BaseModel AI boosts Customer Retention Accuracy x4 in Finance industry

In an increasingly competitive banking environment, few challenges are more critical than engaging with clients at the right time - especially when it comes to knowing who might leave, and when.

Across the US financial sector, leading institutions with advanced analytics capabilities have been actively exploring how data and intelligent automation can strengthen customer relationship management.

To accelerate this transformation, we partnered with one such institution on a Proof of Concept (PoC) focused on two high-impact areas:

  1. Deposit Retention/Engagement diminishment – identifying customers at risk of closing their accounts or diminishing relationship with the bank
  2. Relationship Deepening – identifying customers likely to open additional products in the near term.

The goal: empower the bank’s team with AI-enhanced insights, improve precision of targeting, and reduce the manual effort involved in customer identification - all while keeping the solution fully aligned with the bank’s existing infrastructure and compliance standards.

Use Case 1: Predicting Deposit Churn

The first use case focused on early detection of diminishment behaviors - for example, a significant drop in deposit balances or transactional activity across various product lines. These signals often precede client completely leaving the bank, but can be difficult to catch early using static rules.

The bank’s existing approach, based on well-crafted internal logic and heuristics, provided a strong foundation — and BaseModel was introduced to test how much additional predictive precision could be gained through machine learning.

Our Collaborative Approach

Leveraging the bank’s data (monthly balances, engagement metrics, transactions, product ownership), BaseModel was deployed directly on the client’s infrastructure, with full access to relevant data. Predictions were generated, validated using shared thresholds, and ultimately passed into the bank's CRM platform for action by the bank’s retention team.

Two evaluation strategies were used:

  • Matching the original approach: predicting a fixed-sized prioritizelist of customers most likely to disengage
  • Using BaseModel’s optimized internal threshold to compare lift and coverage

The Results - and What They Enabled

BaseModel delivered a precision improvement of over 4x compared to the bank’s original approach — allowing the team to focus more effectively on customers genuinely at risk of disengagement.

This uplift in targeting accuracy translated to real operational benefits:

  • Less time spent on false positives
  • Higher confidence in outreach
  • Improved resource allocation across retention workflows

Lift analysis showed:

  • The top 10% of ranked customers captured ~40% of all actual diminishers
  • The top 30% covered ~70%

What’s more, a follow-up conducted months after the initial scoring confirmed that only the customers identified by BaseModel continued to exhibit diminishing behavior — reinforcing the long-term predictive value of the model and its potential as an early-warning system.

Use Case 2: Propensity to Open – With Real-Time Impact

The second use case targeted customers who were likely to open another account within the next 30–60 days.

BaseModel was used to score the bank’s existing customers on their propensity to deepen their relationship — through products like term deposits, retirement products, checking accounts, and more. Once scored, the bank’s team reached out to those with the highest predicted intent.

Stories from the Field

The results weren’t just visible in metrics - they played out in real money and real timing:

  • In one case, a client with a single product relationship had planned to move a large sum of money to a competing bank. The competitor failed to respond in time - but the bank, guided by BaseModel’s recommendation, called at exactly the right moment. The client changed course and ultimately deposited a significant six-figure deposit instead.
  • In another case, a bank advisor reviewing the list of high-propensity clients noticed a familiar name - someone they were just about to contact. Moments later, that same client walked into a branch on their own to open a new account. As the advisor remarked:
“It’s amazing - we knew he was going to open the account before he knew it himself.”

These were not scripted scenarios. They were authentic, unscripted confirmations that BaseModel could detect latent customer intent - even before it became visible through behavior.

What Made It Work

Several factors contributed to the success of this PoC:

  • Embedded deployment: BaseModel was deployed on the client’s own infrastructure, maintaining compliance and data privacy.
  • End-to-end pipeline: Scores were computed in batch, integrated into bank's CRM platform, and acted on directly by the bank’s teams.
  • No additional AI tuning required: BaseModel integrated seamlessly into the bank’s environment — allowing their analytics and CRM teams to focus on insights and action, not technical maintenance.
  • Transparent validation: Even without probability scores from legacy models, we aligned thresholds and validated precision using shared criteria.

And perhaps most importantly:

The collaboration between the bank’s team and BaseModel was exceptional - marked by mutual trust, shared ownership, and a clear focus on delivering value. From day one, the joint effort was fast-moving, constructive, and deeply aligned.

This close partnership was key to the PoC’s success - from data onboarding to validation, and from deployment to real-world impact.

A special thanks to the bank’s data, analytics, and relationship teams - their domain knowledge, agility, and customer focus were instrumental in achieving results quickly and meaningfully. This was not just a PoC - it was a real collaboration, with real outcomes.

What’s Next?

The bank is now exploring production rollout of both models, and investigating potential new use cases as well as expanding and executing on the existing ones.

The biggest surprise?

Just how well the engagement diminishment model performed out-of-the-box - and how quickly it was up and running.

Why BaseModel?

When precision matters - and it always does in retention and acquisition - BaseModel delivers. Whether you're running targeted campaigns, managing customer relationships, or fighting churn, it helps you act on the right customers at the right time.