Natalia Connolly

Natalia Connolly is an experimental physicist by training and a director at AlixPartners. She has built data science teams and systems for people making decisions as different as when to intervene in a growing crop and how much a buyer can credibly ask a vendor to cut from a contract. In the latter case, she and Kevin Madura designed their analysis around what the buyer would need at the negotiating table: the contract page or invoice line behind each figure.

From galaxy measurements to medical records

Natalia earned a PhD in experimental high-energy physics and credits Natalie Roe of Lawrence Berkeley National Laboratory as an early mentor. As an assistant professor of physics at Hamilton College, Natalia developed a statistical method with Brian Connolly to analyze galaxies. In 2012, they supervised Hamilton student Spencer Olsson as he tested whether the method could also distinguish patterns of gene expression in two species of fish parasites.

At Cincinnati Children’s Hospital Medical Center, Natalia developed a research program applying machine learning to healthcare. She was first author of a 2016 study linking birth certificates to children’s medical records. With the records linked, the researchers could examine whether maternal metabolic conditions during pregnancy were associated with developmental diagnoses the children received years later.

Data science on the growing floor

As director of data science at 84.51°, Natalia used natural language processing and other machine-learning methods to personalize digital experiences for retail loyalty customers. At EarnIn, she managed a team that modeled which customers were likely to leave and used computer vision to process financial documents.

At the vertical-farming company Infinite Acres, Natalia built the data science and analytics team as vice president of data science. The team forecast harvests and customer demand and monitored crops as they grew. Crop specialists helped set the conditions for each growing zone. If a sensor detected a temperature outside the range for that crop, or a camera picked up a visible anomaly, the system alerted growers so they could intervene. Natalia also saw a use for those measurements beyond the immediate alert. By keeping data from the start of farm operations, the team could compare a current crop with earlier harvests.

Job titles and contract terms in context

• Classifying jobs by reporting relationships. At AlixPartners, Natalia and her co-authors found that matching job titles by wording put project managers and product managers too close together. Even an exact match could mislead: an analyst reporting to the head of finance may do different work from one reporting to the head of marketing. Their job classification system uses keywords and similarity in meaning to narrow hundreds of possible roles to about ten. A language model then considers the employee’s managers, peers and direct reports to rank those roles for a human reviewer.

• Tracing a contract total to the amendment that changed it. In Natalia and Kevin’s reconstruction of a post-merger procurement analysis, Claude Code calculated annual spending correctly but missed roughly $800,000 in contract value because an amendment had lost its link to the original agreement. Even finding every amendment leaves a question of interpretation: an amendment can restate the entire contract value, in which case adding the old and new figures counts the same obligation twice. Natalia and Kevin spent about two weeks building a pipeline that linked agreements to their amendments and extracted rates and terms into tables. They could then check the model’s answers with code. Once the pipeline was ready, it processed more than 40,000 contract documents overnight.