A performance director once paused every Bing campaign across North America for one of his agency’s largest clients. Nobody noticed for three months. The client noticed first.
He told the story on PPC Live The Podcast, as reported by Search Engine Land on October 24. Rémi Kerhoas, now Performance Director at Eskimoz, was early in his paid search career at a 14-person agency, managing a Silicon Valley technology client that spent millions a month across Google and Bing. A few months in, he discovered he had accidentally paused the entire Bing book of business. The campaigns had been inactive for a quarter. The agency had three people on the account plus an internal client team, and still no one on the agency side saw it.
The pause was a human error, and a forgivable one. The three months were not an error. Three months is a system that was never built.
Nobody was assigned to notice
According to the account, the agency ran on CSV uploads of hundreds of thousands of rows, with the team’s attention pointed at campaign execution rather than at spotting unexpected change. The fixes that followed are the most instructive part of the story: more frequent manual account reviews, careful archiving of CSV files, and a request for a dedicated QA hire that was never filled. Every one of those is a person trying harder. None of them is a control that keeps working on the week when everyone is busy, someone is on leave, and a platform changes its UI.
That gap is the normal condition of agency operations, not an outlier. Workflow design usually optimises for the work that has to ship by Friday, and monitoring gets whatever attention is left over. When the process is already fragile, adding AI on top of it does not fix the fragility, it just makes the output arrive faster. We wrote about that dynamic in why AI is exposing broken agency workflows, and a paused account sitting quietly for 90 days is the textbook example.
The failure class is silent state changes, not bad performance
Most agency dashboards answer one question: how are we doing? They are bad at a different question: did something change state overnight, and does anyone know?
- A campaign or ad group flipped to paused, or a line item moved to limited by policy
- A daily budget edited, or a shared budget quietly consumed by a single campaign
- An ad group delivering zero impressions because bids or targeting drifted out of range
- A conversion action deleted, renamed, or double-counting so the numbers look better than reality
- A tracking tag, UTM template, or redirect broken on the landing page
- A product feed or scheduled CSV import that stopped running, so bids are pricing stale inventory
- Ads disapproved, or a landing page now returning a 404 to paid traffic
Each of these produces a small, plausible-looking dip rather than a dramatic drop. In an account spending millions, a missing Bing slice reads as normal variance for weeks. That is why the failure mode is time, not magnitude: the money is not the main loss, the three months of nobody knowing is.
Five checks that would have caught it
- The absolute-zero test. For every active account and platform, if yesterday’s spend was zero while the plan said it should be positive, that is a P1 alert. No thresholds to tune, no judgement required. This single check ends the paused-account class of failure permanently.
- Pacing against plan. Compare month-to-date spend and conversions against the planned run rate. Anything more than roughly 25 percent off in either direction is worth a human look, because both underspend and overspend usually mean a setting moved.
- Conversion health. Clicks above a set floor and conversions at zero for a full day. Also watch the inverse: conversion volume jumping far above the trailing range often means duplicate counting, not a great day.
- Pipeline freshness. Every feed, CSV import, and scheduled job stores a last-successful-run timestamp. If it is older than its own schedule allows, alert. A feed that stopped on a Tuesday will bid badly by Thursday.
- Change log review. Pull platform change history daily and flag status changes that have no matching internal ticket or request. Automated rules and account reps change things too, and those changes are invisible in performance reports until they have done damage.
Write it down as a runbook, not a habit
The difference between an agency that catches these and one that gets caught is rarely talent. It is whether the checks exist somewhere other than a senior person’s memory. Four things make that real: a named owner and a named backup for every account, so leave and turnover do not create blind spots; thresholds stored in one versioned place rather than tuned per person; a single alert destination with a required acknowledgement, so an alert is either resolved or visibly open; and a weekly exception report listing what fired and what was done about it.
Add one counterintuitive line to that report: how many monitors actually ran. Silence should be evidence that the checks executed, not evidence that nothing happened. If you want the wider version of this, our 2026 playbook for agency workflow automation covers the surrounding pipeline.
Agents can run the checks, a human still signs the report
This is the most defensible use of automation in an agency, because it is narrow, read-mostly, and measurable. An overnight run that pulls yesterday’s spend, conversions, job timestamps, and change logs across every account, then posts a single exception list by 7 a.m., replaces a review meeting and removes the reliance on someone remembering. Give the agent read-only credentials, an explicit allowlist of accounts, and a log of every call it makes. Keep writes, budget changes, and pause/unpause decisions behind human approval.
What you should not do is let the agent’s output become the client deliverable without review. Automated reporting is where the return on agency AI is clearest, as we argued in client reporting is where agency AI actually pays off, but the judgement in the commentary is still the product.
What the client remembers
In the reported case, the client stayed, and the director kept his job. The durable damage was not the budget; it was that the client discovered the gap first. Clients forgive mistakes that are found and fixed internally. They do not forgive learning about a quarter of silence from their own dashboard.
So make the check visible before it is needed. One line in the monthly report stating what was monitored, what fired, and what was resolved does more for retention than another chart. And if you cannot answer, today, which specific check would have caught a three-month pause in your largest account, that is the first gap to close. Have us audit your paid media monitoring before the next quiet quarter.