Claude caching cuts API costs 90%—if you structure right
The café two streets over still smells like burnt milk at 7am, and the barista is learning not to fill the portafilter past the line. Outside, delivery vans idle while drivers tap their phones, waiting for route algorithms to catch up with Monday traffic.
Claude’s prompt caching cuts API costs by 90%—but only if you architect for it
Most operators structure prompts the way they think, not the way the cache works—and pay full price every call.
Claude’s prompt caching lets you reuse large chunks of repeated context—style guides, product catalogues, email archives—without paying to re-process them every time. The savings are real: cached tokens cost one-tenth of fresh input tokens. A 10,000-token style guide that you reference in every API call drops from $0.30 per hundred calls to $0.03. At scale, that compounds.
But the cache only triggers when your prompt structure follows specific rules. Context must sit at the start or end of your prompt, stay above 1,024 tokens, and remain static between calls. If you embed your reusable content halfway through a dynamic prompt, or if you tweak it slightly each time, the cache misses and you pay full freight. The feature also only works on Claude 3.5 Sonnet and Opus—Haiku doesn’t support it.
The practical implication: if you’re calling Claude’s API dozens of times a day to analyse subscriber behaviour, draft email variants, or score content ideas against a fixed rubric, restructuring your prompts to front-load static context will cut your bill by more than half. If you’re making five calls a week with no repeated context, caching won’t move the needle.
TACTIC
When Zapier’s filter step costs you more than paths do
Filters and paths both let you control Zap logic, but they count tasks differently. A filter stops a Zap cold and still consumes a task; a path branches without burning an extra run until an action fires inside it. If you’re filtering out 60% of incoming records, you’re paying for every rejected row. Restructuring the same logic with paths and conditional actions can halve your monthly task count—especially on higher-tier plans where you’re already close to your limit and overages cost real money.
WORTH READING
Why Stripe Tax might still cost more than your accountant
Stripe Tax automates sales tax calculation and remittance across US states and forty-plus countries. It costs 0.5% of each transaction, with no monthly fee. That sounds cheaper than an accountant until you run the numbers on a $10,000-a-month subscription business: $50 monthly to Stripe Tax, versus a $75 quarterly filing from a local bookkeeper who also reconciles your bank feed and knows your entity structure. Automation wins at scale—past $30k monthly revenue—but below that threshold, the per-transaction fee often outpaces manual accounting, especially if your subscriber base sits in just two or three tax jurisdictions.
READER QUESTION
Plausible or Fathom—which privacy analytics tool actually earns its keep
Both promise simple, GDPR-compliant analytics without cookie banners. Both charge per site. Plausible starts at $9 monthly for 10,000 page views; Fathom starts at $14 for 100,000 events. The pricing inversion matters if you run multiple low-traffic sites—Plausible bills per property, Fathom offers unlimited sites on every plan. Feature-wise, Plausible ships goals and funnels; Fathom adds uptime monitoring and email reports. If you need one dashboard for twenty WordPress newsletters with modest traffic, Fathom pays back. If you run a single high-traffic publication and want granular goal tracking, Plausible is the cleaner fit.
Know someone who would like this? Forward today’s email—every operator we reach is one closer to running an online business with a little less friction.