Toolkit · Free
Forecast Review Toolkit
Most forecast reviews argue about the number. This one puts the number on the table with its assumptions attached: 4 transparent methods blended at fixed weights, every scenario shock applied only to the weeks that have not happened yet, and the whole thing reconciled to a control total before anyone looks at it. Hand it a weekly file, get back a dashboard you can open in a meeting.
What It Does
Where the year lands, and why
It takes weekly actuals measured against plan and prior year, finds the last week that genuinely closed, and forecasts only what is left. Booked actuals are never rewritten. The forecast is driver-based rather than a trend line, so every part of the answer can be traced back to a method you can name.
Plan remainder, 35%
Trusts the budget's own phasing for the weeks that remain.
Prior-year seasonality, 35%
Scales your booked actuals by how last year's remaining period compared with its own year to date.
Performance to plan, 20%
Takes the plan remainder and tilts it by how the year has actually run against plan so far.
Run rate, 10%
Extrapolates the booked weekly rate across the weeks left. A sanity tether, deliberately weighted light.
The weights are fixed on purpose. A forecast that anyone can re-weight mid-meeting is a negotiation, not a forecast.
The Package
Two files, no email wall
Install the skill once and it is available in every Claude conversation. The template is optional: it exists so you can see the shape the toolkit expects before you go looking for your own data.
Free download: the Forecast Review Toolkit
Install once, then reuse it every month, every quarter, or every time someone asks where the year is heading.
The Skill
forecast-review-toolkit-skill.zip: the engine, the dashboard specification, and a validator that runs the finished file before it reaches you.
Download Skill ↓
Input Template
Forecast_Review_Input_Template.xlsx: the expected column layout, a column guide, and a complete sample year so the toolkit runs before you touch anything.
Download Template ↓
Requires a Claude Pro, Max or Team account. For the strongest narrative alongside the numbers, run it on Claude Opus. The maths is deterministic either way, so the model choice changes the writing, not the forecast. Something not working, or a question about your own data? Get in touch.
Setup
Running it the first time
About 2 minutes to install, then a few minutes per run depending on how much data you hand it.
1
Install the skill
Download the zip and add it to Claude as a skill. It stays available in every conversation afterwards, so this step happens once.
2
Get your data into weekly rows
One row per week per line of business, with plan, prior year and actual side by side. Open the template to see the shape. Leave the actual columns blank for weeks that have not closed, because that blank is what tells the toolkit where the forecast starts.
3
Upload it and ask
Attach the file and say what you want, in your own words. Where do we land this year? is enough. It will map your column names, find the cutoff for each source, and tell you what the data can and cannot support before it builds anything.
4
Open the dashboard
You get back a single HTML file. It opens by double-click, works offline, and has no external dependencies, so it can be emailed to a board member without anything breaking on the way.
The Output
What comes back
One interactive file, built to be read in a meeting rather than studied afterwards.
The landing numberFull-year sales and margin, split into what is already booked and what is still forecast, so nobody mistakes one for the other.
Against plan and against priorEvery headline carries both comparisons. A year that beats last year and misses plan is a different conversation from one that does neither.
The plan-gap bridgeWhere the gap to budget is actually coming from, line by line, rather than as one number to explain away.
Scenario comparisonConservative, base and optimistic, with the shocks applied only to the unbooked weeks.
SensitivityA tornado ranking of what moves the answer most, and a two-way grid that ties back to the scenario cards exactly.
Drill-down that reconcilesEvery breakdown sums to its parent, with an explicit remainder bucket rather than a silent one.
Weekly trajectoryHow the year has tracked week by week against plan and prior, and the shape the forecast expects from here.
Stated assumptionsUplifts and stances appear on screen labelled as assumptions, never dressed up as findings.
Your Data
What it needs from you
If you can produce a weekly sales report with a budget column and a last-year column, you already have enough.
Periodic actualsUsually weekly. Monthly works, with less resolution in the trajectory view.
A plan, a prior year, or bothBoth is better. With neither there is nothing to blend against and the toolkit will say so rather than invent a baseline.
Sales and margin dollarsCost is inferred as sales minus margin, so there is no separate cost column to reconcile.
Some groupingDivision or channel, and a category or line of business. That is the grain the forecast is built at.
Units, if you have themOptional. They unlock a units view alongside dollars. Plan units are never asked for, because they rarely exist.
Several feeds are fineSources that close on different weeks each keep their own cutoff instead of being forced onto one.
What it will not do
- It will not invent a grain your data does not have. With no product field, the finest level is the store, and it says so rather than showing an item view built out of nothing.
- It will not let the narrative overrule the arithmetic. The engine is deterministic and reconciles to a control total. The writing describes the numbers, never replaces them.
- It will not quietly paper over bad data. A margin equal to sales is a missing cost, not a 100% margin, and it gets flagged on the dashboard rather than in a footnote.
- It will not treat an assumption as evidence. A demand uplift you typed in is labelled as yours. Nothing you assumed comes back to you as something it found.
- It is not a replacement for judgment. It settles what the arithmetic says so the meeting can spend its time on the part that actually needs people.
The Thinking Behind It
Why it is built this way
Two pieces explain the choices, and one calculator answers the question this toolkit is usually brought in to settle.
Stop Optimizing Forecast Accuracy
Article
Why accuracy is worth buying only near a decision threshold, and why 3 weeks spent moving error from 12% to 7% can change nothing at all.
Scenarios Are Not Sensitivities
Article
The distinction this toolkit builds into its interface: coherent stories about the world, against one-variable tests of what breaks the number.
Forecast Decision Headroom Calculator
Free · Interactive
Before you run any of this, check whether a sharper forecast would change the decision. Runs in your browser, no download.