May 2026
The Operations Brief
BankIslami
Three opportunities to reduce the cost base as BankIslami rebuilds toward its target efficiency ratio. A starting point for what we believe is a larger conversation.
By Lightbloom · An AI Operating Partner
Potential savings: €1.3-2.1M per year
Cost-to-income ratio · FY2024 vs FY2025
43.5%
Cost-to-income, FY2024
Best efficiency ratio in five years
69.6%
Cost-to-income, FY2025
26-point swing in a single year
Income fell PKR 5.8bn as SBP rate cuts compressed spreads. Operating expenses rose PKR 9.2bn as branch expansion and technology investment landed simultaneously. The gap widened by PKR 15bn in one year.
Fig. 1 · Profit before tax · EUR million · FY2022-FY2025
BankIslami is Pakistan's second-largest dedicated Islamic bank, operating 569 branches across 123 cities with 1.82 million customers and 7,350 employees. From FY2021 to FY2024, it executed a near-textbook efficiency improvement, cutting cost-to-income from 74.7% to 43.5% while growing profit before tax tenfold to PKR 25.5bn. FY2025 reversed both dynamics at once: SBP rate cuts compressed net mark-up income by 23.1%, while branch expansion and the iMAL R14 and Aik digital platform investments pushed operating expenses up 41.8%. The COO, Sohail Sikandar, is publicly calling for AI-driven operational efficiency. The stated target is a return toward 50% cost-to-income. These three opportunities are part of how that happens.
Three key opportunities identified.
Assembled from public sources only: BankIslami Annual Reports 2024 and 2025 (PSX filings), company website, Azentio iMAL case study, and press coverage. Lightbloom has seen no internal data.
What is visible from the outside is typically 40-60% of what a full operational review surfaces once we have access to iMAL transaction cost data, branch-level headcount breakdowns by role, ATM operations logs, and vendor contract schedules. The opportunities below are conservative and defensible from public evidence. The picture from inside is larger.
1
Opportunity One
Replenish the ATM on demand, not the calendar
How an AI demand forecasting agent reads 24 months of per-machine transaction history and replaces static replenishment schedules with an optimised weekly dispatch plan across 643 ATMs.
Finding
BankIslami's ATM network grew 17% in FY2025, from 551 to 643 machines across 123 cities. Each ATM requires regular cash replenishment. In Pakistan's cash-intensive economy, urban ATMs in salary districts draw down heavily around payroll cycles, Eid, and Ramadan, while low-traffic rural ATMs may go days without meaningful withdrawals. The bank's current replenishment approach is scheduled rather than demand-driven: fixed frequencies that cannot adapt to the calendar-driven demand spikes that define Pakistani ATM usage. At 643 ATMs on a twice-weekly average replenishment cycle, BankIslami executes approximately 66,800 armored vehicle trips per year. The cost per trip in Pakistan's security environment runs PKR 10,000-15,000. Neither the logistics cost nor the excess float cost of cash loaded but not withdrawn before the next replenishment appears as a line item in public accounts. Both are structural features of operating a large ATM network without demand forecasting.
66,800 armored vehicle trips per year. Every one of them on the same fixed schedule regardless of what the machine actually needs.
Our Solution
An ATM demand forecasting AI agent reads 24-36 months of transaction data per machine, integrates external demand signals (payroll cycle dates, public holidays, Ramadan and Eid calendar, proximity to commercial districts), and generates optimised cash loading quantities and replenishment schedules per ATM cluster. High-demand ATMs around payroll dates receive pre-loaded additional float. Low-velocity rural ATMs shift from twice-weekly to twice-monthly replenishment during off-peak periods. The output is a weekly replenishment dispatch schedule for the security logistics team, replacing the static calendar. The SBP uptime floor is maintained as a hard constraint throughout. The model does not reduce service levels. It reduces the trips that do not change them.
Estimated annual value
01 / 03
€420-645K
per year
643 ATMs x 2 replenishments/week x 52 weeks = 66,836 trips/year at PKR 12,500/trip average = PKR 835M/year logistics cost. 20% trip reduction = PKR 167M saving. Float efficiency: 643 ATMs x PKR 2M average cash holding x 5% annualised opportunity cost = PKR 77M. Gross PKR 244M. Conservative 55-82% capture (regulatory minimum-frequency constraints) = PKR 130-200M = €420-645K.
2
Opportunity Two
Run one operations hub, not 569
How a shared operations AI agent centralises end-of-day balancing, regulatory data preparation, and exception review from 569 branches into hub clusters, reducing back-office headcount cost without touching customer service.
Finding
BankIslami expanded from approximately 380 branches in FY2021 to 569 in FY2025, adding nearly 200 branches in four years. Total headcount reached 7,319 by year-end FY2024. Each branch runs dedicated operations staff for end-of-day balancing, voucher verification, daily SBP regulatory data preparation, and exception resolution. These tasks are structurally identical across every branch: digital, rule-based, system-to-system processes that do not require physical presence at the specific branch they relate to. Pakistan's branch banking back-office model was designed for paper vouchers. Most banks have not restructured it as digital records replaced physical ones. At 569 branches, BankIslami is running an estimated 850-1,140 operations staff doing work that is centralizable today.
569 branches. Each one doing the same end-of-day operations work independently, every working day.
Our Solution
A shared operations hub model handles all non-customer-facing daily processing for clusters of 25-30 branches: end-of-day reconciliation, regulatory SBP data submission, exception review, and voucher classification. Each branch retains one operations officer for physical cash handling and walk-in customer queries. An AI exception-flagging agent reads daily transaction data from iMAL against rule sets and surfaces anomalies as a prioritised queue for hub analysts, enabling each hub to cover more branches without quality degradation. The model pilots on the 29 most recently opened branches plus high-density metro clusters in Karachi, Lahore, and Islamabad where hub logistics and connectivity are simplest. Branch managers see a daily processing summary pushed to them rather than generated by them.
Estimated annual value
02 / 03
€645K-1.1M
per year
1.5 ops FTE per branch x 569 branches = 854 ops FTE network-wide. Hub model: 23 hubs x 5 staff + 569 x 0.5 retained branch officer = 400 FTE total. Year-1 target (175 branches across newest and metro clusters): 175 FTE reduction x PKR 80,000/month x 12 = PKR 168M. Year-2 extension to 300 branches = PKR 288M. Range PKR 200-350M = €645K-1.1M. Steady-state full-network: PKR 436M/year.
3
Opportunity Three
Process the KYC document, not the paperwork
How an AI document intelligence agent handles straight-through KYC processing and Shariah contract drafting for 1.82 million customers across the Aik digital channel and 569 branches.
Finding
BankIslami serves 1.82 million customers, up 16% in FY2025 alone, and is adding more through the Aik digital platform. SBP regulations require KYC verification at account opening and periodic renewal on approximately 2.5-year cycles. At 1.82M customers, that is approximately 730,000 KYC events per year. Every Islamic financing transaction additionally requires a bespoke Shariah contract document: each Murabaha, Diminishing Musharakah, or Mudarabah facility assembled from structured data already held in iMAL. The Shariah Compliance Department reviewed 199 branches in FY2025 for documentation quality, a signal of how seriously the bank treats this and how manually it is managed today. BankIslami's NADRA biometric integration is already confirmed active in the Aik onboarding flow. The infrastructure exists. The automation layer on top of it does not.
730,000 KYC events per year. NADRA integration is already live in Aik. The automation layer is not.
Our Solution
An AI KYC processing agent reads submitted customer documents via optical character recognition, extracts structured data, validates CNIC numbers against the NADRA API (already live in Aik), pre-fills all iMAL account fields, and routes only genuinely ambiguous cases to a human reviewer. Standard CNIC-plus-utility-bill cases, estimated at 70-80% of volume, process straight through without manual data entry. A parallel Shariah contract document AI agent pulls structured financing parameters from iMAL and auto-populates the standard contract template for each facility type, generating a review-ready draft. The compliance officer reviews and finalises. The assembly is automated. For a bank that markets itself on Islamic financial integrity, Shariah documentation accuracy is not just an operational matter. Getting it consistently right at scale requires a system, not individual vigilance.
Estimated annual value
03 / 03
€225-355K
per year
150-200 KYC/compliance FTE at PKR 75,000/month: PKR 135-180M/year. 70-80% straight-through processing frees 30-40 FTE via natural attrition x PKR 75,000 x 12 = PKR 27-36M. External KYC support reduction (40% of PKR 20-30M estimated) = PKR 8-12M. Overtime reduction at peak cycles = PKR 10-15M. Shariah document prep freed (20 FTE equivalent) = PKR 18M. Total PKR 63-81M; conservative range PKR 70-110M = €225-355K.
Before anything else
We validate the numbers first. Then we build.
Nothing here becomes a commitment until the math is validated against BankIslami's actual iMAL data, ATM records, and branch staffing. If the numbers hold, Lightbloom builds the fixes specific to BankIslami's stack. They are AI agent workflows running on Yield, the AI operating platform we build and use internally, and they keep running. We earn 20% of Confirmed Annual Value per opportunity, once, after the savings land in BankIslami's accounts. Nothing before that.
-
Week 1
01 / 04
iMAL and ATM systems landing
Read-only extraction from iMAL R14 transaction records, ATM management system, and branch operations logs. Data quality assessed against all three opportunity workstreams. No process changes proposed yet.
-
Week 2
02 / 04
ATM replenishment cost audit
Pull 24 months of per-machine transaction data. Map current replenishment schedule against actual demand patterns by ATM cluster. Quantify the logistics cost at current trip frequency and identify the top 100 machines where demand-driven scheduling has the clearest case.
-
Week 3
03 / 04
Branch ops mapping and KYC volume
Structured interviews with 20-25 branch operations officers across branch sizes to confirm daily workflow and FTE per branch. Simultaneously, pull KYC event volume from iMAL by channel (Aik vs branch) and document type to baseline the automation opportunity.
-
Week 4
04 / 04
Joint readout · commitment
Baseline validated against BankIslami's actual books. Scope and sequencing of the build agreed. The engagement begins or it does not. Both ends are honest.
Together, three opportunities
Opportunity One
€420-645K
Opportunity Two
€645K-1.1M
Opportunity Three
€225-355K
· sum ·
€1.3-2.1M
per year, recurring
Identifiable from public sources only. The four weeks will confirm whether it holds. Internal access reveals what public sources cannot.
Where this goes next
A thirty-minute call. Nothing more until it is mutual.
We walk through the brief together: your reactions, what we got wrong, what we missed. If the four-week diagnostic still makes sense at the end, we scope it. If not, we shake hands. Lightbloom only earns when BankIslami does.