Analytics services

Usually because the logic behind each metric lives in reports, SQL views and spreadsheets, and only the data team knows which version is right. Every new question becomes a ticket, and the queue grows faster than the team.

In the sample report below, a fictional insurer's BI team got 600 requests in a quarter, with a median wait of six business days. Its loss ratio had four definitions that differed by up to 12 points.

Yes, once the foundations are in place. A chat tool needs one definition per metric, and it needs those definitions somewhere it can read them, which is what a semantic layer is for. Permissions have to live in the warehouse, where every query passes through them. Most BI tools enforce permissions in the reports, and a chat tool skips the reports.

Without those foundations, two people who ask the same question get different answers, and a chat tool can show data people aren't allowed to see. The diagnostic tells you which foundations you have and which you need to build.

• Options to speed up analytics, from self-service reporting to chat-based BI, with the cost and risk of each
• Better data access and data security: roles that match jobs, masking for personal data, and logs of who queried what
• A process for defining metrics, with an owner for each one and a clear way to agree on changes
• KPI definitions everyone agrees on, built into a semantic layer as tested code

If the data feeding your reports is unreliable, the work starts with data engineering.

Monday is the kickoff and read-only access to your warehouse, BI tool and request log. On Tuesday I interview the teams who ask the questions and the BI team who answers them. On Wednesday I trace your most-used metrics through every report, view and spreadsheet that calculates them. On Thursday I map who can see which data and test how a chat tool answers real questions against your current setup. On Friday I present the findings, the options and a fixed quote for the first phase.

The sample report for Larkspur Mutual Insurance, a fictional insurer, shows the full format (PDF, 13 pages).

I work in the tools you already have, most often Snowflake, dbt, Tableau and Power BI. Selecting and buying a new tool isn't part of the work. If the diagnostic finds that one of your current tools can't do the job, the report says so and explains why.

Client work stays confidential, so here's one of my own. For StockDataAnalytics I analyzed 39,037 symmetrical triangle trades to test the rules traders learn from textbooks. The well-known apex timing rule showed no predictive power, while a few simple filters, such as the size of the first day's drop, separated winners from losers. The article also marks where the sample gets too thin to trust: Symmetrical Triangle Pattern Win Rate: 39,000-Trade Analysis.

The one-week diagnostic costs a fixed $8,000. The work after it is quoted at a fixed price once the diagnostic shows what it involves.