One dashboard per client: a week inside an agency reporting operation

Written by Uptimal, the team that builds and operates Upbuild and runs marketing analytics for multinational clients across the globe. Everything below reflects how the product actually works, including the parts that are deliberately boring.

Lantern is an invented agency, but the week below is not invented. Many clients, nine account managers, a head of reporting called Marta, and one dashboard per client that nobody at the agency has exported, pasted or emailed since spring. Here is how the operation runs from Monday to Friday on the AI Dashboards side of Upbuild, with alerts and exports where they plug in.

Monday, 08:40: nobody built anything over the weekend

Marta opens the dashboards list before her coffee has cooled. One row per client, each with a version number, a refresh cadence and a state. Northwind Retail is on v33, hourly, fresh at 07:04. Acme Health is on v45, every 30 minutes. The Meridian Markets tracker refreshes once a day at 06:00, because that client reads it once a week and a daily cadence is plenty.

Nothing on the list was touched on Saturday. Upbuild refreshed every page on its own, in the agency’s timezone, and each page shows when its numbers were last updated. That is the Monday: the account team and the client read the same page, with yesterday’s figures already in it, and the first meeting of the week starts from the numbers rather than from the question of whether they are current.

The client side is quieter still. Every client opens their dashboard through a private link, view only, with a password the account manager sent them once. No Upbuild account, no seat on the agency’s plan, no login to Google Ads, Meta or GA4. The link shows the published version and nothing else: not the app, not the other clients, not the accounts underneath. If a client forwards it, the worst case is one page of numbers that were going in their deck anyway.

Monday, 14:00: a new client comes on

Acme Health signed on Friday. Onboarding is three moves, and none of them is a ticket.

First, the accounts. Whoever at the agency already has access to Acme’s ad and analytics accounts signs in to Upbuild as themselves, with Google or Microsoft, and those accounts are ready to use. Nothing extra to buy per platform, nothing to paste from a developer console.

Second, the sharing. An admin, Marta in this case, shares the Acme accounts with the Acme account team, one person and one account at a time. The account manager on Acme now sees Acme and nothing else. She cannot open Northwind, and neither can any dashboard she builds, because a dashboard can only draw on the accounts its builder was given.

Third, the dashboard. The agency keeps its template as a prompt, not as a file to copy. The account manager types it into Upbuild AI with the client’s name in it: "Build the Acme Health overview from our retail template: spend, orders, AOV and ROAS across all their accounts, last 30 days ending yesterday, one section per market, campaigns in order of spend. Refresh every hour." Upbuild AI writes the data requests, runs every one of them against the real accounts, draws the page and hands over a draft. She looks, publishes, confirms the cadence, switches on the link with a password, and sends it. Acme used to get a PDF every month, so she attaches last month’s and asks for the same layout on live data. The client’s first Monday looks familiar, and it is already moving.

Wednesday, 11:15: the client asks for a change

Northwind’s marketing director replies to the Monday note: could the overview show Meta and Google side by side, and put the campaigns in order of spend rather than alphabetically? In the old operation that email started a small project. Here the account manager opens the Northwind dashboard, opens its Upbuild AI thread and types the request more or less as it was written: "Split paid media by platform, Meta and Google side by side, and order the campaign table by spend."

The answer is not applied. It arrives as a draft, version 34, with an Open v34 button beside it. The live page keeps showing v33 until she publishes, and the client’s link only ever shows what is published. She opens the draft, checks that the totals still match, publishes, and replies to the email within ten minutes. The change is recorded with her name and a note, next to every earlier version. If the director prefers the old order after all, v33 is one click away, and restoring it is itself recorded as the next version, so the history never lies. The mechanics are in editing dashboards by asking.

Thursday, 16:30: an account manager leaves

Jordi hands in his notice. He runs eleven clients and built every one of their dashboards. In an agency that reports from a BI tool, this is the moment somebody asks who owns the reports, and the answer is usually his personal Google account, which means a week of transferring ownership and re-authorising connectors before his last day.

Here Marta removes him from the organisation on his last afternoon and his access ends. The eleven dashboards stay, refreshing on their schedules, opened through the same client links. So do the connections, the exports and the alerts he set up. Nothing was in his personal chat account, because nothing in Upbuild belongs to a person: everyone signs in as themselves, and the work belongs to the organisation.

The colleague who inherits the clients reads the version history of each dashboard: who changed what, when, and the note that says why. She restores nothing and rebuilds nothing. On Monday the clients notice no difference, which is the only handover a client should ever notice.

Friday, 09:15: where alerts and exports plug in

Dashboards are what the clients see. Two other things run underneath them, on the same accounts and the same refreshes.

The first is alerts. Months ago Marta wrote one sentence: "Warn the media team in Teams when any campaign has spent 80% of its monthly budget before the 20th, every client." Upbuild checks it every time the data refreshes. On Friday morning the message lands: Meridian ES, Search Generic, has spent 84% of its monthly budget on the 12th, and at this pace it runs out on the 14th. The media team moves budget before the weekend instead of explaining a dark week on the following Monday. A naming rule runs the same way each morning, listing the campaigns that break the convention and the name each one should have, without renaming anything itself. More in alerts you set by asking.

The second is exports. Two clients want their own tables: one into their data warehouse, one into a finance spreadsheet. Each export was set up by a prompt and runs every morning, with the history backfilled once at the start. When Acme’s accounts were connected on Monday, the export whose rule says "every account whose name starts with Acme" picked them up before the next run. Nobody at the agency has opened a CSV since spring either.

And when Marta wants a question answered across every client at once, she asks AI Insights: "Which clients saw blended CPA rise more than 15% this week, and which campaigns drove it?" Any answer can be added to a dashboard in one prompt, which is how the group view on her own screen was built.

What the week cost

Add it up: one prompt for a new client, one sentence for a change, one removal for a leaver, no exports, no template copies. The operation does not get harder as the client list grows, which is the whole point of running it this way. Prompt-built, self-refreshing, versioned and shared by a link the client opens without an account: that is the dashboard side of the platform. The two ways in are the same for every agency: get a demo to see the platform live, set up the way you report, or start free and prompt your first client dashboard today.

Frequently asked questions