> ## 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. # Research accounts before outreach > Build a reviewable account-research Play that keeps source evidence, explains the account reason, and records unresolved rows before outreach begins. [Human guide: work with your agent →](https://deepline.com/docs/human/recipes/research-workflow.md) > [!NOTE] > > Produce an account-research file whose reasons link to actual source evidence. Use this guide when you need a reviewable reason to contact an account. The result is a research file with evidence and a clear next action—not generic account prose. ## Choose the research shape | You need to decide | Start here | | - | - | | Why an account is worth contacting now | Continue with this account-research workflow | | How a competitor positions itself in paid channels | [Competitive and ad intelligence](https://deepline.com/docs/plays/play-ad-intelligence.md) | | Which visitor should enter a reviewed outbound queue | [Warm outbound automation](https://deepline.com/docs/recipes/warm-outbound-web-intent.md) | | Which consented contacts should become an ads audience | [Build a paid ads audience](https://deepline.com/docs/ads-audiences.md) | | Which product or warehouse signal should change a GTM action | [Database access](https://deepline.com/docs/database-access.md) | Every research workflow should preserve the input row, source URL or record ID, reason a row passed or needs review, and the next system action. Research that only produces prose is not ready for an automated destination. ## Prerequisites * `accounts.csv` with a stable account identifier and, preferably, a domain. * A definition of the evidence you want, such as a job post, product change, funding event, public record, or company page. * A rule for what should be skipped, suppressed, or sent for human review. ## Define the research contract Give your coding agent this request. It asks for a local Play and makes the review boundary explicit: ```text theme={"languages":{"custom":["/languages/deepline-prompt.json"]}} /deepline-gtm Create account-research.play.ts for accounts.csv. The Play must accept { "csv": "accounts.csv" }. For each account, preserve the source row and return: - the account identifier - source URLs or other evidence for every claim - a concise why_now field based only on that evidence - a suggested next action or review state - a miss_reason when the account is unresolved or should be skipped Apply our suppression and ownership rules before suggesting an action. Do not write to a CRM, sequence, or ad platform. Include a top-level Play description and show me the proposed input and output before running it. ``` ## Validate it without deploying or spending credits Once the agent has written the file, validate its source before it runs: ```bash theme={"languages":{"custom":["/languages/deepline-prompt.json"]}} deepline plays check ./account-research.play.ts ``` `plays check` bundles and validates the local Play. It does not publish the Play, create a run, or call a provider. ## Run a representative pilot Create `accounts-pilot.csv` with the header and three representative accounts: one likely match, one typical account, and one sparse or difficult account. ```bash theme={"languages":{"custom":["/languages/deepline-prompt.json"]}} deepline plays run --file ./account-research.play.ts \ --input '{"csv":"accounts-pilot.csv"}' \ --watch ``` The run can use Deepline credits. Export the completed result for review: ```bash theme={"languages":{"custom":["/languages/deepline-prompt.json"]}} deepline runs export --out account-research-pilot.csv ``` ## Review the pilot before you scale For each returned row, check: * The evidence points to the correct account and supports the stated reason. * A missing source is recorded as unresolved, not written as a confident claim. * Suppressed or owned accounts did not become outreach candidates. * The next action is appropriate for the evidence and has a clear reviewer. * The observed Deepline-credit usage is acceptable for the full list. > [!TIP] > > Do not turn research output into outreach automatically. Review the evidence > and the action boundary first. ## Run the reviewed workflow on the full list Use the same Play and replace the pilot file with the full input: ```bash theme={"languages":{"custom":["/languages/deepline-prompt.json"]}} deepline plays run --file ./account-research.play.ts \ --input '{"csv":"accounts.csv"}' \ --watch ``` ## Next step Use [Warm outbound automation](https://deepline.com/docs/recipes/warm-outbound-web-intent.md) when a visitor event should enter a human-review queue. Use [Build a paid ads audience](https://deepline.com/docs/ads-audiences.md) when the approved outcome is an ads audience rather than an outreach action. ## Verify this task Open the evidence for each accepted reason. Keep unavailable sources, suppression, uncertain identity, and no-results visible before activation. Ask your agent to explain a failed check before repeating work or expanding the scope. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.