The shared picture

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.

SeqIDDecisionOwnerStatusTrigger to move
1D1Which workflows enter the Phase 1 redesign queue, and in what orderCOOCEO sign-offDecide nowThe baseline is complete. No further input is needed to commit Phase 1.
2D5Where the credit decision sits in the new operating model. The sanction stays human H5 permanentCOOCEO sign-offDecide nowThe sanction authority principle is settled. The operating model follows from it.
3D2Approve Phase 1 investment against the per-workflow business caseCFOCEO sign-offDecide nowThe ROI ranges and the assumption register are on the table.
4D3Set the relationship-manager to portfolio ratio as servicing work shifts to AICHROCOO inputDecide nowThe Senior RM task analysis is scored and stable.
5D8Name who owns the freed analyst capacity, and toward whatCEODecide nowWithout a named owner, freed capacity refills with the existing backlog.
6D9Sign off the team verdict for each team before redesign proceedsCEOCHRO inputOn triggerVerdicts stay preliminary until they are checked against the workflow designs.
7D4Reshape the credit analyst team around the capacity that gets freedCHROOn triggerMoves once the renewal redesign ships and the freed capacity is observed, not assumed.
8D6Redesign the early-career analyst path for AI-supported workCHROOn triggerMoves after D4 sets the future shape of the analyst team.
9D7Refresh the hiring profile for relationship-manager and analyst rolesCHROL&D handoffOn triggerMoves once the capability gap map is turned into a learning plan with L&D.
10D10Adopt quarterly recalibration as the rhythm for the business caseCFODecide nowReplaces a fixed multi-year projection. The decision is about cadence, not amount.
Illustrative readout. Illustrative readout. Meridian Bank is a synthetic Indian mid-market commercial bank. Figures are indicative and anchored to public Indian banking benchmarks, not a live engagement. The baseline is a point-in-time picture. Quarterly recalibration reports only what has changed against it. Base case Rs 7 crore to Rs 13 crore a year, summed from the workflows. Readiness 56 out of 100, computed.
CEO · Operating Model

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.

Workflows in Phase 1
4
Sequenced by dependency
Capacity freed
35-45 %
Analyst hours, full Phase 1
Freed capacity with an owner
0 %
Until D8 is committed
D1
Which workflows enter the Phase 1 redesign queue, and in what order
Owner COOCEO sign-off
Decide now
When this can move

The baseline is complete. No further input is needed to commit Phase 1.

SOURCE Leadership Table, decision D1
D5
Where the credit decision sits in the new operating model. The sanction stays human H5 permanent
Owner COOCEO sign-off
Decide now
When this can move

The sanction authority principle is settled. The operating model follows from it.

SOURCE Leadership Table, decision D5
D8
Name who owns the freed analyst capacity, and toward what
Owner CEO
Decide now
When this can move

Without a named owner, freed capacity refills with the existing backlog.

SOURCE Leadership Table, decision D8
D9
Sign off the team verdict for each team before redesign proceeds
Owner CEOCHRO input
On trigger
When this can move

Verdicts stay preliminary until they are checked against the workflow designs.

SOURCE Leadership Table, decision D9
Sanction-term captureOperations. Structures the data everything downstream needs.
automate
Covenant and early-warning monitoringCredit and Risk. Depends on the Operations capture.
agentize
Annual credit renewalCredit and Risk. Frees the analyst capacity.
redesign
Relationship-manager servicingCoverage. Servicing load moves to AI. Selling time grows.
augment
What would change this picture next quarter
Capability shift
A model release that makes covenant reading reliable would move that work from agent-run toward fully automated, and pull the renewal redesign forward a quarter.
Regulatory
An RBI position on AI in credit decisioning would confirm or move the sanction line. Tracked as a standing watch item.
Drift
If the redesigned workflow count stalls, the program is laying tools on top of unchanged work. That is the number to watch.
CFO · Value and Capital

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.

Spend tied to a task change
100 %
Base case only. Quarantine excluded.
Phase 1 horizon
2 qtrs
Committed, then recalibrated
ROI shown as
Ranges
Point estimates hide assumptions
D2
Approve Phase 1 investment against the per-workflow business case
Owner CFOCEO sign-off
Decide now
When this can move

The ROI ranges and the assumption register are on the table.

SOURCE Leadership Table, decision D2
D10
Adopt quarterly recalibration as the rhythm for the business case
Owner CFO
Decide now
When this can move

Replaces a fixed multi-year projection. The decision is about cadence, not amount.

SOURCE Leadership Table, decision D10
Sanction-term capture · automate · AtlasRs 1 crore to Rs 2 crore
0Rs 6.5 croreRs 13 crore
Covenant and early-warning monitoring · agentize · AtlasRs 1.5 crore to Rs 3 crore
0Rs 6.5 croreRs 13 crore
Annual credit renewal · redesign · AtlasRs 2.5 crore to Rs 4.5 crore
0Rs 6.5 croreRs 13 crore
Relationship-manager portfolio servicing · augment · AtlasRs 2 crore to Rs 3.5 crore
0Rs 6.5 croreRs 13 crore
Held outside the base case
Rs 13 crore to Rs 26 crore
Earlier, AI-driven early-warning signals could lower credit costs. It is the largest number in this engagement, and it is deliberately left out of the base case. Why it is held out: It depends on downstream behaviour the bank has not yet shown: acting on the signals in time. Release test: Two quarters of acting on early-warning alerts on schedule moves part of this into the base case.
What would change this picture next quarter
Capture rate
If the renewal redesign ships and freed capacity lands at the top of the 35 to 45 percent band, the renewal return moves toward its upper bound and part of the quarantine can move into the base case.
Data readiness
Sanction-term capture is the gating step. If structured-capture coverage stays below 80 percent, every downstream range widens and the timeline slips.
COO · Work Redesign

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.

In the Phase 1 queue
4 flows
Sequenced by dependency
Available but not yet captured
~30 %
Productivity on the table today
Agents under human oversight
All
The sanction is never delegated
D1
Which workflows enter the Phase 1 redesign queue, and in what order
Owner COOCEO sign-off
Decide now
When this can move

The baseline is complete. No further input is needed to commit Phase 1.

SOURCE Leadership Table, decision D1
D5
Where the credit decision sits in the new operating model. The sanction stays human H5 permanent
Owner COOCEO sign-off
Decide now
When this can move

The sanction authority principle is settled. The operating model follows from it.

SOURCE Leadership Table, decision D5

Each step produces the data the next one needs.

Document intake
H1
H2
Spreading and ratio work
H2
H3
Risk-note drafting
H3
H4
Sanction decision
H5
Nor Human Nor Machine
H1AI entirelyH2AI leadsH3equalH4human leadsH5fully human
The sanction stays fully human (H5), by design. The credit sanction stays fully human. AI prepares the case, structures the evidence, and flags the exceptions. The judgment and its accountability stay with the person. This is a principle of the operating model, not a stage it grows out of. See decision D5 in the Leadership Table.
CapabilityWhat moves to AIWhat people do more of
Credit judgment and risk calibrationData assembly, spreading, and ratio computationConcentrating judgment on the flagged exposures
Early-warning interpretationThe twelve-month reconstruction becomes a continuous digestReading the early-warning signal stream
Exception and anomaly judgmentFirst-pass automation on every stepJudging the exceptions each step routes up
Client and promoter dialogueDocument follow-up and chasingClient and promoter conversations
What would change this picture next quarter
Segment shift
Risk-note drafting is the segment most likely to move toward AI-led as models improve. The sanction segment is fixed and is not on the watch list.
Exception load
If exceptions in covenant monitoring run above the design threshold, the oversight model tightens before the workflow scales.
CHRO · Workforce and Capability

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.

Decisions you lead
4
Of the ten on the table
Readiness
56 /100
Computed, mean of the capabilities below
Where reskilling goes first
Early-warning
The widest gap to close
D3
Set the relationship-manager to portfolio ratio as servicing work shifts to AI
Owner CHROCOO input
Decide now
When this can move

The Senior RM task analysis is scored and stable.

SOURCE Leadership Table, decision D3
D4
Reshape the credit analyst team around the capacity that gets freed
Owner CHRO
On trigger
When this can move

Moves once the renewal redesign ships and the freed capacity is observed, not assumed.

SOURCE Leadership Table, decision D4
D6
Redesign the early-career analyst path for AI-supported work
Owner CHRO
On trigger
When this can move

Moves after D4 sets the future shape of the analyst team.

SOURCE Leadership Table, decision D6
D7
Refresh the hiring profile for relationship-manager and analyst roles
Owner CHROL&D handoff
On trigger
When this can move

Moves once the capability gap map is turned into a learning plan with L&D.

SOURCE Leadership Table, decision D7
AI-assisted spreading
-35
Early-warning interpretation
-38
Exception and anomaly judgment
-20
Covenant structuring
-7
Client and promoter dialogue
-3
Credit judgment and risk calibration
-22
TodayTimeline target

Readiness 56 out of 100, computed as the mean of the scores above. Never typed in.

What would change this picture next quarter
Readiness
AI-assisted spreading and early-warning interpretation are the gating gaps. If readiness on these does not move, the redesign ships into a team that cannot run it.
Pipeline
The early-career redesign sets whether the next analyst intake arrives ready for AI-supported work. The hiring profile follows from it.
Workflow Atlas · The work itself

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.

How to read this layer. Current state comes first, because no executive carries the workflow in their head. Then the shift, segment by segment. Then the redesigned state, with agents marked and human checkpoints fixed. Then what the people do more of. The numbers at the bottom are the ones the CFO page quotes.
Current state Cycle 35 to 45 days · about 26 analyst hours per file · about 1,200 renewals a year
Step 1 · 15%
Renewal trigger and document collection
RM + CSA
H4 today
Two to three weeks of client follow-up. About 60 percent of files arrive incomplete on the first pass.
Step 2 · 20%
Financial spreading
Credit Analyst
H4 today
Manual re-keying from audited financials. Transcription errors surface later.
Step 3 · 10%
Ratio and trend analysis
Credit Analyst
H4 today
Step 4 · 10%
Account conduct review
Credit Analyst
H4 today
Twelve months of statements reviewed once a year, at renewal, after the fact.
Step 5 · 25%
Risk note and appraisal
Credit Analyst
H4 today
A 15 to 25 page note. The largest single block of analyst time.
Step 6 · 5%
RM recommendation and pricing
RM
H5 today
Step 7 · 5%
Sanction decision
Committee
H5 today
Step 8 · 10%
Documentation and limit setup
Operations
H4 today
Sanction terms re-keyed from PDF letters. Unstructured terms break monitoring.
The shift, per segment
SegmentHAS shiftWhat moves to AIWhat people do more of
Document collectionH4H2Account Aggregator pulls statements. Document tools parse the financials. Follow-up is automated.Chasing replaced by checking the exceptions
Financial spreadingH4H1Auto-spread from the parsed documentsReview of exceptions only
Ratio and trend analysisH4H2Computes and flagsInterpreting the anomalies
Account conduct reviewH4H2A monitoring agent runs year-roundReading the digest instead of rebuilding twelve months
Risk noteH4H3Drafts the standard sectionsWriting the judgment sections and owning the recommendation
RM recommendationH5H4Surfaces pricing comparables and relationship historyMaking the call
Sanction decisionH5H5NothingDeciding, with full accountability
Documentation and limit setupH4H2Terms captured structured at source. Limits auto-configure.Clearing the deviations that route to a queue
Future state Cycle 12 to 18 days · about 11 analyst hours per file · event-triggered, not calendar-bound
Step 1
Continuous data feed
AI Agent
H2
Oversight: exception queue for failed pulls and consent gaps.
Step 2
Auto-spread and ratio pack
AI
H1-H2
Analyst spot-checks a sample. Exceptions surface on their own.
Step 3
Conduct and early-warning digest
AI Agent
H2
Year-round monitoring, read at renewal.
Step 4
Analyst judgment on flagged areas
Analyst
H3-H4
Human checkpoint. Risk work concentrated where the signals point.
Step 5
AI-drafted note, analyst-owned judgment
Analyst + AI
H3
Step 6
RM pricing and recommendation
RM
H4
Step 7
Sanction decision
Committee
H5, permanent
Step 8
Structured term capture and auto limit setup
AI + Ops
H2
Exception path: deviations route to a manual queue.
What the people do more of

AI absorbs the load

Data assembly and spreadingCollection, re-keying, and ratio computation. The mechanical 55 to 60 percent of the file.
Twelve-month reconstructionConduct review shifts from an annual rebuild to a continuous digest.
Standard draftingThe boilerplate sections of the note, to the committee template.
Term transcriptionSanction terms captured structured at source, not re-keyed later.

Human work amplified

Credit judgmentAnalyst time concentrates on flagged exposures, not mechanical assembly.
Early-warning interpretationA continuous signal stream is only useful if someone reads it. This becomes the analyst's defining skill.
Exception and anomaly judgmentEvery automated step routes its exceptions to a person. Judging them well is the new core of the role.
Client and promoter dialogueRM time freed from chasing documents goes to the conversations that protect the book.
The value, and what it rests on
Cycle time
35-45d12-18 days
Analyst hours per file
~26~11 hrs
Capacity freed
35-45 %
Annual value
Rs 2.5 crore to Rs 4.5 crore
Assumption register: the rows behind the number
#AssumptionBasis
A-03Renewal volume per year, about 1,200 files.public benchmark, indicative
A-04Analyst hours per file today, 22 to 30 hours.Stage 2 interview range, reconciled
A-07Exception rate on the automated steps, 15 to 20 percent.document-processing benchmark, Indian financials
A-09Adoption ramp to steady state, about 2 quarters.phased rollout, one region first
A-11Structured-capture coverage of 80 percent of sanctions or higher. This is the gating dependency. Below it, every downstream range widens.gating assumption, indicative

Sanction-term capture

automate
Re-keyed from PDF letters today. Unstructured terms break monitoring.
H4H2 Structured capture at source
Exception and anomaly judgmentRegulatory and compliance grounding
Gating dependency for the whole chain · Rs 1 crore to Rs 2 crore
SUMMARY DEPTH · full strips produced in engagement

Covenant and early-warning monitoring

agentize
Periodic and manual today. Breaches surface late, often at renewal.
H4H2 Year-round monitoring agent
Early-warning interpretationCredit judgment and risk calibration
Feeds the renewal redesign · Rs 1.5 crore to Rs 3 crore
SUMMARY DEPTH · full strips produced in engagement

Relationship-manager portfolio servicing

augment
An estimated 50 to 60 percent of the week goes to servicing mechanics today.
H4H2 Servicing load to AI. The relationship stays human.
Client and promoter dialogueCredit judgment and risk calibration
Feeds decision D3, the RM portfolio ratio · Rs 2 crore to Rs 3.5 crore
SUMMARY DEPTH · full strips produced in engagement
Schema note. Current-state summaries are carried in the FlowDNA artifact as a summary-grade baseline block, from Stage 2 interviews. FlowDNA stays a future-state design. The baseline block exists so the contrast renders without asking the reader to hold it in memory.