Charged even when no email is found: what does delivering 1,000 clean leads really cost?
priced a job by delivered leads, overlooking that many local shops have no discoverable email and the old provider charged for every attempt. The first month required an out-of-pocket top-up.
This week's lesson: a fixed quote must account for failed calls, token-price changes and human rework, not merely successful-query price times quantity.
Start with what the buyer will pay for
This issue's 20 cases expose a serious weakness between a working AI service and a profitable one: incomplete cost accounting. A 40%–65% lookup success rate, several clients sharing a model key or changing API prices can wipe out margins even when the technology works.
Missed costs include unsuccessful requests, manual visual/layout repairs, refunds and acquisition. More resilient operators separate tool receipts from service fees, track per-client budgets in Postgres, and accelerate drafts for 16 email clients while retaining human design review.
Take two immediate steps:
- put service fees and actual API usage on separate quote lines,
- and test 50 real target businesses for lookup and bounce rates before committing to a fixed bulk price.
- Opportunity quality this issue
- Separate tool billing and budget controls make services more resilient; untested fixed-price lead packages can lose money.
- Prices and accounting context
- Email lookup $0.011 per match; tool usage passed through at receipt cost; quoted creative retainer $2,000/month; reported independent-app margins 36%–49%
- Who pays
- Marketing clients buying local-business lists, agencies managing AI support or email for multiple accounts, and subscribers to clearly targeted apps
- Frequently used tools
- Agentik、Treg、n8n + Postgres + OpenAI API、Kling 4.0、Higgsfield、Claude Code、Hostinger、Zapier
Billing by delivered results while paying per attempt can force the provider to cover the shortfall
accepted a fixed-price local-business email job, assuming 1,000 delivered leads would mean 1,000 paid lookups. Many plumbers and small restaurants had no usable email, yet the old supplier charged every attempt, pushing costs above the quote. The founder covered the first month's gap.
Switching to Agentik, reportedly about $0.011 per found email with misses free and $0.002 per 100 map businesses, helped; Treg was also mentioned. Batch receipts now expose matches and misses. Even free failed lookups do not remove the labor of duplicate-domain removal and deliverability checks.
Keep service fees separate from unpredictable API usage, and pass through actual receipts
Three Reddit accounts point to a similar approach. After underestimating usage, separated setup and monthly labor from tools billed at receipt cost. Each client gets a dedicated key, daily cap and tool allowlist, with run-ID receipts attached to the invoice.
proposes a Postgres ledger for teams sharing one OpenAI key: check whether a client's monthly spend has crossed a $50 budget before running. Together, cost separation and budget controls address the kind of price volatility that says forced an AI product to close.
A Brazilian founder's R$200k October ledger shows how acquisition and outsourcing consume apparent margins
shares unusually detailed October accounts.
- Enxovaly received R$81,756.40;
- costs included Apple's 15% share of R$12,263.46, R$4,169.58 tax, R$817.56 RevenueCat, R$29,037.02 advertising and R$5,600 in creators, AI content and email.
Reported profit was R$29,478.78, or 36.06%. Bíblia Raízes earned R$118,610.77, spent R$34,000 on ads and retained R$58,455.58, or 49.28%. Advertising is close to final profit in scale. Combining creators, generated content and email into one expense line makes it harder to tell which creative actually earns back its acquisition cost.
Once drafts are fast, acceptance-ready quality becomes the scarce resource
managed email marketing for up to 16 ecommerce clients, cutting the first drafts of eight monthly emails from three or four hours to 20 minutes. Visual design still required people.
demonstrates Kling 4.0 vertical drama with reference images: buyers pay per episode, while the producer pays for each regenerated shot, so repeated face inconsistencies can erase margin. Data engineer built RAG evaluation tools for healthcare and finance but received no replies to 40 CTO InMails.
Technical ability still needs a concrete, immediately assessable offer, such as a ten-question error report, to become a first sale.