Rapidly delivering explainable product recommendations for Credit Unions with PromptQL

Executive Summary

Credit unions want to recommend new products to their loyal customers - not just to drive revenue, but to help members access financial products that fit their needs.

But because the relationship between credit unions and their members is built on trust, they can’t make those recommendations blindly. When a predictive model suggests a product - like an auto loan or a home equity line of credit (HELOC) - frontline staff needs to understand why. This was the problem a Fintech business was trying to solve for their Credit Union customers. Unfortunately, previous approaches they tried produced vague, hard-to-explain, or untrustworthy recommendations.

PromptQL is able to provide human-readable, accurate, and context-aware reasoning that creates trust in otherwise opaque AI recommendation. Even better, with PromptQL in place, this Fintech company can now create and deploy new recommendation systems for their Credit Unions customers in just minutes - instead of weeks or months - without reworking the underlying data!

"We don’t need hallucinated guesses. We need real answers. PromptQL and Hasura helps deliver just that! "

The Challenge

The customer’s goal is to build trust and personalization in credit union interactions by enriching product recommendations with clear reasoning. The team had tried custom OpenAI integrations, other AI vendors, and even raw LLMs like ChatGPT - but all fell short. Custom builds were time-consuming and brittle; in the words of the customer:

"It took me four or five days to write the microservice, and then a few more to push it into production."

Third-party tools lacked transparency and explainability. And generic LLMs introduced too much risk of hallucination. Manual approaches to data mapping, where they manually mapped 100s of fields across a dozen data sources, and model deployment took weeks or months.

The team needed something faster, smarter, and more reliable.

How PromptQL Solved It

Hasura enabled the customer to rapidly deploy AI-powered, explainable product recommendations for their credit union customers – without reworking existing data infrastructure. They were able to go from raw, fragmented data to explainable prompt-driven recommendation with minimal lift – no need for data migration, restructuring, or prep work to get the data “AI ready”.

Key Benefits:

• High accuracy, with near-zero setup
  A strong semantic layer is essential for accurate AI – but traditionally time-consuming to build and maintain. PromptQL’s agentic semantic approach automatically created a unified semantic graph across their multi-domain data in Snowflake, MongoDB, and Salesforce – letting them go from data to reliable recommendations with minimal effort.
• Explainability by design
  Every PromptQL recommendation is backed by clear, plain-language reasoning and confidence scores, helping frontline teller teams communicate the recommendations with confidence – building trust with users.
• Self-debugging AI
  PromptQL detects and corrects logic flaws in real time. For example, it once flagged a recommendation for a member with no transaction history, adjusted its logic, and explained the fix – without manual intervention.
• Frictionless iteration
  New use cases, rules, and data sources can be onboarded in minutes by updating metadata. No pipelines or engineering delays. One team integrated the NCUA 5300 call report in under 20 minutes – no duplication or custom setup required.

Example:

Query: “Pick a member over 25 (in age) and tell me about their next best product.
Prompt QL response: “Personal loan with 43% confidence, based on card transaction patterns.”

PromptQL automatically broke the query down into steps, performed multi-step reasoning and retrieved contextually accurate answers.

Highlight: Upon discovering zero card transactions in that member’s data, PromptQL flagged the inconsistency and self-corrected, showing the system could identify and explain when recommendations didn’t make sense. This ability to detect misalignment between model output and real data was previously impossible without custom code and deep debugging.

This is a game changer - both for the Hasura customer who’s trying to experiment faster and roll out new features and capabilities to credit unions, and for credit union tellers that are trying to confidently cross-sell and improve member management.

"We can develop new recommendation system in minutes - instead of weeks."

Faster Time to Market and Experimentation
With low setup and maintenance effort, PromptQL will enable the customer to go to market in record time and launch new recommendation much faster.

Improved Teller Experience
Credit union staff can cross-sell more confidently, because of the clear reasoning behind each recommendation, helping improve the member experience and cross-sell revenue.

Bonus Benefit: Rapid Credit Union Onboarding
Onboarding a new credit union used to take weeks of manual taxonomy mapping between the credit union's data schema and the customer's platform. With PromptQL, that mapping process now takes just days – saving hundreds of hours and accelerating value realization.

PromptQL’s ease of use, explainability, and ability to generate reliable results in the non-deterministic world of AI is helping the customer actively explore additional use-cases such as:

• Targeted marketing based on AI-generated confidence scores
• Easily identifying underperforming credit unions, and
• Internal automation for source-to-target mapping and anomaly detection