"The account went silent" is one of those phrases that gets said calmly. It sounds like an IT issue — something to look at after the morning standup. It is not. A silent account is a metered loss: it accrues at a fixed rate per hour, and the meter runs from the moment it stops until the moment someone acts, not from the moment someone notices.
This article works through that arithmetic. All numbers below are illustrative examples chosen to make the structure visible — they are not measured results and your own inputs will differ.
What "silent" actually means
Silent is not one failure. It is a family of failures that share a symptom: an account that should be spending is not.
- Billing failure. The card declined, the threshold was hit, the prepay ran out.
- Suspension or policy stop. The account or all its campaigns got restricted.
- Everything limited. Ads are technically live but limited by policy, so delivery collapses.
- Budget or bid change. A pause, a zeroed budget, a bid cap that priced you out of the auction.
- Tracking break. Conversions stopped importing, so a smart bidding strategy correctly concluded there was no value to buy and stopped buying.
- Reporting break. The account is fine, but your data pipeline is not — functionally identical from where you sit, because you cannot tell the difference until you look.
That last one is why "silent" is the right word rather than "stopped". You are not observing the account; you are observing a signal about the account. Absence of signal is ambiguous, and ambiguity is exactly why teams delay acting.
The two costs, and why people only count one
Direct cost is the profit the account would have produced during the outage. It is bounded by the account's normal daily earnings, and it is the number people intuitively reach for.
Recovery cost is everything that happens because the outage happened: the learning phase you re-enter, the auction position you lose, the ranking of a paused campaign that does not come back at the same efficiency, the buyer time spent diagnosing instead of scaling.
Recovery cost is routinely ignored because it is harder to attribute. It is also frequently larger. An account down for six hours costs six hours of profit plus, potentially, several days of degraded efficiency after it comes back. If you only count the six hours, you will consistently under-prioritise detection speed — which is precisely the mistake that keeps teams on manual morning checks.
The arithmetic of reaction latency
Total loss decomposes cleanly:
Loss = (hours undetected + hours to act) x hourly earning rate + recovery cost
Break the first term apart, because the two halves are fixed by completely different things:
- Time to detect is set by your monitoring. If a human checks accounts twice a day, your average detection time is roughly six hours and your worst case is twelve. If a system checks every fifteen minutes, average detection is minutes.
- Time to act is set by your process: who gets told, whether they are awake, whether they have access, whether the fix needs someone else's card or approval.
Most teams try to improve total loss by working on the second half — being more responsive. The first half is where the leverage is, because it is the only one you can drive close to zero without hiring anyone.
Worked example (illustrative numbers)
Take one account with these made-up but plausible inputs:
| Input | Illustrative value |
|---|---|
| Daily spend | 1,200 units |
| ROAS | 1.6 |
| Daily profit | 720 units |
| Hourly profit | 30 units |
Now compare three detection regimes for the same six-hour billing failure that begins at 02:00.
| Regime | Detected | Acted | Hours lost | Direct loss |
|---|---|---|---|---|
| Morning manual check | 10:00 | 10:30 | 8.5 | 255 units |
| Twice-daily check | 08:00 | 08:30 | 6.5 | 195 units |
| Automated alert to Telegram | 02:15 | 02:45 | 0.75 | 23 units |
One account, one incident, and the gap between the first and third row is roughly 230 units of profit. That is unremarkable on its own — it is a rounding error in a monthly P&L. The point is what happens when you multiply it.
Now scale it
Suppose the portfolio is 150 accounts and, in a given month, 12 percent of them experience at least one silent incident. That is 18 incidents. At an average 230 units of avoidable loss each, the month costs about 4,100 units of profit that had nothing to do with media buying skill. It was pure detection latency.
Nothing in that number required a bad decision, a bad creative, or a bad offer. It is the cost of finding out late, and it recurs every month at roughly the same rate until the detection mechanism changes.
The compounding case: a paused winner
The example above assumed a flat account. The expensive version is when the silent account is one of your best.
Consider an account scaling well — call it 2,000 units of daily spend at 2.2 ROAS, so roughly 2,400 units of daily profit. It goes silent overnight because of a payment issue and is restarted the next afternoon. Two things now compound.
First, the outage itself. Fourteen hours at 100 units per hour of profit is about 1,400 units gone directly.
Second, the restart penalty. Bidding strategies use recent conversion history. An interrupted account restarts with a gap in that history and, in practice, frequently needs a period of re-stabilisation before it returns to prior efficiency. Suppose recovery takes four days at an average 30 percent below normal profit:
- Day 1 to 4: 2,400 x 0.30 = 720 units of shortfall per day
- Four days: about 2,880 units
Total for one overnight outage on one good account: roughly 4,300 units, of which only a third was the outage itself. The tail is bigger than the event.
This is the asymmetry worth internalising: outages are most expensive on the accounts that are working best. A losing account going silent is arguably a small favour. A winner going silent for a night can cost a week of that winner's contribution. Any monitoring approach that treats all accounts identically is mispricing exactly the accounts that matter most, and that is a strong argument for weighting alert severity by the account's recent contribution rather than by a flat spend threshold.
What actually reduces the number
Four practical changes, in order of impact per unit of effort.
Detect on absence, not on deviation. Most alerting is built on thresholds — CPA moved, ROAS dropped. Those all require data to exist. A silent account produces no data, so threshold logic never fires. You need an explicit rule watching for an expected event that did not happen: this account is live, it has budget, it has spent nothing in N hours, raise it. If you only have deviation detection, you are blind to the most expensive failure class.
Set the window per account, not globally. An account that normally spends every hour should trigger after two or three hours of zero. An account that spends in a narrow daily window would trigger constantly under the same rule. Use the account's own spend rhythm, or you will generate noise on low-frequency accounts and get too-late alerts on high-frequency ones.
Route critical alerts to a human channel. A dashboard badge does not wake anyone up. Email is fine for the morning review; something with push notifications — Telegram, for instance — is what closes the gap between 02:00 and 08:00. That choice is most of the difference between the first and third row of the table above.
Give every account an owner and route by ownership. Detection with no clear responder just relocates the delay from detection to action. If the alert reaches the buyer who runs the account, time-to-act collapses. Role-based ownership makes this possible without spamming the whole team on every event.
The uncomfortable summary
Silent accounts are not a monitoring nicety. They are a recurring, quantifiable, largely avoidable line item, and their cost is set almost entirely by how long it takes you to find out. The media buying is not the part that is broken. The clock is.
If you want that clock closed without building the detection layer yourself, this is one of the things AdsTracker is built to do: it holds every Google Ads account in one view from the stats you send it, watches for accounts that go quiet against their own spend rhythm, and pushes the alert to email or Telegram while the fix still costs minutes instead of days.