The Report That Ended the Relationship
A performance marketing agency spent six months producing excellent results for a client. Then, in a quarterly business review, a sharp-eyed client VP noticed that the conversion numbers in slide 4 didn't match the conversion numbers in the appendix. The difference was small — a rounding issue in how two data sources were combined. But the damage was immediate.
"If these numbers don't match, how do I know anything in this report is right?"
The agency lost the account within 60 days.
How Data Quality Erodes Agency Trust
Unlike product companies, marketing agencies sell expertise and judgment. When a data error appears in a client deliverable, it doesn't just undermine that one metric — it calls into question every number the agency has ever produced.
The most trust-damaging data errors in agency work:
Stop sending campaigns to bad data — Sohovi validates your list before you hit send — try Sohovi free.
Numbers that don't add up: Row-level numbers that don't sum to the total shown. Percentages that don't add to 100%. These are immediately noticeable and suggest sloppiness.
Inconsistent figures across slides: The same metric reported differently in different slides — usually because two team members pulled data independently from different date ranges or attribution settings.
Wrong attribution windows: Reporting conversions with a 30-day attribution window when the client's benchmark uses a 7-day window. The number looks bigger but is comparing apples to oranges.
Unreachable benchmarks: Claiming "above industry average" using a benchmark that doesn't apply to the client's category, country, or channel.
Missing data that appears as zero: A pixel fires on Monday, stops working Tuesday through Thursday, fires again Friday. The Tuesday-Thursday data shows zero. It looks like performance dropped — when actually the data is just missing.
Building a Data Quality Culture in an Agency
The root cause of most agency data errors isn't carelessness — it's the absence of a QA process.
Agencies that avoid data trust problems do three things:
-
Assign one owner for data per report: Two people pulling the same metric with different settings is the most common source of inconsistency. One person owns the data pull; others work from the same source.
-
Have a non-creator review every report: The person who built the report can't QA it effectively. They see what they meant to create. A fresh reviewer sees what's actually there.
-
Create a report QA checklist: Every number verified against source. Every total checked. Attribution windows documented. This checklist takes 20 minutes and prevents the kind of error that ends accounts.
