated rows.When a joined table repeated a handful of rows that differed only in columns the explore doesn't use, the join was left out of the view along with its fields. Those repeats are now removed during the refresh and the join comes through, with fields such as an order line type or a ticket tag and the measures that rely on them.
- Default filters from every layer of your model.When your own Looker project adds a default filter to a Daasity explore, the view now lists it alongside the filters the explore already had.
- More explores become views.An explore was skipped when one of its fields repeated a setting, or when its SQL generated its own rows, such as a calendar of operating days. Looker accepts both, and your views now do too.
- Large results arrive complete.On the MCP Server, a long result such as two years of daily sales is now sent in a compact format that fits your AI tool's size limit. If a result still has to be cut short, the AI is told how many rows were left out, so it can say so.
- Large views fit, with the full detail one step away.A view with hundreds of fields, such as Sales Report, could overflow what an AI tool accepts in one response, and the AI would guess at field names. The list of views now fits, and the AI can pull one view's complete detail when it needs it.
- Errors the AI can fix by itself.When a query names a field a view doesn't have, the error now names that field and the closest real ones. Month and quarter fields are marked as text labels, like 2026-10, so the AI filters them correctly.
- Retailer Inventory views.The Retailer Inventory explores (Target, Ulta, Sephora, Macy's, Dillard's, Costco, KeHE and Walmart) now become views for the retailers you have data for. Weeks of supply uses the same sales window the retailer's inventory dashboard opens with.
- A connection that only runs scheduled jobs stays connected.A connector used only by automated reports could lose its warehouse access after 30 days and fail with an unclear error. Access now renews every time the connector is used. If it does lapse, the AI says so and tells you how to reconnect.
- Connect Your AI shows each refresh.The page shows a refresh as it runs and when each view was last built. A view that failed to build is listed with the reason.
What's fixed
- One measure with an error doesn't stop its view.If your warehouse rejects the SQL behind one measure, that measure is left out and the rest of the view builds. Before, the whole view failed to build. Ask the AI what was left out of a view and it gives the reason.
- Day names match their dates.An answer that named a day of the week beside a date could name the wrong day. Day names now come from the calendar.
- Optional parts of Daasity's Looker model apply only where you use them.Some optional pieces, such as Shopify B2B, were being applied to every account's views. On an account that doesn't use them, that could change a filter or leave an order count off Sales Report. They now apply only when your Looker model includes them.
- Measures that add up an ID are left out.A measure that sums a product ID or a similar identifier isn't a real total. It's now left out of the view, and the AI gives the reason if you ask for it.
- Your own SQL for a Daasity view is what the view reads.If your Looker project redefines a Daasity view with its own SQL, the view now reads that SQL, so the columns you added come through.
- A comment inside a filter no longer breaks its view.An explore whose filter contained a SQL comment could fail to build.
- AI Analyst considers every matching view.A very large view at the top of the list could crowd out the others, so an answer could come from the wrong view.
What to expect
A rate that's now calculated from totals can differ from a Looker tile built on the same measure, because the tile still adds up the row-level rates. To check the AI's number, divide the two totals for the same period.
What to do
Run Refresh Views from the account menu in AI Analyst. No Refresh Semantic Layer is needed. If your company has several brand accounts, run it on each one. Larger results, clearer errors, the Connect Your AI page and the fix to day names work now, with or without the refresh.
Then try two things. Ask the AI what was left out of your Sales Report view, and why. And ask for ROAS by week on one of your ad platforms, then check one week against that week's sales divided by its spend.