Make.com Vendor Onboarding Workflow: AI KYC, Contracts & ERP Sync

Published by GreenCircuit on

Flowchart illustrating vendor onboarding: intake webhook, OCR extraction, AI KYC scoring, contract generation and e-sign, and ERP synchronization

Problem Context

Vendor onboarding is often slow and error-prone: manual document handling, inconsistent identity and risk checks, contract delays, and missing or mismatched vendor data in the ERP. The objective is a Make.com scenario that accepts vendor submissions, extracts and validates documents, runs AI-assisted KYC/risk screening, produces a contract from a template and e-signs it, then synchronizes a verified vendor master to the ERP — all with deterministic error handling and finance-ready acceptance criteria.

Implementation Workflow

The implementation is a set of modular Make.com scenarios/subflows. Each module uses an intake ID (idempotency key) passed through the pipeline so retries do not create duplicates.

  1. Intake & Document Capture
    • Trigger: Webhook module receives a vendor submission from a form or vendor portal (Typeform/Google Forms). Files are uploaded to a secure cloud folder (Google Drive/S3) via Make modules and linked to the intake ID.
    • Create a canonical draft vendor record in a lightweight store (Airtable/Google Sheets/Make DB) capturing raw inputs and file paths.
  2. OCR & Field Extraction
    • Invoke OCR (Google Vision via HTTP module or a dedicated OCR API) to extract structured fields: company name, registration numbers, tax IDs, bank details, and authorized signer data.
    • Write OCR outputs and confidence scores back to the draft record. If required fields are missing or confidence is low, route the record to human review (Slack/email with file attachments and an action button to correct).
  3. AI KYC & Risk Screening
    • Use OpenAI modules following a strict JSON-output prompt pattern to normalize names and addresses, then call sanctions/PEP APIs (or combined third-party vendor) to look for exact and fuzzy matches.
    • Aggregate results into a risk score with attached rationale tokens (matched-list ID, fuzzy-match confidence, adverse-media flag). Persist the raw evidence snapshot for audit.
    • Decision logic: auto-approve if risk < low threshold; route to compliance when medium; block and auto-notify vendor when risk > high threshold. All escalations require human sign-off before blocking.
  4. Contract Generation & E-Sign
    • For approved vendors, generate a contract using DocsAutomator/TemplateDocs or Google Docs templates. Feed vendor fields, risk tier, and conditional clauses (e.g., additional indemnities for certain jurisdictions) into the template population module.
    • Produce a PDF and initiate e-sign via DocuSign/HelloSign connector. Track signer audit trail and PDF storage location on the vendor record. If e-sign fails, retry transiently; if persistent, route to legal for manual signing.
  5. ERP/Accounting Sync
    • Once contract is signed and finance acceptance is recorded, push vendor master data to the ERP using a native Make connector or REST API module. Use a staging-create-then-promote flow: create in ERP staging endpoint and validate returned ERP ID before promoting to production vendor master.
    • Map critical fields: vendor name, legal entity, tax ID, bank token (never raw account numbers in logs), payment terms, default GL mapping, and onboarding tag. Persist ERP ID and last-sync timestamp.
  6. Notifications & Audit Trail
    • Send templated notifications to procurement, compliance, and finance at each critical milestone. Maintain a secure archive of all artifacts (OCR output, KYC report, contract versions, sign audit) linked to the vendor record for compliance and internal audit.

Error-Handling & Supervisor Pattern

  • Idempotency: generate a unique intake ID at first touch and include it in all downstream API calls to prevent duplicate vendor creation.
  • Retries: standard exponential backoff for transient network/API errors. Implement Make scenario-level retry and a supervisor subflow that attempts N retries then escalates with payload context.
  • Error routing: each failure writes a detailed entry (module name, payload snapshot, error text) to an error queue (Airtable/DB). Automated remediation steps and a link to jump to human remediation UI should be included.
  • Dead-letter handling: records stalled in review for configurable time (e.g., 72 hours) auto-escalate to a manager with recommended remediation actions.

Acceptance Criteria for Finance

  • Mandatory fields: validated tax ID, ERP vendor ID, signed contract PDF with audit trail, payment terms, currency, and GL code mapping. Any missing required field flags the vendor as “payment-disabled”.
  • Bank verification: bank account must be token-verified (micro-deposit or third-party account-verify) before payment-enabled flag is set.
  • Auditability: every onboarding decision must have stored evidence (OCR outputs, KYC report, contract version) accessible from the vendor record for at least the retention period required by finance/regulation.
  • SLAs: low-risk vendors fully onboarded <48 hours; medium-risk within 5 business days with compliance sign-off tracked in the record.

Architecture Notes

  • Keep Make scenarios modular: Intake, OCR, KYC, Contracting, ERP Sync, and Error Supervisor as separate scenarios to simplify monitoring and reusability.
  • Data model: canonical vendor draft with version history and immutable audit artifacts. Store secrets and tokens in a secrets manager and do not log raw PII or banking details.
  • Prefer native Make connectors for Google Drive, DocuSign, and supported ERPs. For unsupported ERPs use secure REST modules with sandbox-first testing and idempotent endpoints.
  • Observability: emit structured events (module, latency, risk score) to a logging/monitoring workspace for trend analysis and SLA reporting.

Risks and Guardrails

  • LLM hallucination: require strict JSON schema outputs and post-validate LLM responses against deterministic checks (regex, list lookups). Never rely solely on LLM output for deny decisions.
  • Privacy: encrypt PII at rest, redact logs, and obtain vendor consent at intake. Minimize PII in LLM prompts; use identifiers or hashed values where possible.
  • Regulatory: keep humans in the loop for adverse actions; retain full audit trails for compliance reviews.
  • Operational exceptions: support a manual intake path for legacy suppliers and ensure the same approvals, audit, and ERP sync gates apply.

What To Do Next

  1. Workshop the minimal canonical vendor data model with procurement, finance, and compliance to align required fields and SLAs.
  2. Build modular Make scenarios in sandbox: Intake → OCR → KYC → Contract → ERP Sync. Test each step with synthetic and pilot vendor data.
  3. Run a controlled pilot with a segment of vendors, measure SLA and error outcomes, tune AI thresholds, and expand after meeting finance acceptance targets.

Sources

  • Make Developer Hub — Make API documentation: https://developers.make.com/api-documentation
  • Make Apps Documentation — Make modules & connections: https://apps.make.com/make
  • OpenAI modules for Make: https://apps.make.com/openai-modules
  • Google Docs + OpenAI integration examples: https://www.make.com/en/integrations/google-docs/openai-gpt-3
  • DocsAutomator — templated documents in Make: https://apps.make.com/docsautomator
  • TemplateDocs integration with Make: https://templatedocs.io/docs/integrations/make
  • Cadmon Systems partner examples for ERP integrations: https://www.make.com/en/partners-directory/cadmon-systems
  • Scalevise case study on Make onboarding automation: https://scalevise.com/resources/automating-onboarding-with-make-from-2-weeks-to-2-hours/

Related Reading