White-Label Client Reporting for Agency Portfolios

A client-facing report has earned its place when the client's own operations lead can reconcile it without phoning the account manager. For the agency portfolios we work on, that has meant one semantic layer and three report templates covering roughly forty accounts, rather than forty dashboards each with its own version of cost per acquisition. The structure: Reports read from the modelled layer, never from a live platform connector. A connector-backed dashboard silently rewrites its own history when a platform restates, and the client screenshots the version they saw. Ours reads yesterday's frozen table, so a figure quoted in a quarterly review can be reproduced six months later, row for row. Nine tiles a page is our ceiling, and every tile sits under the question it answers, written in a sentence. A definitions appendix ships inside the report itself rather than in a separate document that goes stale: how a lead is counted, when it is counted, what is excluded, and which figures come from the platform's own attribution as opposed to your warehouse. Freshness and gaps: Every page carries the timestamp of the last successful load, and a banner appears when the newest row is more than six hours old. When a platform's API returned nothing for a day, the chart shows the hole. We do not interpolate a missing day, and we do not carry the last value forward, because both make an outage look like performance. Access: Tenant separation is done at the connection, not with a filter in the report. A report-level filter is one URL parameter away from showing tenant B's spend to tenant A, and we have found that live in production twice. Depending on your stack that means sandboxed connections in Metabase, per-tenant service accounts and authorised views in BigQuery, or a Lightdash instance mapped to the dbt project. Custom domain and SSO, scheduled PDF snapshots on the first working morning of the month. What we refuse: We will not strip platform names to make an agency's reporting look like proprietary technology; every client eventually finds out and the trust cost lands on you, not on us. We will not build an attribution tab whose model your own account directors cannot explain out loud in a meeting. We will not put a metric on a client-facing page when the agency both controls the metric and can redefine it without a change record; anything client-facing is version-controlled and diffable. Commentary: The written interpretation stays yours. We build the layer that makes commentary cheap to produce, including a monthly variance table that flags which movements are large enough to be worth a paragraph, but we do not write client narrative and we do not generate it automatically. An agency that outsources its own point of view has nothing left to charge for. Who this is not for: In-house teams reporting to themselves, who need an internal analytics layer instead. Agencies wanting a white-label front end over data they have not modelled; the work in this listing starts from a warehouse, and if you do not have one that is a different engagement and a different conversation. Anyone whose main requirement is that the report look impressive in a new business pitch. Delivery: Templates are built against three real client accounts chosen to be awkward, not representative: your messiest tagging, your longest sales cycle, your most sceptical client. Rollout to the rest of the portfolio is scripted, and your team runs it with us watching once and then alone. Documentation covers how to add a tenant, how to change a definition, and what to say when a client asks why a number moved.