> ## Documentation Index > Fetch the complete documentation index at: https://deepline.com/docs/llms.txt > Use this file to discover all available pages before exploring further. # Understanding enrichment cost > How marginal attempts, acceptance rates, cache behavior, and billing modes affect cost per useful result. The unit price of one source request is only part of enrichment economics. The workflow may make several attempts, validate candidates, find no result, or produce a value that still needs manual repair. Cost per accepted outcome connects those activities to the reason the work was run. ## The denominator changes the conclusion Imagine two illustrative pilots, each with 100 input records. Route A consumes 100 Deepline credits and yields 50 accepted results. Route B consumes 80 credits and yields 20. Their accepted-result costs are 2 and 4 credits respectively. These are invented numbers, not Deepline prices. They show why a smaller total bill can still represent a more expensive useful result. The team must also consider whether the uncovered records matter to the intended process. Acceptance is part of the measurement. If one route counts unverified candidates and another counts only approved values, their ratios do not measure the same outcome. ## A fallback's value is marginal A source used after earlier attempts serves the unresolved population. Its useful contribution is the additional accepted results it supplies, together with the usage and delay needed to obtain them. That population may be harder than the original list. A provider's standalone coverage is therefore not automatically its incremental coverage in a waterfall. Overlapping results can provide redundancy without improving the remaining gap. Stopping after acceptance avoids later calls on those rows. It does not make earlier requests or validation free. The action's billing contract determines whether requests, returned results, pages, or other units carry usage. ## Failed requests and legitimate misses have different economics A legitimate no-result is evidence that the route did not establish the needed fact. A failed request may instead reflect invalid input, access, throttling, or execution failure. Counting both as lack of coverage can lead to buying another source when the existing integration needs repair. A retry can consume work even when it eventually produces the same accepted value. Repeated manual investigation is another cost, although it is not a Deepline usage field. Keeping these categories separate helps identify which change could actually improve the result. ## Reuse is conditional Deepline can reuse matching completed runtime work. An unfinished receipt is not a completed cache hit, and independent misses can execute independently. Intentional refresh and cache policy can also create new work. A customer table can hold a historical accepted result for reuse, but that is a workflow freshness policy. Saving a value does not make it indefinitely current, and a database read is not a universal guarantee that the provider will never be called again. ## BYOK changes who receives part of the bill Under BYOK, provider-account usage follows your agreement. Deepline compute or other applicable charges can remain. Under managed access, the action's Deepline pricing applies. A team may need both its own provider invoice and Deepline's recorded charges to evaluate total cost. Deepline does not expose internal provider spend. A missing usage event or a pending amount should not be treated as a zero charge. Estimates describe a future input mix; recorded outcomes describe actual work under its contract. ## Better economics can require a different boundary A cheaper request may reduce quality enough to increase manual repair. A more selective acceptance rule can lower immediate coverage while protecting a costly downstream action. Conversely, a low-risk research task may accept a broader class of candidates than an automated CRM write. The right optimization follows the business action, not a universal preference for the smallest API price. ## Apply this understanding * [Evaluate a waterfall](https://deepline.com/docs/guides/b2b-data-enrichment-waterfall.md). * [Measure a representative pilot](https://deepline.com/docs/guides/waterfall-enrichment-claude-code.md#evaluate-coverage-and-usage). * [Pricing contract](https://deepline.com/docs/pricing.md) and [provider rate cards](https://deepline.com/docs/_generated/pricing-schedule.md). * [Direct-call billing recovery](https://deepline.com/docs/sdk-v2/tool-execution-recovery.md#billing-and-plays). This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.