Why per-word pricing is the wrong model for PPC localization

Per-word pricing exists because it made sense for the work translation agencies used to do: long documents, legal contracts, user manuals, where volume and effort scale together in a fairly predictable way. Charge a rate per source word, multiply by the document length, and both sides know roughly what they're paying for. The model is simple, auditable, and, for that kind of work, defensible.

It falls apart the moment you apply it to PPC ad copy, and the reason is structural, not a matter of degree. A Google Search headline runs to 30 characters. A Meta primary text hook might be a single sentence. Priced per word, that headline costs almost nothing, a few cents, regardless of how much the CPA swings on whether those seven words are right. Meanwhile the actual cost driver in performance localization isn't word count at all. It's how many variants a campaign needs, how fast they need to turn around for weekly creative refreshes, and how much risk each string carries before it can ship. None of that is captured by counting words.

There's a second mismatch, and it's newer. AI can now generate a large first pass of ad copy variants at close to zero marginal cost. If a pricing model still bills per word, it's charging for the part of the process that has become nearly free, while giving away the part that actually costs money and determines whether the campaign works: the risk-routing decision and the human review it triggers.

Three specific ways per-word pricing breaks for PPC:

1. It misaligns incentive with impact. A seven-word headline that determines whether an ad account clears policy review in a new market carries far more weight than a seven-word phrase buried in page eleven of a manual, but per-word pricing charges the same for both.
2. It doesn't fit iteration speed. Paid social needs creative refreshed weekly, sometimes daily during a launch window. Invoicing per word for that cadence turns every testing cycle into a procurement exercise, which is exactly the friction that makes teams stop testing as often as they should.
3. It ignores where AI already does the work. Once generation is largely automated, billing per word effectively bills for typing, not judgment. The judgment, deciding what needs a human, and doing that review well, is where the actual cost and the actual value sit.

The alternative is pricing tied to what actually drives cost and outcome: flat campaign bundles or retainers scoped to language pairs, platforms, and expected risk-tier volume, so the invoice reflects the work the engine is actually doing, not a word count left over from a different industry.