How to use Claude for pitch decks, month-end close, and KYC

Anthropic shipped 10 ready-made Claude templates for finance teams on May 5, 2026. ^[1] Three of them are worth your Monday morning. Seven aren't.

It's the same vertical-product play Anthropic ran with Claude Design: pick a high-frequency surface, ship a template, ship the audit log with it.

Most AI finance demos die between the pilot and the audit log. Anthropic's new templates ship with the audit log on day one, the thing a regulator asks for. That changes the calculus for a specific subset.

TL;DR

• Three Claude finance templates are ready for this week: Pitch agent, Month-end closer, KYC screener. All 10 ship as open-source plugins in github.com/anthropics/financial-services.
• The other seven need more handholding than they save. Watch, do not ship.
• Claude scored 64% on a finance task test (Vals AI). Best in class today. 64% is not 99%.

How do you install any of these templates?

You install the same way for all 10. Pick one of three paths: ^[1]

• Claude Cowork (paid plan, easiest): Settings → Plugins → Add plugin → paste https://github.com/anthropics/claude-for-financial-services. Pick the agents you want.
• Claude Code CLI (developer-friendly): claude plugin marketplace add anthropics/claude-for-financial-services, then claude plugin install financial-analysis@claude-for-financial-services (the core skill bundle). Add specific agents after that.
• Claude Managed Agents (Platform, public beta, headless): export your ANTHROPIC_API_KEY, clone the repo, run scripts/deploy-managed-agent.sh kyc-screener (substitute whichever agent name you want).

Install the financial-analysis core plugin first. Every agent depends on it. Then add one specific agent and try it on real work.

How do you use Claude for pitch decks?

Pitch decks are the highest-frequency, lowest-risk finance task. Start here.

The Pitch agent reads a target's filings and produces three artifacts: a comps model in Excel, a pitch deck in PowerPoint, and a cover note in Outlook. ^[1]

Slash commands inside the agent: /comps (comparable companies), /dcf (discounted cash flow), /lbo (leveraged buyout model), /cim (confidential information memo), /ic-memo (investment committee memo).

Try this week: install the Pitch agent. Pick one upcoming meeting. In Cowork or Claude Code, type /comps and hand it the target's last four quarters of public financials. Compare its draft to what your analyst produced last week. If it gets you 70% of the way there, the template earns its place.

How do you use Claude for month-end close?

The monthly accounting close is recurring, structured, and audit-trailed. It is the second-best place to start.

The Month-end closer reads your close checklist, drafts journal entries, reconciles accounts, and produces close reports. ^[1] The audit log records every action, so each entry traces back to a source if an auditor asks. Permission controls let you scope what the agent can write to. ^[1]

Try this week: install Month-end closer alongside the financial-analysis core. Run it in parallel with your normal close for one cycle. Compare what it flags against what your team flags. The first month is calibration, not replacement.

How do you use Claude for KYC and customer checks?

KYC stands for "know your customer." It is the regulatory check every bank, broker, and fintech runs before opening an account. It is high-volume, structured, and document-heavy. That is exactly the shape of work where a model earns its keep.

The KYC screener parses entity files and onboarding documents, runs them through the rules engine, flags gaps, and packages escalations for compliance review. ^[1] The compliance officer still signs off. Claude removes the document shuffling, not the judgment call.

Try this week: route one new-customer onboarding through Claude alongside your normal process. Time both. If Claude finishes document review in a fraction of the time, the template is ready for a wider pilot.

What about the other seven templates?

Watch them. Do not ship them yet. Each links to its README: ^[1]

• Meeting prep agent. Qualitative work. Needs human context.
• Earnings reviewer. Analyst nuance. Hard to verify quickly.
• Model builder. Models hinge on assumptions. Those are the human's job.
• Market researcher. Open-ended task. Hallucination risk is real.
• Valuation reviewer. Judgment call. Not a template task.
• GL reconciler. Close to ready. Worth piloting if your close is mature.
• Statement auditor. Compliance stakes too high for a first try.

Re-evaluate this list after one major model update. The line moves fast.

Is Claude actually good enough at finance work?

Claude Opus 4.7 scored 64.37% on Vals AI's Finance Agent benchmark. ^[2] That is the highest of any model tested as of May 2026. Sonnet 4.6 follows at 63.33%, then Muse Spark and DeepSeek V4 at around 60%. ^[2]

The benchmark uses 537 SEC-filings questions written by analysts at Goldman Sachs, Silver Lake, and Citadel. ^[2]

64% is not 99%. Use Claude as a first-draft generator for the three tasks above. Do not use it as a replacement for the senior person who signs the work.

Key takeaways

• Three Claude finance templates are ready for Monday: Pitch agent, Month-end closer, KYC screener. Each is a structured task with a human reviewer at the end.
• One install path for all 10 templates: Cowork plugin, Claude Code CLI, or Managed Agents. Install the financial-analysis core plugin first.
• Seven others need more time: meeting prep, earnings review, model building, market research, valuation, GL reconciliation, statement audit.
• The audit log is the unlock. Earlier AI finance tools could not be defended to a regulator. These can.
• 64% is best in class, not perfect. First-draft generator, not a replacement for sign-off.

If this was useful and you want more plain-English takes on what AI is actually ready for in finance and ops, follow on LinkedIn. I drop one of these a week.

Footnotes

1. Anthropic, "Introducing finance agent templates," May 5, 2026. https://www.anthropic.com/news/finance-agents and the open-source repo at https://github.com/anthropics/financial-services [↩] [↩ ^[2]] [↩ ^[3]] [↩ ^[4]] [↩ ^[5]] [↩ ^[6]] [↩ ^[7]]

2. Vals AI, "Finance Agent v1.1 leaderboard," accessed May 6, 2026. https://www.vals.ai/benchmarks/finance_agent [↩] [↩ ^[2]] [↩ ^[3]]