# Rolling Forecast Builder > Blends actuals-to-date with a forward forecast via a bundled calculator, bridges the result against the original plan, and flags when the forecast's implied run rate jumps well above what's actually been achieved so far without a named reason. Use whenever the user needs an updated full-year view partway through the year, wants a rolling or continuous forecast rather than a once-a-year budget, or… - **Type:** Skill - **Install:** `agentstack add skill-natan-mohart-24-finance-skills-for-claude-rolling-forecast-builder` - **Verified:** Yes — security-reviewed for prompt injection and unsafe behavior - **Seller:** [Natan-Mohart](https://agentstack.voostack.com/s/natan-mohart) - **Installs:** 0 - **Category:** [Agent Skills](https://agentstack.voostack.com/c/agent-skills) - **Latest version:** 0.1.0 - **License:** MIT - **Upstream author:** [Natan-Mohart](https://github.com/Natan-Mohart) - **Source:** https://github.com/Natan-Mohart/24-finance-skills-for-claude/tree/main/skills/rolling-forecast-builder ## Install ```sh agentstack add skill-natan-mohart-24-finance-skills-for-claude-rolling-forecast-builder ``` Requires the [AgentStack CLI](https://agentstack.voostack.com/docs/cli). Works with Claude Code, Cursor, and any MCP-compatible agent. ## About # Rolling Forecast Builder ## When to use Use whenever the year is partway through and the full-year view needs updating with real actuals, replacing a forecast that either ignores actuals entirely (still showing the original budget) or quietly assumes an accelerating pace for the remaining months with no stated reason. ## What it does Blends actual results for completed months with a forecast for remaining months via a bundled calculator, computes the resulting full-year view, bridges it against the original plan, and explicitly flags when the forecast run rate for remaining months is meaningfully higher than the actual run rate achieved so far — the most common way rolling forecasts quietly turn optimistic without anyone noticing. ## Method 1. **Pull actuals for every completed month**, not an estimate — the forecast is only as trustworthy as the actuals it's anchored to. 2. **Build the forecast for remaining months from real drivers**, using the same driver-based-budget-builder discipline, updated with anything learned from the actuals so far (a channel that's underperforming, a cost that's running higher than planned). 3. **Run the bundled calculator** (`scripts/rolling_forecast.py`) to get the full monthly view (actual and forecast months together), the resulting full-year total, and the bridge against the original annual plan. 4. **Read the run-rate comparison explicitly.** If the forecast's monthly run rate for remaining months is more than 15% above the actual run rate achieved in completed months, that's a flag worth resolving before the forecast goes out — either there's a specific, named reason (a signed deal, a seasonal pattern, a planned launch) or the forecast is quietly assuming things get better with no basis. 5. **State the bridge to plan in both dollars and percentage**, and be explicit about whether the business is tracking ahead, on, or behind the original commitment — a rolling forecast's job is to catch a miss early, not to smooth it into invisibility. 6. **Update the forecast on a fixed cadence** (monthly or quarterly), not ad hoc — a rolling forecast that only gets refreshed when someone remembers to do it defeats its own purpose. 7. **Carry forward the reasons behind material changes** from one forecast cycle to the next, so a pattern of consistently over-optimistic remaining-month forecasts becomes visible over time, not just within a single cycle. ## Inputs - Actual results by month for completed months - Forecast for remaining months, built from updated drivers - The original annual plan to bridge against - Config saved as JSON matching the format documented at the top of `scripts/rolling_forecast.py` ## Output format Month-by-month table of actuals and forecast with running cumulative total; full-year forecast total; bridge to original plan in dollars and percentage; actual vs. forecast monthly run-rate comparison with an explicit flag if the forecast assumes unexplained acceleration. ## Example Five months of actuals average $1.9M per month; the forecast for the remaining seven months averages $2.24M, an 18% jump the calculator flags automatically. The team traces it to a genuinely signed enterprise deal starting month 7, a specific, defensible reason, so the forecast stands, but the flag forced that justification to be stated explicitly rather than the acceleration passing through unquestioned. ## Common pitfalls - Forecasting remaining months at a faster pace than actuals have shown, with no specific driver behind the acceleration. - Refreshing the forecast irregularly instead of on a fixed cadence, so it drifts out of date exactly when it matters most. - Reporting the bridge to plan only in dollars or only in percentage, when both tell a different part of the story. ## Source & license This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code. - **Author:** [Natan-Mohart](https://github.com/Natan-Mohart) - **Source:** [Natan-Mohart/24-finance-skills-for-claude](https://github.com/Natan-Mohart/24-finance-skills-for-claude) - **License:** MIT Install and usage instructions live in the source repository linked above. ## Pricing - **Free** — Free ## Security capabilities Automated source analysis of v0.1.0 — what this tool can access: - **Network access:** no - **Filesystem access:** no - **Shell / process execution:** no - **Environment & secrets:** no - **Dynamic code execution:** no *"Yes" means the capability is present in the source — more access means more to trust, not that it is unsafe.* ## Versions - **0.1.0** — security scan: passed — Imported from the upstream source. ## Links - Listing page: https://agentstack.voostack.com/l/skill-natan-mohart-24-finance-skills-for-claude-rolling-forecast-builder - Seller: https://agentstack.voostack.com/s/natan-mohart - Browse the marketplace: https://agentstack.voostack.com/browse --- Listed on AgentStack — the marketplace for AI agent skills and MCP servers. Every listing is security-reviewed. Creators keep 70%.