Understanding the data completely— a skill becoming more and more important with AI— along with storytelling and simplicity differentiated the winners of this year’s Statistics Datathon.
The 10th anniversary of the Illinois Statistics Datathon is in the books. Nearly 600 students across more than 200 teams hailing from the Department of Statistics, the Siebel School of Computing and Data Science, the iSchool, Molecular and Cellular Biology, Psychology, and many others gathered at the Campus Instructional Facility on the weekend of March 27 to wrangle Synchrony’s industry call center data in just 36 hours.
“It’s heartwarming to connect with so many students ready to take on a real-world challenge, students from incredibly diverse academic backgrounds, leveraging their skills to not just compete but learn,” said J.D. Graham, Outreach and Engagement Specialist for the Office of Data Science Research, after attending the event.
The annual competition, sponsored by Synchrony, DataBricks, and Amazon Web Services, invited students from all majors and skill levels to predict call center activity across August 2025, broken down into 30-minute windows. Students had daily totals and interval-level data from April through June 2025, but not how those calls were distributed across each day in August.
More specifically, they were asked to predict three things for each slim window:
- Call Volume (CV) — how many inbound calls would arrive in that slot
- Customer Care Time (CCT) — the average time an agent would spend on each call, in seconds
- Abandon Rate (ABD) — the fraction of callers who hang up before an agent answers
Three metrics, 1,488 rows, asymmetric scoring that punished underprediction, and an uncommon request these days: don’t use AI.
Throughout the day and a half, thousands of submissions came through, but only one team could come out on top.
“For standing out for their creativity, technical skill, and ability to tackle a real-world forecasting challenge,” per the Datathon page, Team 225 was crowned first. For more information on the second and third place teams, please visit the Datathon page.
The winning formula
Eshan Velidandla, a sophomore in Information Science + Data Science, joined first year Master’s student in Computer Science Rishab Shah, senior in Statistics Charlotte Sampson, and sophomore in Information Science + DS Raphael Chen to become victorious by doing less than everyone else: No machine learning, complex models, or AI tools, and they beat teams with 10x more submissions.
Their core formula was Interval CV = Daily CV × Shape × Bias.
Their insight: it wasn’t a forecasting problem, it was a shape transfer problem— instead of trying to figure out the shape of a day’s call intensity, they transferred over the spring data by summing up all calls in a given time slot across all days and dividing by total calls or ratio of sums. This meant busier, more representative days automatically carried more weight than, say, a low-volume Sunday. They also excluded extreme outliers like holidays and post-holiday “return days.
When they tried reshaping the abandon rate the same way they did call volume, the math broke down at the edges of the day. They realized that caller patience is driven by the staffing-to-volume ratio, which is stable across seasons, so they just applied spring’s abandon rate to August.
Because the competition penalized underprediction more than overprediction, the
mathematically optimal prediction wasn’t their best estimate; it was something slightly higher.
After enough submissions, they used their leaderboard data of known volume error (EV), care time (EC), and abandon rate (EB), and put it through an OLS regression. This revealed the scoring weights — roughly how much each metric contributed to the composite score, which told them where to focus. Knowing
the penalty structure, they ran a sweep of multipliers across several submissions until the
composite score bottomed out at 1.044x. They reverse-engineered the competition. Every call volume prediction was then submitted 4.4% higher than their model said.
With an EV of 34.134, EC of 14.04%, and EB of 1.26%, their composite score ranked 6, but they beat first place on two of three metrics. Where they really differentiated themselves was when framing the numbers.
“What would a customer in this instance do? What is Synchrony’s incentive in this issue? We were able to say, we chose this algorithm because of this reasoning.”
-Velidandla
Data Storytelling
The team said it came down to leaning into a consulting mindset, capitalizing on their soft skills, storytelling, and a consumer-focused presentation. Instead of just defending their algorithms, they asked deeper questions.
“What would a customer in this instance do? How would a customer approach this problem? What is Synchrony’s incentive in this issue?” Velidandla said.”We were able to say, we chose this algorithm because of this reasoning.”
During the presentation, they were able to showcase this, and onlookers like Graham took notice.
“What struck me was how much time they spent on the why behind every decision, not just the how. Using their intuition and lived experience to tackle a complex problem,” he said. “To me, that’s exactly the kind of thinking we want students walking away with.”
Shah, the most experienced professionally, saw this opportunity as a reflection of real‑world data work.
“I really wanted to hone in on the data storytelling. That’s just the most important skill by far, I see it so much in work,” he said, adding that the judges noticed it too. “They gave us the most compliments on that by far.”
“We were pretty touchy about the AI subject…The best thing is to really get your head wrapped around the data. AI can never replace doing those exploratory steps.”
– Chen
AI-less
When asked about AI, Velidandla put it simply: “You can’t be good at using AI in your workflow if you’re not good at the workflow itself.”
The team admitted that the no-AI constraint was ultimately a gift, allowing them to slow down and think about their approach. They built visualizations, talked about what those visualizations actually meant, and that process revealed they didn’t need AIML at all.
“We were pretty touchy about the AI subject. I think in general, for me, when it comes to data science, especially like the simpler machine learning stuff, I’m very hesitant to use AI,” Chen said. “The best thing is to really get your head wrapped around the data. AI can never replace doing those exploratory steps.”
Velidandla added that it’s a common pitfall for teams in this kind of competition.
“Most of the time, people are just very trigger-happy, so to speak. The first thing they do is compile the files and throw them into a chatbot and say, ‘Let’s build something, let’s do something, find the best solution here.”
You’re playing the role of the product manager, as Chen puts it, where you understand the entire pipeline, then apply AI to help with individual implementations. But even then, they didn’t use it, and the 2nd place team 015 proved why not: part of their presentation focused on how using a pure AI prediction system actually performed the worst out of any type of curve extraction that you could use.
Chen reinforced the core argument for simplicity: “Since it didn’t use ML, that’s what made it so powerful, because it generalized well. A more complex machine learning prediction model would have overfit or fit to the noise of the data and would not have generalized very well to newer data.”
Their Story
Team 225 came into the weekend looking to get in one last hoorah before saying goodbye to two of their graduating members and maybe having some fun along the way.
“We wanted a chance to work together in a collaborative, group-based environment at least once before Rishab and Charlotte leave,” Velidandla said. “We came with the mindset of, hey, let’s go and have some fun. Let’s go design something cool.”
This doesn’t mean they didn’t have a competitive edge, for some of them, they were out to establish themselves as capable data science leaders on campus.
“We wanted to prove our skills and be like, this is what we can do. You can kind of trust us to give advice on data science-related matters,” Chen said. “Going in, I was like, yeah, we kind of have to win this.”
In the beginning, the team wasn’t fully in it. It was a distracting weekend with Illini Men’s Basketball far along in the tournament, and Team 225 was positioned around 20th before a single submission launched them to 8th place, and they decided, “this is a project we want to focus in on.”
“I do think we still have a long way to go,” Velidandla added. He notes the experience of presenting to Sid Chakravarty from Synchrony and receiving his feedback as a gift that added to their data storytelling journey. “There’s always another step forward”.
Though it meant most to Veridandla to collaborate with this group before they took that step forward.
“I had so much fun working with these guys.”
Learn More and Get Involved
Read more about the Datathon here.
Student organization leaders and faculty advisors: Visit our student portal and contact ODSR if you would like to have your group’s data science or AI-related activities featured here. ODSR is a campuswide convening organization that facilitates collaborations, resource sharing, and public engagement focused on data science research activities at the University of Illinois.