Forecast revenue from your own history

Separate the trend from the seasonal noise, project it forward, and get margins of error you can defend.

Feed the engine your monthly revenue. It decomposes the series, forecasts forward with intervals, and marks the dates where the behaviour of your business actually changed.

Run this with your AI assistant

meelu-analytics-mcp is a free, open source MCP server that runs on your machine. One command installs it and configures your assistant.

meelu-analytics-mcp View on GitHub All open source

The problem

Why this is hard to answer today

A spreadsheet forecast is a growth rate you picked, compounded. It is a statement of hope wearing the costume of an analysis.

Picking one growth rate hides the fact that your trend and your seasonality are moving separately.
A single line implies a certainty you do not have — the useful part is the range, and spreadsheets rarely produce one.
Nobody remembers which month the business genuinely changed shape, so old data gets weighted the same as the data that still describes you.

The solution

What the engine does instead

The engine pulls your series apart before it projects it, so the forecast rests on the underlying trend rather than on whatever last month happened to do.

Trend separated from seasonality, each reported on its own.
A forecast with margins of error that widen honestly the further out it goes.
Changepoints — the dates the series genuinely shifted — found rather than guessed.
The method it chose, recorded, so a board member can ask and you can answer.

Built on Time series, one of nine tool categories in the server.

How it works

Step by step

Step 1
Export your revenue history
Monthly is fine; weekly is better if you have it. Two years beats one, because seasonality needs at least two cycles before it can be told apart from a trend.
revenue.csv
```
month,mrr,new_customers,churned_customers
2024-01,15000,50,5
2024-02,16500,55,8
2024-03,18200,60,10
2024-04,19850,65,10
2024-05,21200,70,12
```
Step 2
Install the MCP server
Install meelu-analytics-mcp once, then restart your assistant. The install page has the one command and the README covers the rest.

Step 3

Ask for the forecast

Name the file and the horizon. Ask for the reasoning in the same breath — you will want it later.

In your assistant

Using meelu, load ~/Downloads/revenue.csv and forecast MRR
for the next 12 months. Show the trend separately from any
seasonality, give me the margins of error, and tell me
whether the series changed behaviour at any point.

Step 4
Interrogate it before you present it
A forecast you cannot explain is worse than no forecast in a board meeting.
Follow-ups
```
Which method did it use, and why that one?
How much should I trust the twelve-month figure
versus the three-month one?
```

Example output

What comes back

Shape of an answer — the numbers below are illustrative, the structure is what the engine actually returns.

Decomposition
  Trend:        +7.4% per month, steady since 2024-08
  Seasonality:  weak; December +4%, January -3%
  Changepoint:  2024-08 — growth rate stepped up here.
                Months before it are a different business.

Forecast (from 24 months of history)
  Month     Central    Range
  +1        $52,400    $50,100 – $54,800
  +3        $60,900    $54,200 – $68,400
  +6        $74,800    $61,500 – $91,000
  +12       $95,600    $66,000 – $138,000

Trust: medium — 24 points is enough for a trend, thin
for seasonality. The twelve-month range is wide because
it deserves to be.

The widening range is the honest part. A forecast that claims the same precision twelve months out as it does next month is not a forecast.
Changepoints are worth as much as the projection. Knowing that August was the month the business changed tells you which history is still describing you.

In practice

Who asks this

Founder preparing to raise

An investor will ask where revenue lands next year, and a made-up number will not survive the follow-up question.

Bring a range with the method attached. Being able to say why the interval is wide is more convincing than a confident single number.

Operator planning headcount

You want to hire in Q4 and need to know whether revenue supports it.

Plan against the lower bound rather than the central line. That is the decision the interval was for.

Anyone reporting to a board

Last quarter's forecast missed and you have to explain it.

The changepoint analysis usually explains it — something shifted, and now you can name the month it happened.

FAQ

Common questions

How do you forecast revenue?

Take the historical series, separate it into trend, repeating seasonal pattern and noise, then project the trend and seasonal parts forward. For subscription businesses the alternative is a bottom-up build: current recurring revenue, plus expected new business, minus expected churn and contraction, plus expansion. Do both when you can, because the gap between a statistical projection and a bottom-up plan is where the optimism hides.

How much history do you need to forecast revenue?

Twelve data points give you a trend. Twenty-four give you a chance of separating seasonality from that trend, because a yearly pattern has to repeat before anything can tell it apart from a one-off. Below twelve you are extrapolating a line and should say so out loud rather than presenting a figure with a decimal point.

What is the difference between a forecast and a target?

A forecast is what the evidence says will probably happen. A target is what you have decided to aim for. Confusing the two is the most common failure in revenue planning: a target dressed as a forecast makes everyone downstream plan hiring and spending against a number nobody actually believes. Produce the forecast first, then set the target against it, and keep the gap visible.

How do you forecast a seasonal business?

Measure the seasonal pattern rather than guessing at it. Decomposition splits the series into trend and repeating seasonal effect, so you can see how much of December is a pattern that happens every year and how much is real growth. Without that split, a strong season reads as growth and the following quarter reads as a collapse, and both readings are wrong.

Why should a forecast include a range?

Because a single number is always wrong and gives no sense of how wrong. A prediction interval says where the actual figure is likely to land, and it widens the further out you look, which is honest: next month is far more knowable than next year. A forecast presented without a range invites people to plan as though it were a fact.

Keep going

Related use cases

Churn prediction

Work out what predicts a customer leaving — and who is at risk now

Cohort retention

See which signup months held up, and find the month things changed

Stripe analysis

Turn a Stripe export into churn drivers, forecasts and cohorts

Try revenue forecasting on your own data

Install meelu-analytics-mcp once and every use case on this site is available to your assistant — it is one server with forty-five tools, not six separate products.