EASE 9.0 · Innovation · Business Process Re-engineeringThe Delay Was Architecture, Not People
How agents enable process redesign.
Turning sequential, multi-desk workflows into parallel, agent-based execution that collapses a multi-day cycle into minutes.
A loan application at a typical PSU bank moves through a sequential chain: intake desk, documentation check, credit desk, compliance desk, sanction committee. Each desk works on its own queue and hands the file forward only once its own step is complete.
Sequential Handoffs Versus Parallel Execution
Business Process Re-engineering, in the EASE 9.0 sense, means redesigning the sequence itself, not making each desk faster: replacing a chain of sequential handoffs with parallel, agent-based execution, where multiple checks run simultaneously instead of queuing behind each other.
The Path to Progress
The delay in a five-desk approval process is architectural, not a matter of any one team working slowly.
The Business Case
One PSU bank's own deployment is the cleanest illustration, and it is a live PSU precedent rather than a hypothetical one. Collapsing sequential desks into one coordinated pass changed more than the speed of the same approvals. It changed which applications got approved at all, as cases that used to fall through the cracks between desks (an intake-desk note that never reached the compliance desk, a document flagged at one step but not re-checked at the next) surfaced instead of silently stalling. The reported qualified-response gain from this deployment is detailed in full on the Customer Experience Transformation page, since it is one number measured there, not a separate proof point to re-derive here.
The same mechanism shows up outside India too. Absa Bank in South Africa rebuilt SME onboarding on Salesforce Agentforce, collapsing an 87-field, roughly 420-minute manual KYC process into account-opening that completes in minutes, with issue resolution 88% faster, independently recognized with a 2026 Celent Model Bank Award. Different bank, different process, the same underlying redesign: parallel evaluation of everything the old process checked sequentially, rather than a faster version of the same sequential chain.
Source: Salesforce, “Absa Bank's Blueprint for the Agentic Enterprise” and Absa Group customer story (2026).
Where AI Agents Fit
A customer never sees the internal sequence of desks that used to process their file; they only see how long the answer took. The deployment cited above ran on Appice's real-time campaign layer, not a full agentic re-engineering of the underlying approval workflow; extended into that next step, this is what Decide applying compliance gates as encoded policy evaluated in parallel, instead of a sequential sign-off passed desk to desk, would look like, with Inform replacing the batch reporting that used to hide how the process was actually performing. This is backend restructuring work; its entire value would show up as speed and cost somewhere else, most visibly in Cost Optimization & Income Growth and Customer Experience Transformation.
Take the specific 87 fields Absa's onboarding process checked one desk at a time. Identity verification, a credit-bureau pull, AML and sanctions screening, and document authenticity have no actual dependency on each other; a sequential process runs them one after another only because the desks were originally organized around people, not around what each check requires. An agent firing identity verification, the bureau pull, sanctions screening, and document checks simultaneously, then reconciling all four results against policy, is what actually collapses 420 minutes into single digits: not a faster version of the same four-desk queue, but the queue itself becoming unnecessary once nothing is genuinely waiting on anything else. Exception handling is the one piece that deliberately stays sequential: a mismatched field or a sanctions hit still routes to a person, because that is the one decision this category does not automate away.
What Makes Re-engineering Hard
The technical redesign is usually the easier half. The harder half is organizational: each desk in the old sequential chain has an owner, a headcount allocation, and a set of performance metrics built around processing volume at that desk. Collapsing five desks into one automated pass means renegotiating what those roles do next. Banks that treat this purely as a technology rollout, without a change-management plan for the affected desks, find the old process quietly persists in parallel with the new one, because no one has been given a mandate to retire it.
The Skeptic's Answer
Every desk staffer in the old sequential chain has heard the word "re-engineering" before, and it usually meant one thing: a consulting deck that ends in headcount reduction, dressed up in process language. Staff who recognize that pattern will slow-walk the pilot, and a programme the affected desks are quietly working against does not ship.
The fear is sometimes justified and sometimes not; the difference lies in what happens to the people at the collapsed desks. This deployment's results, and Absa's, go beyond faster processing. Absa's issue resolution ran 88% faster specifically because staff moved from manually checking 87 fields to supervising exceptions the system flagged, a different and arguably more skilled job, not fewer of the same job. A redesign with no answer for where desk staff go next is the one worth resisting. A redesign with a specific redeployment plan, into exception handling rather than unemployment, is a different proposition, and it is the one that gets adopted rather than sabotaged.
Redesign Around What the Bank Is Actually Deciding
Appice's platform models the decision a bank needs to make, then lets the agent execute every check that decision requires in parallel; the old sequential chain of desks was never the thing worth preserving, only the checks each desk happened to perform. Because the decisioning engine runs on the bank's own infrastructure, that model belongs to the bank once it is built. When the next process comes up for redesign, and in a PSU bank there is always a next one, the engineering already paid for gets reused, and the bank sets the shape of its own workflow going forward.