> ## Documentation Index > Fetch the complete documentation index at: https://docs.apollospace.ai/llms.txt > Use this file to discover all available pages before exploring further. # Financial services & lending > Origination, document collection, and follow-up that don't go cold for lack of hands — with the credit decision always in a person's hands. It's 5pm on a Friday. Three credit applications came in this week and none of them is complete: one is missing proof of income, another stalled on the incorporation papers, the third sent the wrong document and nobody followed up. The analyst will sit down Monday to a queue that was born stale — and every day a file stays incomplete is an applicant rethinking whether it's worth continuing. **What leaves your plate:** guiding the applicant through collection, chasing what's missing, assembling the file for the analyst, and keeping the pipeline current all start happening on their own — with context, within budget, and with **the credit decision always waiting on a human.** ## How it works, in one line Credit origination is, at bottom, **a complete file that reaches the decider's desk on time.** The agent doesn't decide credit — it takes the friction out of the funnel so the person who decides gets a case ready to analyze, not a half-empty folder. ```mermaid theme={null} flowchart LR A[Trigger
new application, missing doc,
follow-up date] --> B[Agent pulls context
from Company Brain + CRM] B --> C[Decides + uses tools
Chats, CRM, Documents, Routines] C --> D{Credit decision,
pricing or money?} D -->|No| E[Collects / organizes / updates status] D -->|Yes| F[Proposes + calls the analyst
with the complete file] E --> G[Becomes memory
the next application starts faster] F --> G ``` ## Concrete flows ### 1. Guide the applicant through document collection **Trigger:** a new application enters the [pipeline](/en/features/crm) — from the site, a referral, or opened by an advisor. [Athena](/en/agents/athena), the organization's operator, pulls the checklist for that credit product from the [Company Brain](/en/features/brain) — which documents the product requires, in what format, which eligibility rules apply. She opens a [chat](/en/features/chats) or [WhatsApp](/en/integrations/whatsapp) and walks the person through it **one document at a time**: "got the ID, now I need the last 3 months of income proof." Each file received becomes a [document](/en/features/documentos) organized in that application's [folder](/en/features/pastas), and the [CRM activity](/en/features/crm) logs what's arrived. If the person asks something the checklist doesn't cover — a specific rule, a policy exception — the agent searches the Brain; if it doesn't find it, it **doesn't make it up**: it logs the question and escalates to a human. Collecting a document is funnel work; interpreting credit policy isn't. ### 2. Assemble the complete file for the human analyst **Trigger:** that application's checklist is complete — every document is in. The agent pulls everything into a cover [document](/en/features/documentos): the applicant's data, the list of attachments with what each one proves, and the points it asked the Brain to check (consistent records, dates within validity, amounts that match across papers). It opens a [task](/en/features/boards) on the analysis board **assigned to a person**, with the file attached and the pipeline stage moved to "ready for review." The agent **organizes and checks completeness and consistency** — it does not issue a risk opinion, compute a score, or recommend approve or deny. It hands over a case ready for a person to decide; credit judgment begins exactly where the funnel work ends. ```mermaid theme={null} flowchart TD A[Checklist complete] --> B[Agent assembles cover file
data + attachments + checks] B --> C[Opens task assigned
to a human analyst] C --> D{Analyst decides} D -->|Approves / denies / prices| E[Human decision logged
in the auditable trail] E --> F[Agent updates status
and notifies the applicant — if approved] ``` ### 3. The follow-up that recovers a stalled application **Trigger:** a [routine](/en/features/rotinas) the agent schedules for itself — every morning, it sweeps applications with a document pending for more than X days. On fire, the agent pulls from the [CRM](/en/features/crm) whoever's funnel has stalled, drafts a gentle reminder of what's missing, and sends it on each applicant's channel ([email or WhatsApp](/en/features/outbound)) — with **human approval before the send** if you want to keep a hand on outbound. Anyone who doesn't reply after a few attempts becomes a [task](/en/features/boards) for the advisor, and a [proactive notification](/en/features/rotinas) flags the book owner's bell. **Why this matters:** an application stalled on a missing document is the biggest revenue leak in a credit funnel — not because the applicant gave up, but because nobody chased on the right day. A routine has no hectic Friday: it chases every time, on time, and turns the silent queue into an actionable signal. ### 4. Keep the pipeline and statuses alive — and answer the "how does it work" **Trigger:** any change on an application — a document arrived, the analyst decided, the applicant replied. The agent updates the [CRM](/en/features/crm) right after: adjusts the stage, logs the activity, attaches what changed. The pipeline stays **always current** — whoever opens the board sees today's truth, not yesterday's snapshot. And when an advisor or the applicant asks "which documents does this product require?" or "where does Maria's application stand?", the agent answers from the [Company Brain](/en/features/brain) and the CRM, with information that already exists — instead of becoming one more internal email queue. ## What stays human Money is handled with maximum care. The agent **proposes and prepares, never decides alone**, whenever the subject is: * **The credit decision itself** — approve, deny, set a limit. Always a person's call. * **Pricing and terms** — rate, tenor, collateral, any term of the offer. * **Disbursement and money movement** — releasing, transferring, refunding. * **Collections and recovery actions** — reporting default, renegotiating, any collection contact. * **Contract and signature** — credit instrument, amendment, any commitment that becomes an obligation. * **Regulatory or legal communication**, governance decisions, and destructive actions. In those cases the agent prepares everything — the file, the draft, the context — and **waits for a person to approve.** Every action lands in the auditable trail, with author and justification, isolated per organization. ## Where to start Upload each credit product's requirements into the [Company Brain](/en/features/brain) — required documents, formats, eligibility rules, and the process answers you repeat every week. That's where the agent pulls what to request and what to check. Connect [WhatsApp](/en/integrations/whatsapp) or [chats](/en/features/chats) and let the agent guide the applicant document by document, filing each one in the application's folder. Start with human approval before every external message. Create a [routine](/en/features/rotinas) that sweeps stalled applications and chases what's missing. One agent, one cadence, one channel — and autonomy grows like a ratchet, never a leap. ## Next steps The same collect, check, and organize pattern applied to everyday paperwork. Where the application pipeline lives — stages, activities, and history always current. Origination and qualification at the front — how an application enters the funnel. Why each customer's sensitive data stays isolated per organization.