Customer Success, Implementation, and Operations Specialist with 10+ years managing enterprise post-sales lifecycles, institutional client portfolios, and complex system integrations across SaaS, fintech, and IoT platforms.

At ButterflyMX (March 2025 - May 2026) my core role was Implementation Project Specialist: I owned implementation lifecycle health, on-site activation, and the CARR-to-Live-ARR billing gate across thousands of customer implementations. I am currently seeking Customer Success and Implementation roles where operational scale intersects with applied AI workflows. Alongside that core role I volunteered as the team's Credal.ai power user, building Salesforce-backed agent workflows that automated the billing audits and account triage I previously handled manually. That split runs through everything below: core delivery first, AI as operational leverage I earned the right to build.

Before ButterflyMX I spent over a decade leading customer onboarding, executive relationship management, and revenue retention across tier-one financial data and trading platforms (Refinitiv, Fitch Solutions, Enfusion).

Before and alongside my applied AI work, my career is grounded in high-volume enterprise customer delivery and account management:

• ButterflyMX (PropTech & IoT Access Control, Implementation Project Specialist, March 2025 - May 2026 - core role): Governed implementation lifecycle health, on-site activation milestones, and CARR-to-Live-ARR billing gates across an active pipeline of 5,162 customer implementations and 8,214 live property records. Maintained a 35+ daily intervention cadence across contractors and property managers. This was my core role; the AI systems below were volunteered on top of it.
• Refinitiv / Thomson Reuters (Risk Intelligence & Compliance): Managed a $4.8M ARR ($400K MRR) institutional portfolio across 30+ enterprise accounts and 2,000+ end users (World-Check KYC/AML, Sanctions, PEP screening), driving adoption and renewal outcomes through proactive usage-based risk analysis and QBRs, and supporting mission-critical U.S. government agency workflows.
• Fitch Solutions (Credit Ratings & Macro Intelligence): Customer Success Manager, Global Accounts supporting a $42M institutional book of business across tier-one banks and financial institutions, leading onboarding, usage-monitored outreach, and renewal strategy for fixed-income intelligence feeds.
• Enfusion (Fintech / Investment Operations): Provided technical and integration support for 15+ hedge funds and asset managers on a cloud-native investment management platform, covering onboarding support, reconciliation-break triage, and third-party connectivity coordination.

Architected an enterprise multi-agent retrieval and intelligence engine separating nondeterministic business reasoning from deterministic, injection-safe execution:

• Orchestration Tier: Evaluates natural language business requests, handles user disambiguation, maps operational vocabulary to API field taxonomies, and enforces human-in-the-loop write approval gates.
• Delegated Execution Tier: A strictly read-only sub-agent compiling mapped requests into validated, injection-safe SOQL queries executed against live Salesforce REST APIs.
• Scale: Automated account triage, billing audits, and lifecycle risk across 5,162 active customer implementations and 8,214 live property records.
• Customer Operations Impact: Enabled instant, zero-friction account health audits across 5,000+ implementations without waiting on engineering sprint cycles.

• Diagnosed an out-of-the-box 96% semantic search failure/warning rate when using standard dense vector similarity on programmatic API field names and schemas.
• Replaced naive vector RAG with a structured, metadata-gated Smart Filtering v3 retrieval architecture enforcing strict similarity scoring thresholds to eliminate LLM hallucinations.
• Validated against a mined production corpus of 17,086 SOQL queries across 96 Salesforce objects and 1,534 fields.
• Customer Operations Impact: Ensured 100% reliable data retrieval for customer-facing teams, preventing hallucinations during account reviews.

• Solved silent context-window truncation and data loss caused by massive unstructured email reply chains (5–20KB per record in CRM activity logs).
• Engineered a Two-Pass Task Triage Algorithm: Pass 1 scans lightweight subject lines and risk trigger words (cancel, permit, TCO, refund, go-live); Pass 2 selectively pulls deep conversational history only for high-signal records while filtering out internal notes.
• Result: Compressed per-project activity payloads by 94% (~165KB down to ~10KB) prior to model ingestion.
• Customer Operations Impact: Prevented dropped communication history across multi-year accounts, surfacing critical churn and go-live signals in seconds.

• Uncovered and resolved a 95% data omission rate in customer reference discovery caused by continuous building management turnover and polysemic CRM schemas.
• Engineered multi-source relational traversal logic across operating accounts, physical building entities, and corporate parent hierarchies.
• Customer Operations Impact: Prevented customer misattribution during rapid building management transitions, preserving continuous account ownership.

Enterprise TypeScript framework for LLM tool calling, secure data connectors, and agentic workflows.

• salesforce/getCleanActivityRecords (PR #559 — Open, under review)
  • Built an end-to-end data normalization and security pipeline for AI agents to ingest Salesforce email communication histories.
  • Implemented AST-style (Abstract Syntax Tree) SOQL injection guards, regex-based HTML/quote-chain stripping, and deterministic thread deduplication.
  • Engineered a 133 Jest unit test suite validating edge cases, malformed payloads, and injection defense.
• readCommentsOnDoc (PR #572 — Open, under review)
  • Engineered Google Docs comment and anchor text extraction with deterministic $O(n)$ merging of Google Drive API metadata and raw OpenXML (OOXML) structures.
  • Implemented decompression stream security guards to protect against zip-bomb vulnerabilities.
  • Code lives in my working fork.
• getSpreadsheetMetadata (PR #534 — Merged)
  • Designed metadata extraction using field masks to minimize memory footprints and reduce API token overhead.
  • Built a fallback-driven XLSX download pipeline to bypass native Google Sheets API timeout limits on large workbooks.

• Batch 422 Fallback Handling (PR #661 — Merged)
  • Fixed GitHub API 422 review fallback handling in Alibaba's automated AI code review platform.
  • When batch review creation fails due to line validation mismatches, surviving comments are grouped into a single consolidated fallback review rather than degrading into $N$ separate comments.

• Restored stale municipal endpoints and metadata for Boston and New York City open data portals within the global catalog (PR #418 — Merged).

Note on private work: the property-intelligence tooling behind my Socrata SODA / ArcGIS REST claims (municipal permits, ownership, violations, TCO research) currently lives in a private repository — sanitized architectures and case studies available on request.

• LinkedIn: linkedin.com/in/matthew-betancourt-csm
• Email: me@mattbetancourt.com