The problem
The national unemployment rate and GDP describe the country as a whole. They say little about why one regional economy adds jobs while the next one stalls. The Institute for Strategy and Competitiveness (ISC) at Harvard Business School studies those differences through clusters: regional concentrations of related industries. The data behind that research was spread across roughly 12 open sources, including the Census Bureau’s County Business Patterns and StatsAmerica, and added up to more than 50 million records. Policymakers, economic development teams, academics and business owners needed to find the numbers for their own state, metro area or county, then compare them, without stitching sources together themselves.
Context
ISC, the Economic Development Administration (EDA) of the U.S. Department of Commerce and Phase2 set out to build Clustermapping.us, a public, interactive map of America's economic geography. The national initiative was funded by an EDA federal grant, and the cluster categories came from an algorithm developed by researchers from Harvard, MIT and Temple, and published in the Journal of Economic Geography. The research was ISC's, and its value depended on the data staying consistent and rigorous at every level, from states down to counties. Success meant a site where any user could compare regions, draw their own economic areas, follow a single industry cluster across the country, and find the related research and reports in the same place.
Our approach
Work began in May 2013. The beta launched in June 2014, with the full site following later that year. What started as a two-phase build grew into a three-year relationship, with ISC extending the engagement nine times through launch, post-launch support and a managed services agreement covering 2015 and 2016.
Twelve open sources, one data set
The work started with the data. The site imported raw data from roughly 12 open sources, and brought it into one data set of over 50 million records, combining cluster research with 52+ economic indicators for every state, metropolitan area and county. Instead of forcing sources into a fixed database schema, the team stored each region and each cluster as its own record, with every measure carried as a property on that record. A new source added new properties, not a schema change or a code release. ISC staff uploaded new data sets through the site’s admin console, and a data service turned them into records without a developer in the loop. Pre-built summary records, calculated with methods ISC approved, kept queries returning in milliseconds.
A semantic data model for clusters and regions
Users mixed and matched measures of economic development and combined counties into custom economic zones. That worked because every record encapsulated what it described and where it sat: the industry, the cluster that industry belonged to, the geography, and the state around it and the counties inside it. Region shapes came from Census geography, and economic areas were assembled by combining counties, so a user-defined zone was resolved the same way an official one did. The site’s content tags stayed in sync with the regions in the data, so research and numbers described places the same way.
Dashboards that build data on request
The team built a set of custom dashboards that generate new sets of data on request rather than serving fixed reports. A user can define a custom economic zone, pick the measures that matter to them, and open cluster dashboards by industry. Charts on the same page drew on the same data, so selecting a region on the map updated everything around it. A search for a state returned the state, every region inside it and the related research, and a public API returned the same data for anyone running their own analysis.Design and user experience were a central focus from the start, so the maps stay clear even with this much data behind them. The site also gathers academic research, tools, technical documents and reports from other organizations next to the data.
Outcome
Clustermapping.us turned roughly 12 open sources and over 50 million records into one public tool for regional economic data. Policymakers could then see how their region performed and what drove wages and job growth there, then shape policy to local needs. Economic development professionals and academics were able to test which factors make a region strong or weak. Business owners got regional data to inform strategy. Regional organizations put it to work. ISC demonstrated the site to regional development leaders at NADO’s 2014 conference, Skagit County’s economic development association published its county dashboard as a resource, and Hanover County, Virginia’s 2023 industry cluster analysis drew on its data nine years after launch. Under the managed services agreement, the team added dashboards and visualizations that widened the site’s users and use case. One consistent model behind scattered sources is what made this work, and it’s what makes data ready for AI today.