Breaking Down Machine Learning Silos to Maximise Value

Building walls to tear them down 

Traditional FSI organisations have been institutionally siloed mainly due to regulatory, legal and operating pressures. Business units and functions try to unify across customer journey transformations to stop the short-term fix of stitching processes and data flows together to bridge customer touchpoints, often unsustainable, carrying higher cost implications further down the road. Technology operating models and entity structures historically have inhibited customer experience, business unit consolidation is enabling a fuller spectrum of products and services to be served across the customer lifecycle. For example, linking retail, private banking, insurance and wealth to increase customer wallet share and retention. Commercial banks and capital markets are also tightening market positioning to enable businesses through a wider product portfolio with tailored risk or climate-adjusted pricing. These shifts aim to maximise upsell and cross sell strategies through a single customer view. 

This is changing as customers demand a bank any way and anywhere experience leveraging universal expectations driven by big tech (Amazon, Google, Facebook), FinTechs and challengers. Digital business and operating models are redefining how banks service their customers with diversified strategies including partnerships and banking-as-a-service modules to protect market share and grow new business, for example through standalone digital banks. Risk processes integrating with external factors such as climate are causing further disruption whilst banks try to adapt legacy risk frameworks impacted by modern pressures to protect the balance sheet. For example, green mortgages, climate stress tests, or climate-adjusted pricing for commercial bank portfolios aligned to physical and transition risk exposures. Delivering a superior customer experience in a time where consumer change is volatile and cyber-crime is catalysed by the eruption of pandemic-driven e-com payments, integration of advanced, real-time risk segmentation and richness of insights from stress tests into day-to-day operations is on the horizon for most banks, but no easy task. Due to the growth of datasets and simulations involved, to add material value requires advanced analytics, automation, flexible cloud-based technologies and talent to enable collaboration in line with governance as well as the agility to change with the evolving business, regulatory, consumer and technical landscape. Machine Learning will be critical to the success of modern AI banks; however, it is mostly active in pockets and silos hindering full value creation. Agile or fluid operating models centred around MLOps will be needed in order to truly scale ML and data science value across financial institutions. Talent is critical, building multi-disciplined teams to formulate networks and cross-functional scrums to deliver against critical group-wide ML initiatives i.e customer retention or acquisition strategies translated to common strategic goals involving multiple functions aligned to key customer journeys. Getting ahead of the curve rather than waiting for scale or governance to become a problem is pretty obvious, but a bit of a chicken and egg story with other factors such as ethical AI and explainability adding layers of complexity. Leveraging MLOp structures strengthens key business drivers, for example; customer experience through hyper-personalisation, operating margins through robotic process automation or computer vision, sales/agent productivity through smart routing and human centred AI, ultimately, increasing sustainable returns on capital.  

Goals vs realities 

The reality is that most traditional business systems and units lack digital maturity, skills and transparency to run complex ML deployments like multi-arm bandits or ensemble ML model chains to automate and impact customer journeys, let alone across complex ML techniques, cloud ecosystems and business processes to yield higher customer value, forecast accuracy or cost optimization. Model catalogues, versioning, role-based access and audit trails enable stakeholders to collaborate with agility across the MLOps lifecycle, enabling the right skills to access the right information at the right time in line with change controls and governance procedures. Often creating silos and barriers to what should be simple ‘business sense’ decisions, accountability and sponsorship needs to be driven from executives or key business stakeholders to ensure sustainable value. 

Even when the organisation seemingly is ‘one’ to the consumer, large gaps remain operationally, spurring unnecessary work required in the mid/back office to maintain SLA’s. In the context of machine learning, this sees data scientists often working in isolation from devops and engineering teams with poor workflows and handoffs, miscommunication, portability and errors occur throughout the deployment process causing delays – one of the reasons only 1 in 10 machine learning models make it into production. Seldon enables the data scientist or machine learning engineer to 1-click deploy complex inference graphs without having deep expertise in Yammal, K8s or container infrastructures. The data scientist or end user also has the ability to apply real-time, advanced monitoring for drift or outlier detection across any cloud or training framework creating deeper transparency and collaboration with DevOps and wider business teams. In-turn this begins the journey of the democratisation of AI and ML across the business through a single deployment and monitoring platform. 

Sometimes these gaps are by design due to regulatory purposes that are required to execute strategies at the business unit level without compromising ethical or capital risk boundaries. However, consumers wanting wealth or trading products or businesses needing commercial banking products expect a seamless experience. Moving forwards, the FSI Industry will need to better apply data driven insights to capitalize on cross/up-selling to extend relationships and tailor risk adjusted pricing and discounts increasing customer lifetime value and wallet share. 

Moving in the right direction 

Most banks are aware of these challenges, actively implementing change management protocols to bring processes, ML technology and people together under Agile operating models or hybrid versions as customer journeys evolve. MLOps is a key strategic enabler allowing greater collaboration to deploy and scale models feeding critical business applications, the use of advanced monitoring and enterprise-wide optimisation factoring in drift and outlier detection enables proactive risk mitigation, creates better feature insights and overall transparency across the model deployment lifecycle. 

Into the mix comes Seldon. The progression of machine learning proliferating FSI comes with vast benefits but also as we’ve seen before from Cloud can become unwieldy and counter-productive if not flexible to change and controlled. Our cohesive platform offers guardrails that allow FSI players to integrate to any cloud, ML framework or toolkits to scale, increasing speed and productivity of machine learning deployment with advanced monitoring, explainability and governance.  

Consider a modern customer journey to gain a real-time loan – customer acquisition (hyper-personalised offer>customer retargeting>propensity to buy score>channel mapping) and credit decision value chains (credit qual>limit asses>pricing optimisation>fraud prevention). Risk, marketing, channel, infrastructure, payments, document processing and KYC are all areas ML model chains require to deliver value; these sit across different disciplines, teams, functions, systems and governance models, hence why traditional banks have struggled to compete with the agility of challenger banks agile structures. Furthermore, leveraging feedback loops can optimise real-time traffic based on customer behaviour through re-segmentation modelling integrated with digital marketing models enhanced further by cohort and micro segmentation, leveraging unstructured data from social media, geo-location or browsers.  

However, silos between platforms, data pipelines and teams are limiting progress. Cultural and strategic differences between data scientists, DevOps and MLOps are creating hurdles and bottlenecks that limit efficiency and productivity. Ultimately, bank operating models are not moving fast enough to keep pace with the desired rate of change. 

Again, Seldon can help. By bridging the gaps between silos, our tools enable operations to collaborate around critical strategies evolving in line with market and customer change – more quickly. 

A platform for the future 

With the ability to deploy ML models, monitor and optimise at scale with confidence, Seldon will help your business maximise returns on investments across data science, machine learning and enhance value downstream to the business applications they serve. Creating agility across MLOps to deploy, learn, optimise against key business priorities will enable better resource allocation with clients experiencing up to a 92% increase in productivity, increasing ML returns to scale. 

Leveraging machine learning at scale to better address the changing priorities of customers, market risk and underserved categories will maximise revenue, retention and growth whilst optimising cost. 

To learn more about how Seldon can help your business do more with machine learning, please get in touch. 

Contents