What the leadership team must decide
Ten decisions sit between Meridian and a captured AI transformation. Each has one owner, a defined trigger, and evidence drawn from the same baseline. This is the page the leadership team works from together. Everyone reads the same task-level picture, cut four ways.
The decisions are ordered by data dependency, not by department. Structured sanction-term capture in Operations makes covenant automation possible. Covenant automation sets the data quality for early-warning signals. Clean early-warning signals make the renewal redesign possible. The renewal redesign is what frees analyst capacity.
The work behind every decision lives one layer down, in the Workflow Atlas: the work as it runs today, where AI takes the load, the redesigned future state, and the human work each redesign grows. Open the Workflow Atlas.
| Seq | ID | Decision | Owner | Status | Trigger to move |
|---|---|---|---|---|---|
| 1 | D1 | Which workflows enter the Phase 1 redesign queue, and in what order | COOCEO sign-off | Decide now | The baseline is complete. No further input is needed to commit Phase 1. |
| 2 | D5 | Where the credit decision sits in the new operating model. The sanction stays human H5 permanent | COOCEO sign-off | Decide now | The sanction authority principle is settled. The operating model follows from it. |
| 3 | D2 | Approve Phase 1 investment against the per-workflow business case | CFOCEO sign-off | Decide now | The ROI ranges and the assumption register are on the table. |
| 4 | D3 | Set the relationship-manager to portfolio ratio as servicing work shifts to AI | CHROCOO input | Decide now | The Senior RM task analysis is scored and stable. |
| 5 | D8 | Name who owns the freed analyst capacity, and toward what | CEO | Decide now | Without a named owner, freed capacity refills with the existing backlog. |
| 6 | D9 | Sign off the team verdict for each team before redesign proceeds | CEOCHRO input | On trigger | Verdicts stay preliminary until they are checked against the workflow designs. |
| 7 | D4 | Reshape the credit analyst team around the capacity that gets freed | CHRO | On trigger | Moves once the renewal redesign ships and the freed capacity is observed, not assumed. |
| 8 | D6 | Redesign the early-career analyst path for AI-supported work | CHRO | On trigger | Moves after D4 sets the future shape of the analyst team. |
| 9 | D7 | Refresh the hiring profile for relationship-manager and analyst roles | CHROL&D handoff | On trigger | Moves once the capability gap map is turned into a learning plan with L&D. |
| 10 | D10 | Adopt quarterly recalibration as the rhythm for the business case | CFO | Decide now | Replaces a fixed multi-year projection. The decision is about cadence, not amount. |
Is the change real, and is it in the right order?
You approved AI investment to change how the bank runs. This page shows whether the work is being redesigned, or whether tools are being laid on top of unchanged work. Three calls are yours: the pace, the order, and who owns the freed capacity.
The baseline is complete. No further input is needed to commit Phase 1.
The sanction authority principle is settled. The operating model follows from it.
Without a named owner, freed capacity refills with the existing backlog.
Verdicts stay preliminary until they are checked against the workflow designs.
Which numbers can we defend, and what does waiting cost?
Every number on this page traces to a scored task or a redesigned workflow. The numbers we cannot defend that way are held outside the base case, on purpose. The base case is Rs 7 crore to Rs 13 crore a year. You are signing off a position you can hold in front of anyone.
The ROI ranges and the assumption register are on the table.
Replaces a fixed multi-year projection. The decision is about cadence, not amount.
Which work changes next, and where does authority sit?
You know which tools are deployed. This page shows which workflows have actually been redesigned around them, where people keep decision authority, and which AI agents run in the loop and under what watch. The sanction stays fully human, by design.
The baseline is complete. No further input is needed to commit Phase 1.
The sanction authority principle is settled. The operating model follows from it.
Each step produces the data the next one needs.
- 1Sanction-term capture Atlasstructures the sanction terms. Nothing downstream works without it.
- 2Covenant and early-warning monitoring Atlasbecomes possible once terms are captured as structured data.
- 3Annual credit renewal Atlasbuilds on clean early-warning signals and frees analyst capacity.
- 4Relationship-manager portfolio servicing Atlasfrees selling time as servicing shifts to AI support.
| Capability | What moves to AI | What people do more of |
|---|---|---|
| Credit judgment and risk calibration | Data assembly, spreading, and ratio computation | Concentrating judgment on the flagged exposures |
| Early-warning interpretation | The twelve-month reconstruction becomes a continuous digest | Reading the early-warning signal stream |
| Exception and anomaly judgment | First-pass automation on every step | Judging the exceptions each step routes up |
| Client and promoter dialogue | Document follow-up and chasing | Client and promoter conversations |
What shape does the team take, and is it ready in time?
You turn the change into people decisions: team shape, role redesign, hiring, and the early-career path. The gap that matters is between capability today and the capability the timeline needs. Readiness is 56 out of 100. The work is to close that gap before the redesigned workflows arrive.
The Senior RM task analysis is scored and stable.
Moves once the renewal redesign ships and the freed capacity is observed, not assumed.
Moves after D4 sets the future shape of the analyst team.
Moves once the capability gap map is turned into a learning plan with L&D.
Readiness 56 out of 100, computed as the mean of the scores above. Never typed in.
What actually changes in the work
Every decision on the other pages points to a workflow. This is where the workflows live: the work as it runs today, where AI takes the load, the redesigned future state, and the human work each redesign grows. One workflow is shown in full. The engagement covers them all this way.
| Segment | HAS shift | What moves to AI | What people do more of |
|---|---|---|---|
| Document collection | H4→H2 | Account Aggregator pulls statements. Document tools parse the financials. Follow-up is automated. | Chasing replaced by checking the exceptions |
| Financial spreading | H4→H1 | Auto-spread from the parsed documents | Review of exceptions only |
| Ratio and trend analysis | H4→H2 | Computes and flags | Interpreting the anomalies |
| Account conduct review | H4→H2 | A monitoring agent runs year-round | Reading the digest instead of rebuilding twelve months |
| Risk note | H4→H3 | Drafts the standard sections | Writing the judgment sections and owning the recommendation |
| RM recommendation | H5→H4 | Surfaces pricing comparables and relationship history | Making the call |
| Sanction decision | H5→H5 | Nothing | Deciding, with full accountability |
| Documentation and limit setup | H4→H2 | Terms captured structured at source. Limits auto-configure. | Clearing the deviations that route to a queue |
AI absorbs the load
Human work amplified
| # | Assumption | Basis |
|---|---|---|
| A-03 | Renewal volume per year, about 1,200 files. | public benchmark, indicative |
| A-04 | Analyst hours per file today, 22 to 30 hours. | Stage 2 interview range, reconciled |
| A-07 | Exception rate on the automated steps, 15 to 20 percent. | document-processing benchmark, Indian financials |
| A-09 | Adoption ramp to steady state, about 2 quarters. | phased rollout, one region first |
| A-11 | Structured-capture coverage of 80 percent of sanctions or higher. This is the gating dependency. Below it, every downstream range widens. | gating assumption, indicative |