# Your AI agent is brilliant. Now make it trustworthy. Rockhopper is the second-set-of-eyes and sign-off layer that sits between AI agents and your financial spreadsheets — so your team can move at agent speed without losing human control. ## AI Agents - Claude - GPT - Gemini - Custom ## Agent output Governance · Control · Transparency - Cell-level diff - Change classification - Materiality gates - Audit log - Revert - Governed output Excel Google Sheets ## The trust gap AI agents are coming for your spreadsheets. The controls haven't caught up. 94% of enterprise spreadsheets already contain errors — and that's before AI agents start touching them. When an agent builds a model or updates a forecast, who's accountable? How does the Controller sign off? What goes in the audit trail? ## The Future of AI Assurance A CFO's case for independent oversight of agentic AI workflows. If AI is building your spreadsheets, should it also be checking them? We'd argue no — and we're not alone. ### 01 · Invisible work #### No audit trail for agent work AI agents modify cells, update formulas, and restructure models without leaving a traceable, approvable record. Teammates, managers, and auditors can't review what they can't see. > Agent updated 412 cells in Revenue Model v7 > No reviewer assigned > No change log entry created ### 02 · Accountability gap #### AI can't take on liability Claude can build the model. Claude cannot sign off on SOX controls. Someone still has to. Rockhopper is the bridge between agent output and human accountability. > "Who approved the agent's forecast?" > "Which control owner signed off?" > "Where's the SOX evidence?" ### 03 · Missing frameworks #### Governance frameworks don't include agents SOX controls, IFRS compliance, and internal review processes were designed for humans. They have no provision for an AI that changed 400 cells overnight. - SOX 404: No control for agent-initiated changes - IFRS 16: Lease model updated by agent — no review - Internal audit: Agent bypass not in risk matrix ### 04 · Machine speed #### Errors compound at machine speed When an AI agent makes a mistake in a financial model, it doesn't make one mistake — it makes the same mistake across every connected cell, tab, and downstream report, instantly. - 1 formula error → 847 affected cells - 3 downstream reports corrupted - Discovered 14 days later at close ## What Rockhopper does Infrastructure for the agent-augmented finance team. Rockhopper doesn't compete with your AI agent. It makes your AI agent deployable — by giving every change, regardless of who made it, a transparent, governed, sign-off-ready record. ### 01 — Transparency #### See everything the agent touched Every cell changed by an AI agent is logged, summarized in plain English. What changed, how much, by whom — human or machine — and what it means for your numbers. - Cell-level diff - Agent attribution - Impact narrative ### 02 — Control #### Human in the loop, always Route agent outputs through the same review and approval workflows your team already uses. Set materiality thresholds, require sign-off above a dollar amount, flag changes that cross SOX control boundaries. - Review workflows - Materiality gates - SOX controls - Sign-off routing ### 03 — Governance #### Audit-ready from day one Rockhopper maintains a complete, immutable record of every change — human or agent — with timestamps, version snapshots, and reviewer sign-offs. Your auditors get a log they can actually use. - SOX compliance - Immutable log - Version snapshots - Auditor export ## How it works The review layer every agent relies on. 1. **Agent performs analysis or builds the model** Your AI agent — Claude, a custom model, or any third-party — runs its task. It updates assumptions, builds a forecast, reconciles data, or flags anomalies. All changes are captured by Rockhopper in real time. 2. **Rockhopper classifies and summarizes every change** Rockhopper automatically classifies each change — structural vs. input, material vs. immaterial — and generates a plain-English change log summary and visualizer for the reviewer. No hunting through cells. 3. **Material changes are routed for human review** Based on your configured thresholds, changes above a materiality level or crossing a SOX control boundary are automatically routed to the right reviewer — Controller, CFO, or auditor — for sign-off. 4. **Full audit trail created automatically** Every action — agent and human — is written to an immutable log with timestamps, reviewer identity, and version snapshots. Your auditors get a complete, exportable record without anyone building it manually. ## Enterprise governance Built for companies that can't afford to get it wrong. Rockhopper was designed for organizations with real compliance obligations — SOX-controlled environments, IFRS reporters, and regulated industries where every number needs a name behind it. ### SOX control integration Define control thresholds by dollar amount, formula type, or tab. Any agent or human change that crosses a boundary triggers an automatic review workflow. ### IFRS & GAAP compliance tracking Rockhopper flags changes to cells tied to recognized IFRS or GAAP line items, ensuring disclosures stay accurate when agents update underlying assumptions. ### Segregation of duties — preserved for agents AI agents inherit the same workflow controls as human users. An agent cannot self-approve a material change any more than a preparer can self-approve a journal entry. ### Immutable audit log — auditor-ready export Every change, approval, and sign-off is written to a tamper-proof log. Export a full package for external auditors in one click, including agent attribution for every action. ## The future of finance Agents and humans. Both need Rockhopper. Scaling laws haven't ended. Models will be 5x more capable within two years. Every financial document will be touched by an agent. The question isn't whether AI enters your finance team — it's whether you have the infrastructure to let it in safely. > **5×** More capable AI models within 2 years. Finance teams that adopt agents now build an insurmountable productivity advantage. > **100%** Of financial documents will be touched by AI within 5 years. The governance layer is not optional — it's inevitable. > **$0** Liability that any AI agent can accept. Rockhopper holds the accountability layer that models legally and practically cannot. > **1** Review layer that works with any agent — Claude, GPT, or custom — and any workflow. Bring your own AI. Rockhopper handles the rest. > “The future of audit is a collaboration between agents and humans — two separate species working together. The tools they use must welcome both equally.” > **Sebastian Perez Lawrence** > Cofounder & CEO > Rockhopper Inc. ## The future of finance Rockhopper is the assurance infrastructure your AI-augmented finance team needs before the models get any faster.