Most reporting problems start upstream. The CRM counts a customer once per contact, billing counts one per invoice address, and the two never agree on how many customers you have. We find those mismatches before anyone builds a forecast on top of them.

What we check

• Duplicates and orphaned records
• Fields that changed meaning over time, like a “region” that was a state until 2021 and a sales territory after
• Gaps: days, stores or products with no rows at all
• Refresh times: which report is a day old and which is a week old

What you keep

The checks are plain SQL that runs in your warehouse every night. When one fails, it emails the owner of that source, not us.

Most companies we meet have forty dashboards and three definitions of revenue. We start by writing down what each number means, then build the smallest report that answers the questions asked in your weekly meeting.

How we build it

• Interview the five to ten people who use the numbers
• Write one definition per metric, next to the query that computes it
• Build the dashboard in Looker Studio, Power BI, Metabase or Tableau, whichever you have
• Retire the reports nobody opened in the last ninety days

We forecast whatever drives your costs: units, orders, calls or staff hours. Each forecast comes with a back-test, so you can see how the model would have done last year before you trust it this year.

What we model

• Seasonality and holidays, including the ones that move every year
• Promotions and price changes, if you have the history
• Stock-outs, which hide real demand in your sales data
• New products, using the closest existing ones as a guide

When we say no

With less than eighteen months of history, a forecasting model rarely beats a well-kept spreadsheet. We tell you that in the first week and bill for the first week only.

Most price lists were set once and adjusted by feel. We use your order history to estimate how volume moves when price moves, then set prices where margin is highest without losing the customers you want to keep.

What we usually find

• Discounts given to customers who would have paid full price
• Products priced below cost once shipping and returns are counted
• Price steps that push customers down to the cheaper tier

Half the tests we are asked to review were too small to detect the effect they were looking for. We fix the sample size and the stop date before launch, so nobody ends a test early because the first week looked good.

What we test

• Prices, offers and free-shipping thresholds
• Checkout and sign-up changes
• Email timing and content
• Sales scripts, by randomizing which reps use them

Segments only help if someone treats each one differently. We build them from order history, not demographics, and agree with your team which action goes with each segment before we hand them over.

Questions we answer

• Which channel brings the customers who stay longest
• What the last three orders of a customer who left look like
• How much you can spend to win back a lapsed account