New York-based event-driven investment manager
From three to four hours of deal research to a draft sheet in under eight minutes.
A coordinated system of seven specialist agents brings more than five sources into one cited deal file while keeping calculations, exceptions and investment judgment visible for analyst review.
Measured result
Under 8 min
to produce a single-page draft that previously took at least three to four hours.
- Previous analyst time
- 3–4 hours
- Sources in the recurring process
- 5+
- Time to a draft
- Under 8 min
- Reviewable deal file
- 1 page
An event-driven analyst previously moved across multiple websites and more than five data sources to assemble one merger-arbitrage deal sheet. A coordinated system of seven specialist agents now produces the draft in under eight minutes as a single page. The analyst can spend the remaining time reviewing evidence, resolving judgment calls and deciding what matters for the investment.
The speed comes from gathering, structuring and calculating the material. It does not remove analyst review. A wrong share count, a mishandled collar or a confident sentence that outruns its source can distort the discussion even when the page looks polished.
The first idea was a template for Claude. The work became an orchestrated multi-agent workflow that could prepare a repeatable deal file around the manager's method, calculate defined terms and leave the investment judgment visible for the analyst.
Inside the system
Seven agents. One controlled run.
The workflow coordinates parallel research, deterministic calculation, an analyst methodology checkpoint, narrative analysis and independent verification before rendering the cited deal sheet.
Illustrative deal run
Fictional transaction · production workflow structure
Source layer
Authorized inputs are collected with their provenance intact.
Primary filings
Deal documents · current filings
Licensed market data
Prices · estimates · fundamentals
Deal configuration
Terms · lenses · methodology
Current developments
News · dockets · material changes
Provenance carried forward
Every material claim and figure can retain a path back to its source.
Orchestration layer
State, dependencies, retries and review points
Parallel research fan-out
Runs concurrentlyDeal terms
Transaction structure and primary documents
Complete
Market & company
Prices, estimates and company snapshots
Complete
Current developments
Filings, news and transaction changes
Complete
Analysis & decision
Human checkpointPeer analysis
Relevant comparison set and downside inputs
Complete
Methodology
Analyst-selected downside and probability method
Selected
Synthesis & control
Independent gateNarrative analysis
Rationale, regulation, risks and catalysts
Complete
Verification
Re-derive figures and test cited claims
1 review flag
Run timeline
Completed under eight minutes
00:00
Run opened
00:48
Sources joined
02:36
Math complete
04:10
Method selected
06:32
Narrative complete
07:42
Verified & rendered
The agent team
Seven specialist agents. One reviewable deal sheet.
An orchestration layer coordinates distinct research and verification jobs, runs independent work in parallel and brings the results into one consistent file. Deterministic calculations and analyst checkpoints sit between the agents rather than being hidden inside generated prose.
- Agent 01
Deal terms
Extract the transaction structure, dates and non-standard terms from primary deal documents.
- Agent 02
Market & company data
Assemble prices, fundamentals, estimates and the target and acquirer snapshots.
- Agent 03
Peer analysis
Build the relevant comparison set and inputs for downside and relative-value work.
- Agent 04
Current developments
Review recent filings, configured news sources and material changes around the transaction.
- Agent 05
Methodology
Apply the selected downside and probability methods with deterministic calculations.
- Agent 06
Narrative analysis
Use the chosen sector lens to organize rationale, regulatory issues, risks and catalysts.
- Agent 07
Verification
Recheck figures and evidence, then surface conflicts and unresolved items for analyst review.
Configured to the firm
Put the manager's method into the workflow.
The analyst supplied the rules that a generic deal summary would miss. Ticking fees and collars belong in deterministic calculations. Share count must be chosen under an explicit rule. Governance issues need their own treatment, and conflicting sources need a hierarchy the team can inspect.
A deal template captures the companies, status, consideration, dates and non-standard terms. A lens adds questions relevant to the sector or buyer type. A methodology choice sets the downside and probability approach for that transaction. The deal can then enter scheduled monitoring.
That structure lets the same core workflow handle different transactions without pretending every sector should be analyzed under one checklist. The lens library remains editable as the firm's practice evolves.
Verification
Review the narrative as carefully as the numbers.
A calculation can be tied to a formula and source inputs. Interpretive prose needs a different check. Does a statement about why a deal might change course follow from the evidence? Is it the source's claim or the firm's investment view?
Testing found that a polished sheet could still contain an unsupported narrative sentence. Verification therefore became an active part of the build. When a source conflicts or a claim cannot be supported, the analyst must be able to resolve it before the sheet is relied on.
Iteration on live transactions changed which figures the agent used, how it selected a method, what stayed as a cited third-party link and where the sheet asked a person to make a choice.
What changed
A faster, more consistent starting point for deal review.
The firm has a working, configurable seven-agent deal-file workflow that turns a process taking at least three to four hours into a single-page draft in under eight minutes. Analyst review remains part of the process, supported by source links, deterministic calculations and visible verification flags.
This is a workflow result. It is not an investment-performance claim.
The service behind this work
Managed AI Agents
Agent scoping, building, evaluation, deployment, monitoring and ongoing operation across research and operations.
See more agent workflows →Read next
Managed AI agents
Firm-wide research agents run as a managed service
Earnings and news agents covered by deployment, security, versioning, tracing, evaluations, monitoring and ongoing optimisation.
AI training
PDF reports into model-ready data
Three working sessions on specialist extraction, Excel model updates and agents the investment team could test.
Does your team rebuild the same deal file repeatedly?
Start with a real transaction, the existing analyst worksheet and examples of decisions that are easy to get wrong. The first agent should show its sources, compute what can be computed and mark the point where a person decides.