All workflows
Compliance

Counterparty and KYC document review

Bulk ingestion of counterparty packs, ownership structures, and AML docs with redundant retrieval patterns typical of compliance workflows.

KYC review on a Box + Workday + Salesforce stack: $0.82/run, 69% invisible across vendor invoices and connector sync. Firedog is the first tool to join them per review run.

Model 20%Prompt 11%Orchestration 7%Retrieval 27%History 27%Connectors 9%
ComponentWhy it is unattributedcalls$ / run
Model · 20%Extraction and risk scoring (GPT-4.1)Each document batch is sent to GPT-4.1 for structured entity extraction (beneficial owners, jurisdictions, sanctions flags) and a risk-tier score.tokens in: 62,000 · tokens out: 4,800Visible on the invoice.5$0.16
Prompt · 11%Compliance checklist and jurisdiction rulesA 9,000-token prompt encodes FATF categories, sanctions list references, and ownership threshold rules. Sent per call without caching.tokens in: 45,000 · tokens out: 0Visible on the invoice.5$0.090
Orchestration · 7%Document classification and ownership graph resolutionAn OCR/classification agent pre-processes each document; a graph-resolution agent traces ownership chains. Both produce billed model calls that no workflow monitor attributes to this job.tokens in: 18,000 · tokens out: 2,400Ownership graph resolution is run by a separate compliance microservice; its LLM calls land on a different cost center and never appear in the KYC workflow P&L.12$0.055
Retrieval · 27%Sanctions database and precedent-case lookupsEach entity triggers an embedding lookup against the sanctions database and a precedent-case vector store. Lookups are not cached between documents in the same counterparty pack.tokens in: 55,000 · tokens out: 0Sanctions lookup embeddings are managed by the compliance platform vendor; costs appear on the compliance SaaS invoice, not the AI workflow budget.44$0.22
History · 27%Prior review and exception history re-injectionPrior KYC review summaries for the counterparty are prepended to every call to enable change detection. Annual refresh runs accumulate 2-3 prior review summaries in context.tokens in: 32,000 · tokens out: 0Prior review summaries are injected by the case-management system as 'counterparty history'; the resulting token cost is not tracked against the AI workflow budget by any current tooling.5$0.23
Connectors · 9%Box, Workday, Salesforce, and Zapier connectorsThe review pulls the counterparty pack from Box, resolves entities and signatories against Workday, writes the risk tier back to Salesforce, and triggers onboarding Zaps. Each external record is normalized by an LLM mapping call, and entity data is re-pulled per document because it is not cached across the pack.tokens in: 15,000 · tokens out: 1,200Box reads bill to the storage tier, Workday lookups to the ERP API contract, and each onboarding Zap to the Zapier plan; the case-management system frames all of it as 'counterparty data,' so none reaches the KYC workflow budget.8$0.070
KYC document review costs $0.82/run but 69% is split across separate vendor invoices - sanctions lookups, ownership graph calls, prior-review history, and Box/Workday/Zapier connector sync that no single dashboard joins.
$0.82
cost per run
$6.6k
per year
69%
unattributed today
$3.28
per user, per day
Scenario: companies compliance team, weekly cadence. 8 active users, 4 runs/day, 250 working days. 22 documents/run. Avg doc: 14,000 tokens.

Default model: gpt-4-1

Trigger: New counterparty onboarding (sporadic, high-doc volume), annual KYC refresh cycles, and credit event reviews. Compliance teams run 5-20 reviews per week.

Who runs it: Prime brokerage, credit trading, and compliance teams at companies and asset managers. Triggered by new counterparty onboarding, annual re-certification, and credit line reviews. Packs are pulled from Box, entities resolved against Workday, outcomes written to Salesforce, and onboarding routed through Zapier, so each review round-trips through that connected stack under Workday's permission model.

What Firedog shows

  • Unified per-run view joining LLM, embedding, and compliance-platform costs for the first time.
  • Which counterparty documents are triggering redundant sanctions lookups due to entity overlap within the same pack.
  • Dollar impact of caching sanctions lookup results for 24 hours given the low rate of sanctions list updates.
  • A unified view of the Box, Workday, Salesforce, and Zapier connector cost per review alongside the LLM, embedding, and compliance-platform spend.

Decisions this informs

  • Cache sanctions lookup results per named entity for a 24-hour TTL; sanctions lists update daily at most, not per-document.
  • Deduplicate entity extraction across documents in the same counterparty pack before triggering ownership graph resolution.
  • Store prior review summaries as structured change-detection objects rather than full text to cap history re-injection at 5,000 tokens per counterparty.
  • Resolve each entity against Workday once per counterparty pack and cache it, rather than re-pulling and re-normalizing per document.

Connected systems

  • Box · Document storage
    Counterparty packs, ownership structures, and AML documents are pulled from the secured Box repository for each review.
  • Workday · ERP / HR
    Legal-entity, authorized-signatory, and internal-relationship data is pulled from Workday to resolve ownership and control chains; the review runs inside Workday's permission model.
  • Salesforce · CRM
    Risk tier and review outcome are written back to the counterparty record in Salesforce.
  • Zapier · Automation
    Onboarding and re-certification routing runs on existing Zaps that the review must trigger and extend.

Understand every AI dollar.

Book a demo to see how Firedog attributes every AI request across your organization, and what your leadership team can do with the data.