Research, modeling and operating standards
Take a problem. Go to the primary literature, not the summary of it. Build the model from the best data available, public where that lets you check it yourself. Validate it against something independent, and say where it fails. The sector changes. The standard of proof does not.
The group
Analytics answers where to put it and whether the economics hold. Advisory answers how it runs once it is there. Learning builds instruction from what the research actually supports. Different problems, different sectors, sold separately, and built on the same rule: name the source or do not make the claim.
Demand, unit economics, competitive position and where the money is actually going. Whether an expansion pays, what a site has to do to clear its cost, and which costs are structural rather than habit. Current product: Meridian.
Organizational transformation on current best practice: culture, workforce and management development, written standards with a KPI against each one, and the change work that gets them adopted. Current product: the Pillars of Excellence playbook.
Purpose-built learning tools for a specific population, a specific skill and a specific gap, built from the practice guides and the intervention literature rather than from what sells. Current product: Able Minds Lab.
We tested the published site-ranking models our own industry sells. They do not predict outcomes, p = 0.86. We publish that rather than sell one.
The method
01
The technical report, not the summary of it. No name, number, date or rate appears without the file it came from. Public data wherever possible, so a client can check the work.
02
A separate method on separate data, not the data that built the model. Our break-even estimate lands within 1.6% of an independently published actuarial study.
03
Every deliverable marks what was verified, what is professional judgment, and what only the client can confirm from their own book.
04
Out loud, in the deliverable. A finding that kills an idea is worth more than one that flatters it, and a model reported only on its wins is not a model.