Our history is short, or the tracking changed halfway through.
I forecast anyway and tell you how much to trust it. A forecast with a stated weakness beats a hopeful line, and the first graded window tells us how good the data really is.
The problem
You plan next quarter's budget with one number and a hope.
How it works
Three steps, in order.
- Baseline.I take the history of the number you care about, find its seasonality and its breaks, and tell you what a forecast can and can't promise on your data.
- Forecast and plan. You get the 28-day forecast with its range, the scenarios, the odds per option, and the design for the test that will grade the next big decision.
- Measure and grade. The campaign runs. The lift is measured against the control, the forecast is graded at each checkpoint, and the decision goes on the record.
What you end up with
- A forecast with a range. Your main number for the next 28 days, with a 90% band and the history that produced it.
- Odds per option. Hold, push or cut, each with its probability of clearing the target.
- Tests designed before the spend. Geo or holdout, with a power check, run by you or with me.
- A record that grades the quarter. What was expected, what was decided, what happened.
Proof
Two examples, as published.
A marketing team ran out-of-home campaigns across three Canadian cities and needed to know if they worked. I built counterfactual cities from a donor pool to isolate the causal lift in branded web traffic, with directional confidence intervals and a framework for designing measurement into future buys. Genuine lift was separated from background noise.
The forecast on this page is how TrueLift, the tool I built, grades its own forecasts: a 90% band, four checkpoints, misses on the record.
Who you'd work with
I'm Martin Guzman. I've spent twelve years building marketing measurement: warehouses, attribution, forecasting, and production machine learning. I currently run analytics for the marketing team at a B2B AI company. You work with me directly, with no account managers and no handoffs.
The offer
Measurement Assessment
You get
- A first 28-day forecast of your main number, with a range
- The decisions worth measuring in the next two quarters, with the method for each
- A test design, with a power check, for the next big one
- A written plan for the foundation work, if the data needs it first
What this won't do
- It won't give you a forecast to bank on if the history is under a year old. I'll forecast anyway and say how much to trust it.
- It won't fit a mix model. If your spend history supports one, the plan says so and what it takes.
Book a quick discovery call
Questions
Do we need marketing mix modeling?
Probably not yet. MMM needs a few years of spend variation across channels. For most companies under 500 people I get further with a forecast, pipeline attribution and one well-designed test, and I'll say so.
How accurate is the forecast?
It carries a 90% band and gets graded against reality at days 7, 14, 21 and 28. You see the hit rate, misses included. Accuracy is a number on the record, not a promise.
Do you need our ad spend data?
For attribution and mix work, yes: spend by channel by day. For the forecast alone, the history of the number you want to forecast is enough to start.
Can we run the tests ourselves?
Yes. The test design, the power check and the readout template are written to be run by your team. I can run the first one with you.
What comes after, and what comes first
AI for marketing analytics
Marketing data and attribution, if the foundation needs work first
Start with the assessment.
In three weeks you'll have a forecast with a range and a plan to measure the next big decision.