The first core banking software (CBS) appeared around the 1980’s in the west and 90’s in India, and then rapidly took off from there. CBS-era was a significant departure from the times of handwritten ledgers and accounting entries made manually, double entry style. Today many of those are precious artefacts of the banks’ museum. The benefits of core banking have brought in significant efficiencies along with system integrity, audit-ability and greater systems-driven control.
The CBS systems were essentially designed for branch based banking and operations. However the banking paradigm has shifted dramatically over this horizon - banking product’s have evolved , delivery modes have shifted, and servicing has seen an orthogonal shift over the last 30 years. And what has emerged on the horizon is a new set of customer experiences (and, more importantly, expectations) that the existing CBS systems can just not match up to. There has been an unnatural level of force-fit on these systems since them. After 30 years of being patched/re-patched/extended, and re-architected, the core banking systems are now being pulled from one clinic to another.
The key players - Oracle’s Flexcube, Temenos, FIS, Infosys Finacle dominated the landscape… and, surprisingly, continue to do so! With the notable exception of Thought Machine (founded by google engineer Paul Taylor), it’s not difficult to see why there has not been a Google like re-invention for the CBS packages. The answer is simple - It’s too painful for their customers to migrate.
The core banking software was never designed for always on + high throughput + low latency + digital (faceless) transactions. Over the years, some of the short-comings have been addressed with new versions of the systems, but others remain. As a result of the loading-up, the pressure on CBS systems is now evident, where during the reporting or EoD or heavy load seasons (like month-end) the systems see significant downtime. And this is even before we look at what the deluge of AI data will do!
Simply put, we are just not prepared for the era of AI.
- First, technically, one constraint is the DB that the CBS systems ride on (which is predominantly structured databases - sql). There is little chance that an sql db can be repurposed for a vector db - which is essential to running embeddings and similarity searches. Or for processing unstructured data - akin to drinking from the fire hose - from digital channels, satellite systems, and partner platforms.
Conceptually, the db needs to be re-architected for the following:
- Second constraint is the customizations that have been performed since. With each regulatory/govt/product-driven change, there are customizations that are required to be performed. These customizations are often so entrenched, (and without adequate documentation), that tinkering with them could have unintended consequences and send ripples across the downstream systems. Think about the rules, sub-rules, cross-product linkages, etc, etc. For instance, theoretically there could 500,000+ permutations on the account types (account category: 6 types mode of operation: 10 types service level: 6 levels KYC Level: 4 types digital access: 5 levels Customer Category: 10 types Special Features: 8 combinations * Documentation Type: 5 types). Though only about 50-100 are practically implemented.
Imagine the possibilities if we could indeed implement, and market, all the necessary possibilities - which would suit the customer needs the closest.
- Third constraint is that the “hollow-out-the-core” strategy of the last 10 years, has given rise to satellite systems, which have just gone manifold over the years. Large banks could be running anywhere from ~100-500+ software systems, which lends itself to an incredible amount of complexity to be able to latch on to another core. Making a shift from the current paradigm is akin to taking off in an airbus 380 to land in a boeing 787.
And the primary reason banks should be very concerned is that rapid digitalization is leading to embedded-banking, where the customer experience layer is rapidly moving over to the non-banks. These non-banks are operating at internet-scale and could possibly deploy the ai-arsenal at a much faster clip - thereby creating a fundamental advantage for themselves.
What can the banks do? Like it or not, the wave of AI enabled systems is coming - and needs to be thoughtfully architected. Mindlessly adding to the tech debt of the organization is bound to metamorphose the spaghetti into a soup.
Setup the ai- 10 point agenda- which I believe needs to be driven from the top to be able to rise up to the challenge. As always, there will be haves and have-nots in this transition, with the difference, this time around the wedge is expected to be 10x wider.
Will share some pointers from the agenda in a subsequent post.
Aside, there’s a lot for the fintechs to do here too. This change from a deterministic world of banking needs fintechs to usher in products and services in every part of the banking world. For the right products, there is a ~Rs. 2000cr revenue opportunity for the tech-fins and ~Rs 15,000crore revenue on the business products. More on this later too.
The ecosystem should push together and make sure Indian banking emerges at the top of the curve, and gets ahead rapidly. Sun, willing.
