Case Study

How a Major Islamic Bank Deployed an On-Premises AI Engine Across Web, WhatsApp, and Email

How a major Islamic bank deployed an on-premises AI engine to answer product and policy questions across web, WhatsApp, and thousands of daily emails

  • Industry: Banking and Finance

  • Deployment: On-premises

  • Engine: AutomatiCworX’s reasoning engine, deployed as a question-answering layer, not paired with action execution

  • Where it runs: the bank’s website, WhatsApp banking, and as an internal tool for the service quality team answering customer email

  • Scope: Product information, financing calculations, account services, and policy questions across the bank’s full retail and digital offering

The customer

A large Islamic commercial bank with an extensive branch network, WhatsApp banking, a mobile and digital account platform, and a service quality team that handles a high daily volume of customer email. Its product range spans everyday retail banking alongside Shariah-compliant financing products, house financing, Car and Bike Ijarah, and digital account offerings aimed at resident and non-resident customers alike. That breadth is also the support challenge: a customer might ask a SWIFT code question on the website one minute, an Ijarah instalment question over WhatsApp the next, and email the service quality team about something else entirely, and all three need to be right.

The challenge

The bank’s product catalogue is wide and the questions customers ask about it are specific, and they arrive through three different channels with three different expectations. A website visitor wants an immediate answer. A WhatsApp customer expects the same conversational speed they get from anyone else on the platform. And the service quality team was working through thousands of emails a day, many of them the same handful of questions asked in slightly different words, financing terms, account opening documentation, profit rates, schedule of charges, each one requiring an agent to look up or already know the correct current policy before replying.

Each of those questions has one correct answer, drawn from a policy or a schedule of charges or a financing structure, and a wrong answer on any of it, especially anything touching Shariah compliance or a financing calculation, is not a minor error. It is the kind of mistake that costs trust in a way a generic customer service slip does not. At the same time, a large share of the volume across all three channels was genuinely repetitive, consuming time from both customers waiting on a reply and agents who could have spent that time on the emails that actually needed judgement.

Why AutomatiCworX

The bank deployed AutomatiCworX’s engine as a question-answering layer, live on the website, inside WhatsApp banking, and as an assist tool for the service quality team working through email, not the full agentic platform. There is no action execution here, this system does not move money, block a card, or change an account. It answers, accurately, and knows when not to, regardless of which of the three channels the question arrives through.

Two things made that possible across all three. It runs entirely on-premises, inside the bank’s own environment, so no customer data or account information leaves the bank’s infrastructure to answer a question, whether that question comes from a website chat, a WhatsApp message, or an email in the service quality queue. And it runs on AutomatiCworX’s confidence and grounding system, so an answer is only given when the system is actually confident it is correct against the bank’s own policies, schedules, and financing terms. On anything it is not confident about, especially fine-grained Shariah compliance questions or edge cases in a financing calculation, it says so and hands off, rather than producing a fluent but wrong answer.

AutomatiCworX in action

A visitor on the bank’s website asks for the SWIFT code. The system answers directly and correctly, without the visitor needing to search or call.

A customer on WhatsApp banking asks how much their Ijarah instalment would be for a given vehicle price and tenure. The system applies the bank’s actual Ijarah structure and calculates it correctly, rather than approximating with a generic loan formula that would misrepresent a Shariah-compliant product.

A member of the service quality team opens one of thousands of daily emails, a customer asking about documentation required for a digital account aimed at customers living abroad. Instead of looking the policy up manually, the agent gets a grounded, correct answer from the system immediately and can reply to the customer in a fraction of the time it used to take, with the policy detail already right.

A customer asks a nuanced question about whether a specific financing arrangement is Shariah-compliant in their situation. Recognising this sits outside a straightforward factual answer, the system declines to guess and directs the customer to the appropriate channel instead.

Across all three surfaces, the system answers the full spread of what customers and staff actually ask about: everyday banking (debit card, mobile app, internet banking, WhatsApp banking, cheque books, bill payments, fund transfers, remittances), Islamic financing products (Car and Bike Ijarah, house financing, financing for customers living abroad), digital account services (account opening, accounts for non-resident customers, freelancer accounts), account and product policy (deposit accounts, profit rates, schedule of charges, term deposit certificates), and general service information (branch locations, account servicing, complaints intake, non-resident customer support).

The results

The bank now has a single, accurate, on-premises answer layer covering the full breadth of its retail and digital product set, live on its website and WhatsApp for customers, and used directly by the service quality team to work through its email volume faster and more consistently, without any customer data leaving its own infrastructure and without the system guessing on anything it is not grounded to answer.

What is next

The question-answering layer establishes the grounded, on-premises foundation the bank can extend toward action execution, verified transactions and account actions across the same website, WhatsApp, and email surfaces, once the bank is ready to take that step.

See what AutomatiCworX can resolve for your customers.

Explore more Case Studies