Reported operating use · Investment banking and M&A advisory
Buyer sourcing, deal coordination, diligence and banker materials
OffDeal
Production internal banking workflows; daily production-data benchmarking reported by its co-founder/CTO.
A published third-party example, not a Nuvepro customer case or evidence of a partnership.
What the source reports
OffDeal consolidated more than twelve agentic workflows into Archie. Its engineering account documents frozen production-data slices, isolated PostgreSQL schemas, in-process MCP and webhook routing for repeatable testing. It describes missed tasks and plausible but incorrect financial answers. The account explicitly says manually specifying sufficient deal context was impractical. This is implemented business work and evaluation, not a proposed demonstration.
Reported technology: Anthropic Claude Agent SDK and Claude Platform; other models also used. Scope and outcomes are source-reported.
What the enterprise environment would need
The following is a proposed project design for an enterprise doing similar work. Published details are marked; the remaining systems, data and scenarios would be agreed for your project.
- Systems and tools
- Published: PostgreSQL, BigQuery, GCS, Docker, MCP and external callbacks. Proposed: compose approved existing services.
- Data and state
- Published: OffDeal describes relationally complete production slices. Proposed: use approved extraction and de-identification while preserving relationships between deal records. If using synthetic cases, check that they retain the relationships needed for the tasks being evaluated.
- People and handoffs
- Proposed: bankers review ambiguous deal interpretation and incomplete task creation; assess agent and banker contributions separately.
- Rules and authority
- Proposed: encode confidentiality and approval boundaries without fixing the reasoning sequence.
- Normal and exception cases
- Published failures include omitted follow-up tasks and wrong valuation. Proposed: add cross-deal identity confusion and ambiguous instructions.
- Reset and replay
- Published: frozen SQL restored into per-test schemas, then teardown. Proposed: version external-provider responses separately from database state.
A proposed simulation
An illustrative practice design derived from this work, not an executed assessment.
Take a new seller mandate through buyer research, valuation questions, deal tasks and banker handoff. Change the mandate midway and introduce incomplete financial evidence.
Review gate: Reconcile valuation inputs, supported conclusions and every requested follow-up against banker-qualified outcomes; check confidentiality and missing-evidence escalation.
Challenge the result: Withhold a related deal record or change a mandate. If the assessment still rewards a confident but unsupported valuation or omitted follow-up, its readiness result is invalid.
What needs checking
- Published: comparisons, precision/recall and model judgments with concrete failure inspection.
- Proposed: banker-qualified expected outcomes and separate checks for every requested downstream task.
- Published: production-derived relational context rather than hand-authored context.
- Proposed: compare staging tool effects and human handoffs with live operating traces.
Review environment fidelity, human capability and agent performance separately. The enterprise’s accountable owners decide acceptance of the shared workflow and any operating use.
Shape this around your enterprise
Bring a similar workflow and the people responsible for it. Together, define the records and systems to use, the result your team will build or configure, and how its owners will check it. Nuvepro scopes the simulation, environment and preparation. Agree any existing-agent evaluation, implementation assistance and production responsibilities separately.
Sources and limitations
Curation review: . This is the date this source example was reviewed, separate from source publication dates.
- No independent outcome audit or demonstrated equivalence of generic scenarios to production-derived deal slices. Different published evaluation vintages should not be combined into a single score series.
- OffDeal powers every stage of M&A advisory with one Claude-based agent
- How Do You Grade an AI Investment Banker?Source publication: 2026-02-17