Why farms fail

Farms rarely fail because the automation stopped working. They fail on content supply, instrumentation, network design and scaling too fast. The five failure modes, in order of frequency.

Reviewed by John Carroll · Updated 7 October 2026

In nearly every case the automation was fine. Something around it was not.

1. Scaled before instrumented

The most common by a wide margin. Accounts are dying, output is below target, and nobody can say which phones, which network group or which creation batch is responsible. Growth multiplies a leak nobody can see. The fix is unglamorous: record every ban against every variable, then look.

2. Content supply never solved

A fleet is distribution. Capacity built without a pipeline to feed it is capacity paid for and idle, and the agency concludes the fleet does not work when what failed was planning.

3. Network designed last

Phones bought, rack built, then the realisation that every account sits behind one IP. Retrofitting separation onto a built fleet is expensive and slow, and in the meantime accounts are being lost in correlated batches.

4. Warming skipped under pressure

Accounts sitting idle for a week feels like waste, so they get pushed early, so they die early, so the agency pays the creation cost repeatedly for accounts that never produced. The saving is illusory and the cost compounds.

5. Deployed at full size

Two phones working proves the automation logic and nothing about two hundred, which adds concurrency, heat, queueing and detection at scale. Agencies that jump straight to full size lose the first cohort entirely and then distrust the entire approach.

The pattern underneath all five

Each one is a decision taken in the wrong order. Network before hardware, instrumentation before scale, content before capacity, measurement before expansion. The technology is rarely the thing that fails.