about the record
firstmileproblem.com is written and maintained by Marcos Novo. It is an independent project — no sponsor, no vendor, nothing for sale. It exists to define one term clearly enough that people can use it, cite it and argue with it.
One term, defined in public.
Most AI projects that stall do not stall at the model. They stall earlier — where data is created, collected and prepared — and the people closest to that failure rarely have a name for it. The first-mile problem is that name. This site records what it means, where it came from, the evidence for it, and a way to check your own data against it.
The definition is meant to be reused. Quote it, link it, adapt it — the field definition carries ready-made citations in APA, MLA and BibTeX.
How the pages are checked.
Primary sources first
Every figure links to the publisher’s own page. Where none is public, the outlet that first reported it is named, and the page says so.
Label what kind of claim it is
Measured results, analyst forecasts and repeated estimates are marked differently. Vendor-sponsored surveys are flagged as vendor-sponsored.
Credit honestly
Ideas are credited to the people who said them, including uses that predate this site. Nobody here claims to have coined the term.
Stable definitions
Definitions and their anchors (for example #semantic-drift) do not move, so links and citations keep working. Changes are logged below.
No paid placement
No sponsored content, no affiliate links, no vendor rankings. Companies are cited when they published something worth citing.
AI-assisted, human-checked
Research and drafting use AI tools. Every claim and figure is checked against its source before publication, and the sources are linked on the page.
Corrections log.
Found an error, a misattribution or an earlier use of the term? Write to the contact listed in humans.txt. Corrections are made on the page and logged here with the date.
· Origin
Earlier uses of “first mile” in data work (2011, 2017) added. Anu Jain is now credited with the sharpest current wording of the idea, not with coining it.
· Definition
Two unsourced claims softened (“half a trillion dollars of fiber”, “most data never reaches model-ready state”). Sourced Gartner and S&P Global figures added.
· Glossary
Layer ordinals corrected — four “See” links had counted the seven layers from zero and from one at the same time.
· Field pages
Structured-data FAQ answers realigned word-for-word with the visible answers.