Tide Drives Innovation, Scale, and Savings With AI Observability

London, UK
Models in production
Use cases
  • Fraud detection - transaction monitoring
  • Credit approval - probability of default 
  • Existing customer risk assessment
  • New customer risk assessment
Tech stack
  • Training: Databricks notebooks on AWS
  • Packaging: MLflow - GitHub workflow
  • Registry: MLflow 
  • Data testing: Databricks 
  • Feature store: Tecton
  • AI Observability: Fiddler
Download PDF

Fiddler AI Observability helps Tide scale their machine learning (ML) solutions to support the company’s growth, gain a deeper understanding of model outcomes, and align data science and business teams by tying together model performance with business KPIs. 


Since Fiddler, Tide has been able to: 

  • Save one day of troubleshooting for every performance issue found by Fiddler
  • Prioritize high-value ML projects to support the company’s growth
  • Increase understanding of model outcomes for better decision making

Reactive model monitoring and lacking standardization

Tide, the UK’s leading business financial platform, needed to scale ML solutions, including automated bookkeeping and integrated invoicing. Tide wanted to provide accounting solutions to small businesses who did not have a finance function. 

Maintaining production ML models, however, hindered the data science team from developing new solutions to unlock business opportunities. Data scientists spent days writing and managing scripts to troubleshoot model performance issues instead of focusing on innovation. What’s more, each data science team implemented their own monitoring tool, resulting in disparate views and interpretations of model performance. This separation of work created silos for decision making amongst data science, machine learning, and business teams. 

“Model monitoring was extremely reactive to the point that we wouldn’t know if we were sitting on a problem. Some of the pain points are mean time to detect concept or data drift, root cause analysis and explainability,” said Suryanarayana Ambatipudi, Head of Data Science at Tide. “We needed standard tools and processes in place that everyone could stand behind to scale our ML efforts and support the company’s growth strategy.” That’s when Ambatipudi knew he needed an enterprise-grade model monitoring tool to empower his data science team to scale the company’s ML initiatives. 

Driving innovation and productivity with AI Observability

Ambatipudi sought to build out an MLOps framework with tools to support his team’s ML efforts. Early on Tide considered building a monitoring tool in-house. But Ambatipudi knew that procuring a tool would not only be cost-effective in the long run but also adhere to ML industry standards. “Building requires expertise and time that my team does not have. My team is hired to give value to the business, not to build a tool,” said Ambatipudi. “Time and expertise aside, there are also high infrastructure and maintenance costs involved in such an exercise.”

After evaluating various monitoring tools, Ambatipudi selected Fiddler as their AI Observability platform of choice. 

“Fiddler’s robust AI Observability platform is a powerful, yet flexible addition to our MLOps framework. Since using Fiddler, we have saved one day worth of troubleshooting for every issue that Fiddler detects. Fiddler has taken full care of our monitoring needs in the background, while the data science team is focusing on what matters in the foreground.”

— Suryanarayana Ambatipudi, Head of Data Science, Tide

Cost savings from automated performance tracking

Tide trusts Fiddler to handle monitoring operations for their business-critical models that move the business forward, from detecting fraudulent transactions to evaluating loan approvals to assessing customer risks. The data science team is now better equipped to immediately deep dive into issues that impact the business, and provide quick post-mortem analysis as needed.

Connecting model predictions to business decisions using explainable AI

The team is also embarking on using explainable AI to provide answers to Operations to help reduce the mean-time-to-resolve (MTTR) business metric. 

“Fiddler’s unified view, coupled with monitoring and explainability, will enable us to work closer with business teams…This will help change the conversation from being just model providers. We are confident in our decisions by aligning ourselves with the same ML metrics, and gaining insights on what the models are doing and why they are behaving that way.”

— Suryanarayana Ambatipudi, Head of Data Science, Tide

Scaling up ML initiatives in the future

With Fiddler in place, Ambatipudi is confident that his team can fully support the business by bringing models into production to solve ML use cases over time. “At Tide, our models are categorized as high or low risk based on the business impact of decisions that can go wrong as well as the long lead cycle in getting the ground truth. Right now we are focusing on monitoring high-risk models that drive the business forward in Fiddler. As our MLOps become more mature, sophisticated and automated, we can optimize business performance by taking the low-risk models to Fiddler,” said Ambatipudi

“Fiddler is an extended team for the ML organization at Tide, and also a trusted advisor in the space of model monitoring and explainability. We look forward to continuing our relationship with them,” said Ambatipudi.

With Fiddler, Tide enjoys:

  • Higher productivity without spending time monitoring and troubleshooting performance issues
  • Unified view of model performance metrics 
  • Improved collaboration between data science and business teams