Articles
Articles
Each article answers one question a company asks about its data or about using AI for analysis, with its sources listed at the end.
Is AI-made analysis cheaper?
Not if it is checked. AI speeds up doing the work, not checking it, so a five-day report becomes four, and one used to go twice as deep takes seven. A worked example priced against a decision, and the cases where AI does cut the cost.
How to check an AI-generated report against its sources
List the claims, let a script confirm each quote is really in its source, then read what no script can: whether it supports the sentence.
Can AI check AI's work?
Partly. Three AI reviewers working apart and a judge ruling by one test can catch work built wrong. Not a wrong brief or a shared blind spot, which is why a person still signs.
How many customers do you need to test a price change?
It depends on what you measure: the customers who leave after a price rise need far smaller groups than a change in sign-ups does. The formula, a worked example on an invented company, and what to do when the numbers are out of reach.
What KPIs should a Series A startup track?
For a subscription business, five, each defined exactly, with a worked example on an invented company and where each one usually goes wrong.
Can AI replace a data science team?
For questions defined in advance, with the data shaped to answer them, increasingly yes. For the open questions a data science team exists to answer, not yet. The benchmarks on both sides, and what they mean for a company with no data team.
How to check the numbers in an AI-generated report
Reading the report will not show you which numbers are wrong. Six checks that will, from tracing each figure to its source to re-running the analysis on a second sample, each with a worked example on invented numbers.
How long does it take to build a customer segmentation model?
The model is the short part; the data, the decisions and the checks take the rest. A step-by-step estimate for an invented online shop.
What is reducing churn by 1% worth?
One point less churn keeps one more customer in a hundred each month. The arithmetic, why the answer is a range rather than a number, and what it leaves out.
Start here