Finding the next retail expansion opportunity in existing data

The question

Additech sells fuel additives through dispensers installed at retail stores. The product was available in only a handful of states, so the expansion question came down to two things: which markets next, and what would they be worth? For years Additech had collected terabytes of transaction data from the stores its pumps sat in, millions of transactions a day, with no data-warehousing solution to analyze it.

At the end of 2024, Fulton Ring partnered with Additech’s technology and leadership team and started by extracting sales data from one of Additech’s retail partners.

The build

The first attempt stored the data in TimescaleDB. As millions of transactions per day accumulated, query times stopped scaling. Fulton Ring moved to a data-lake architecture instead: transaction data loaded into Amazon S3 as Parquet, queried with Athena. Queries over terabytes of data dropped from several minutes to under 8 seconds, fast enough to drive live dashboards. The serverless setup also cut expected infrastructure costs by more than 92% compared with running databases on EC2.

Dagster orchestrated the ingestion pipeline. Running it on EC2 rather than a serverless, per-processing-billed option kept the cost of backfilling the historical data down.

What the team can do now

Amazon QuickSight dashboards put the data in front of Additech’s decision makers:

• Loyalty-account insights showing where and how often customers shop, to test whether recurring transaction volume correlates with Additech sales.
• Average transaction volume per site, with high-volume days flagged.
• An adjustable projected-revenue calculator that estimates expected sales at locations where Additech is not installed, and reports how reliable each estimate is. It gives the team a repeatable way to compare expansion candidates nationwide.

Public record

We published the full write-up, including the architecture trade-offs, in the Fulton Ring Medium publication on March 25, 2025.