Case studies | Zentrova Technologies

Next-generation mobile banking for a top-10 joint-stock bank

gioithieucongty08.mauwebdep.com.vn — Sat, 24 Oct 2026 13:12:23 +0000

In 2021, one of Vietnam's ten largest joint-stock banks set itself a clear goal: move 70% of retail transactions to digital channels within three years. Its mobile app, first built in 2015, could no longer keep pace with the product roadmap the board expected. Zentrova was brought in to redesign both the customer experience and the platform behind it, from the login screen down to the integration layer with the core banking system.

The challenge

The legacy app was a monolith tightly coupled to the core, and every release needed six to eight weeks of testing. On payday, response times regularly exceeded eight seconds and failed transactions tripled. Opening an account online took 25 minutes, and almost half of applicants abandoned the process and went back to a branch. The bank also needed a platform its own engineers could run and extend once the programme ended, not another black box dependent on an outside vendor.

Our solution

Microservices architecture with more than 120 APIs, separating accounts, payments, cards and lending so that each team can ship on its own schedule.
Kafka event streaming as a buffer between the app and the core, absorbing traffic spikes without putting posting logic at risk.
eKYC with OCR and face matching, reading ID documents and verifying biometrics directly on the customer's device.
Layered security: device binding, end-to-end encryption, jailbreak detection and real-time transaction risk scoring, all aligned with PCI DSS.
Kubernetes deployment with fully automated CI/CD and a nightly regression suite of more than 3,000 scenarios.
Co-delivery model: 85 Zentrova engineers working alongside 30 bank specialists in two-week sprints, handing over operations to the in-house team step by step.

Results

The new app went live after nine months, and every customer was migrated within six weeks with no planned downtime.

Online account opening reduced from 25 minutes to 4 minutes.
Monthly active users reached 3.2 million, a 2.4x increase within 18 months.
Average response time below 400 ms, even on peak days.
App availability of 99.98%.
Release cycle shortened from six weeks to two weeks.
App store rating up from 2.9 to 4.7/5.

The bank's in-house team now owns day-to-day operations and ships new features independently, while Zentrova provides a small architecture and security squad under a support agreement. The same platform is being extended to small-business customers in the next phase.

What we value most is that Zentrova didn't just write software for us. They changed the way we build digital products. — Head of Digital Banking, client bank

Core banking migration to a hybrid cloud

gioithieucongty08.mauwebdep.com.vn — Sat, 24 Oct 2026 13:12:23 +0000

The primary data centre of a mid-sized joint-stock bank was approaching the end of its lease, and most of its servers were already out of warranty. Rather than fund another hardware refresh, the bank chose a hybrid cloud model: sensitive data stays in a certified private facility, while everything else runs on a public cloud region located in Vietnam. Zentrova owned the programme end to end, from the initial assessment to post-migration operations.

The challenge

The bank ran 46 systems holding roughly 180 TB of data, with many undocumented dependencies between applications. Provisioning a single test environment took three weeks on average, slowing every new initiative. Regulation required disaster recovery within four hours, yet the existing recovery plan had never been rehearsed in full. Any plan also had to keep branches, ATMs and digital channels running throughout, and leadership wanted infrastructure spending to move from large up-front purchases to a predictable, usage-based model.

Our solution

Application assessment sorting every workload into one of six paths: retain, rehost, replatform, re-architect, replace or retire.
Standardised landing zone built with Terraform, with network segmentation by data sensitivity and centralised key management.
Continuous data replication between old and new environments, allowing a full rollback within 30 minutes if anything went wrong.
28 applications containerised and run on Kubernetes with load-based autoscaling.
Five migration waves, each executed over a weekend and rehearsed twice beforehand.
Compliance controls mapped to ISO 27001 and PCI DSS, with centralised access logs ready for audit at any time.

Results

Within 14 months all 46 systems were running on the new platform, and the bank's 1.8 million customers did not lose a single business day of service.

Annual infrastructure cost reduced by 32%.
Test environment provisioning cut from three weeks to 45 minutes.
Core system availability of 99.98% in the first 12 months.
Disaster recovery measured in rehearsal at 1 hour 50 minutes.
End-of-day batch processing reduced from five hours to 2 hours 10 minutes.

Beyond the numbers, the migration changed how the technology team works. Infrastructure is now defined as code and reviewed like any other change, new environments are requested through a self-service portal, and the bank's own engineers run the platform day to day, with Zentrova acting as a second-line support partner.

For the first time, we ran a full disaster recovery drill without holding our breath for the outcome. — Chief Technology Officer, client bank

Omnichannel e-commerce platform for an electronics retailer

gioithieucongty08.mauwebdep.com.vn — Sat, 24 Oct 2026 13:12:23 +0000

A consumer electronics retailer with 140 stores across Vietnam saw that most customers researched products online before visiting a store, yet its website and point-of-sale systems behaved like two separate businesses. Prices, stock levels and promotions often differed from one channel to another, eroding customer trust and costing store staff time in explanations.

The challenge

The old website slowed down every sale season and once had to suspend orders on the busiest day of the year. Inventory synced only once a night, so around 9% of online orders were cancelled because the item was out of stock. The marketing team needed three to four days to push a new promotion to every channel. Leadership wanted one platform where a product, its price and its availability were defined once and shown consistently everywhere.

Our solution

Headless commerce architecture: a single business layer serving the website, the app and in-store screens, with a React front end rendered on the server for speed.
Real-time inventory aggregated from 140 stores and three central warehouses with under five seconds of latency, enabling buy online, pick up in store.
Elasticsearch-based search that understands Vietnamese typed without diacritics, typos and the everyday names customers use for products.
Central promotion engine that lets marketers configure campaigns themselves and publish them to every channel at once.
Load testing before each peak season at five times real traffic, combined with autoscaling on cloud infrastructure.

Results

The platform went live after eight months and has since handled four peak shopping seasons without a major incident.

Online revenue up 68% in the first year.
Out-of-stock cancellations down from 9% to 1.2%.
Average mobile page load time of 1.6 seconds.
31% of online orders now collected in store, driving extra add-on sales at the counter.
Promotions published across all channels in under an hour.
Loyalty programme grown to 5.6 million members.

Because the business layer is shared, new channels are now far cheaper to add. The retailer has since launched in-store kiosks and a business-to-business ordering portal on the same platform, each in under eight weeks, and its in-house team handles most front-end changes without outside help. Zentrova still runs performance engineering ahead of every major sale.

Our customers now see the same price and the same stock figure whether they are on the app or standing in one of our stores. — Head of E-commerce, client company

Retail analytics platform for a Japanese supermarket chain

gioithieucongty08.mauwebdep.com.vn — Sat, 24 Oct 2026 13:12:23 +0000

A Japanese supermarket chain with 420 stores generated enormous volumes of sales data, yet most replenishment decisions still depended on each store manager's experience. Reports were compiled by hand in spreadsheets and typically reached the executive team a week late. The company partnered with Zentrova's Tokyo centre to turn its data into a tool for everyday decisions.

The challenge

Data lived in 38 separate sources: point of sale, loyalty cards, suppliers, logistics and even weather feeds. Fresh produce accounted for 35% of revenue but suffered high waste, while core items regularly ran out on weekend shelves. Every department had its own definition of metrics such as gross margin or sales per square metre, so meetings started with arguments about the numbers. Head office wanted a single version of the truth, delivered early enough each morning to change what was ordered that day.

Our solution

Data lakehouse architecture in the cloud, ingesting all 38 sources at intervals ranging from 15 minutes to daily.
Shared semantic layer built with dbt, standardising more than 160 business metrics with automated data quality tests.
Demand forecasting models in Python for every item in every store, accounting for seasonality, promotions and weather.
Role-based dashboards for executives, regional managers and store managers, designed to work well on tablets.
Access control by store and region, with loyalty member data encrypted in line with personal data protection rules.
Phased rollout: a six-week pilot in 20 stores, measured carefully before extending to the whole chain.

Results

The platform was in use across the entire chain after seven months, with 2,300 internal users logging in every week.

Average inventory value reduced by 18%.
Fresh produce waste down 27%.
Shelf out-of-stocks on core items down 34%.
Executive reports ready at 7 a.m. every day instead of a week late.
Store-level demand forecast accuracy of 91%.

The forecasting models are retrained every week and their accuracy is tracked openly on the same dashboards, so store managers can see when to trust a suggestion and when to override it. The client's analysts now build new reports themselves on the shared semantic layer, and the next phase extends forecasting to supplier ordering and promotional planning, with Zentrova's Tokyo team on hand for model tuning.

Our store managers now start the day with data instead of guesswork. — Chief Operating Officer, client supermarket chain

Smart factory IoT and MES for an electronics component maker

gioithieucongty08.mauwebdep.com.vn — Sat, 24 Oct 2026 13:12:23 +0000

A foreign-invested electronics component manufacturer runs three plants in northern Vietnam, supplying global device makers that demand lot-level traceability for every shipment. When order volumes nearly doubled in two years, paper travellers and spreadsheets could no longer keep production under control. Zentrova was engaged to build a smart factory platform that connects machines to a manufacturing execution system and to the company's existing ERP platform.

The challenge

The plants operated more than 1,400 machines spanning several generations and suppliers, most without any data interface. Downtime was recorded only when a shift leader remembered to write it down, so efficiency figures were unreliable at best. Each time a customer asked to trace a defective lot, the quality team spent two to three days digging through records. Management needed accurate, real-time visibility without halting production to replace equipment.

Our solution

Machine connectivity layer using industrial IoT gateways and MQTT, retrofitting sensors on older equipment instead of replacing it.
Edge computing on every shop floor, processing data locally so that each plant keeps running if the connection to the central platform drops.
Manufacturing execution system covering work orders, routing, materials and quality, with two-way synchronisation to the ERP platform.
Lot-level traceability from incoming raw materials to finished goods, searchable in seconds.
Line-side tablets replacing paper: operators receive work instructions, report defects and confirm output right at their station.
Segmented industrial network separating operational technology from the office network, with access controls aligned to ISO 27001.

Results

After 11 months, all three plants were running on the new platform, used every day by 4,500 operators and engineers.

Overall equipment effectiveness (OEE) up from 64% to 81%.
Lot traceability reduced from three days to under one minute.
Defects caught at final inspection down 38%.
Printed production paperwork reduced by 95%.
Product changeover time on the lines shortened by 27%.

The rollout followed a deliberate sequence: one pilot line, then one full plant, then the remaining two. Each step was signed off against measured OEE before the next began, and plant engineers were trained to add new machines to the platform themselves. The same data now feeds Zentrova's predictive maintenance models, currently being trialled on the plants' most critical equipment.

We can see the status of every machine in real time and, more importantly, we know why it stopped. — Plant Director, client company

AI-driven predictive maintenance for a steel mill

gioithieucongty08.mauwebdep.com.vn — Sat, 24 Oct 2026 13:12:23 +0000

In a steel mill, a single hour of unplanned downtime on the rolling line can cost hundreds of millions of dong, not to mention the safety risks of emergency repairs. A domestic steel producer wanted to move away from fixed-interval maintenance and run-to-failure repairs towards maintenance based on the actual condition of each asset. Zentrova is delivering the programme in two phases; phase one covers the 620 most critical assets in the rolling and raw materials areas.

The challenge

Equipment runs around the clock amid extreme heat, dust and heavy vibration. Fixed maintenance schedules meant many healthy components were replaced too early, while bearing and gearbox failures still struck without warning. Diagnostic know-how rested with a handful of senior engineers who were approaching retirement, and the company had no structured way to capture what they knew. Every month without change meant more lost production and more overtime spent on emergency repairs.

Our solution

Industrial sensor network measuring vibration, temperature and motor current on motors, mills and conveyors, built to survive harsh conditions.
Time-series data platform capturing billions of data points a day, with pre-processing at the edge to reduce network load.
Machine learning models in Python that detect anomalies and estimate the remaining useful life of each asset, enriched with historical repair records.
Expert knowledge captured digitally: diagnostic rules from senior engineers are encoded in the system and used to validate model output.
Alerts built into the maintenance workflow, automatically raising work orders with the predicted cause and the parts to prepare.
Wave-based validation: each equipment group runs in parallel with the old process for three months before it goes live.

Results

Rolling line 2, the first pilot area, has delivered measurable results after six months of operation:

Unplanned downtime reduced by 41%.
Failures flagged on average nine days in advance.
Spare parts spending down 17% by avoiding premature replacement.
Alert precision of 87%, with false alarms below 8%.

Phase two will extend coverage to the reheating furnaces and the compressed air system. Over the same period, responsibility for retraining and monitoring the models will move step by step to the client's own data team, supported by playbooks and joint review sessions with Zentrova engineers.

Our maintenance crews no longer chase breakdowns. They plan replacements for the scheduled stops instead. — Head of Mechanical and Electrical Engineering, client plant

Unified hospital information system for a four-hospital network

gioithieucongty08.mauwebdep.com.vn — Sat, 24 Oct 2026 13:12:23 +0000

A private hospital group with four general hospitals and 1,100 beds had grown quickly through acquisitions. Each hospital still ran its own management software, so the records of a single patient could sit in four different places. Leadership set two goals: a complete electronic medical record, and one shared platform for the whole group.

The challenge

Patients had to repeat their details every time they visited another site. Lab results and imaging reports moved between departments on paper or through chat messages, delaying treatment decisions. During the morning rush, the wait from registration to seeing a doctor sometimes exceeded 90 minutes. Billing teams also struggled with health insurance claims, which were frequently rejected because of incomplete or inconsistent data. Whatever the solution, it had to be rolled out without interrupting round-the-clock clinical care.

Our solution

Unified HIS and EMR platform on a microservices architecture, with a single patient identifier shared by all four hospitals.
HL7 FHIR integration with the laboratory system (LIS), imaging archive (PACS) and the national health insurance claims gateway.
Tablet app for physicians to review history, prescribe and place orders at the bedside.
Smart scheduling and queuing that notifies patients when their turn is near, so they no longer wait in the lobby.
Healthcare-grade data protection: role-based clinical access, a full audit trail of every record viewed, and encryption at rest and in transit.
Hospital-by-hospital cutover, with an on-ward support team at each site for the first two weeks of operation.

Results

After 12 months, all four hospitals were live on the new platform, used every day by 3,800 clinical and administrative staff.

Average outpatient waiting time reduced by 45%.
Lab results available in the medical record 60 minutes sooner.
Health insurance claim rejections down from 4.1% to 0.9%.
System availability of 99.97%.
100% of inpatient records digital, with paper charts fully retired.

Clinicians were involved from the very first design workshop, and more than 120 doctors and nurses served as super-users who trained their colleagues. That early ownership is a large part of why adoption reached every ward within weeks rather than months. The group is now using the same platform to open a fifth hospital and a network of satellite clinics.

Our doctors now have a patient's complete history at hand, wherever that patient was treated before. — Chief Medical Officer, client hospital group

Telemedicine app for a multi-specialty clinic chain

gioithieucongty08.mauwebdep.com.vn — Sat, 24 Oct 2026 13:12:23 +0000

A chain of 26 multi-specialty clinics in Vietnam's major cities wanted to reach patients in the provinces, as well as people who needed regular follow-ups but struggled to find time for a clinic visit. It needed a telemedicine app reliable enough for doctors to diagnose with confidence, yet simple enough for elderly patients to use without help.

The challenge

Before the project, remote advice happened over the phone and personal messaging apps, with no records kept and little control over quality. Appointments were managed separately at each clinic, and no-show rates reached 22%. Prescriptions still had to be printed, so patients travelled to a clinic just to collect a slip of paper. Doctors were also wary of any tool that might drop calls mid-consultation or expose patient data, so trust had to be earned from the first release.

Our solution

Flutter mobile app for both major mobile platforms, with large type and a pared-back flow designed for older users and tested with patients over 65 before launch.
WebRTC video consultations that adapt quality to the connection and stay stable even on weak 4G networks.
Central booking across all 26 clinics, with automatic reminders by push notification and text message.
Electronic prescriptions digitally signed by the doctor and routed to partner pharmacies for home delivery.
Personal health record synchronised via HL7 FHIR with the clinic management system, so doctors can see results from earlier visits.
End-to-end encryption for calls and data; video is never recorded without the patient's explicit consent.

Results

The app launched after six months and has become the chain's second-largest patient channel, after walk-in visits.

More than 1.1 million remote consultations completed.
640,000 registered patients, 38% of them outside the cities where the chain has clinics.
No-show rate down from 22% to 7%.
Post-consultation satisfaction score of 4.8/5.
96% of video calls completed without interruption.

Zentrova continues to release updates every month based on feedback from doctors and patients. Recent additions include family accounts, so adult children can book and join consultations for their parents, and structured follow-up plans for chronic conditions such as diabetes and hypertension, which now account for a third of all remote visits.

Patients on regular follow-ups no longer have to take half a day off work for a fifteen-minute appointment. — Medical Director, client clinic chain

Real-time shipment tracking platform for a 3PL provider

gioithieucongty08.mauwebdep.com.vn — Sat, 24 Oct 2026 13:12:23 +0000

A third-party logistics (3PL) provider operating in four Southeast Asian countries manages 32 warehouses and a fleet of 2,800 vehicles, serving shippers in consumer goods and electronics. Its customers increasingly want to know exactly where their goods are at any moment, not just receive an end-of-day report. Zentrova is rolling out a real-time tracking platform in stages, and the first 18 warehouses are already live.

The challenge

Each country ran a different warehouse management system, and transport data sat with individual subcontractors. Order status typically lagged reality by six to twelve hours. Monthly manual stock counts disrupted warehouse operations, yet inventory discrepancies still stood at 2.5%, regularly leading to disputes with shippers. Customer service teams spent most of their day answering calls that simply asked where a shipment was. Replacing every system at once was out of the question, given the cost and the risk to daily operations.

Our solution

Central Kafka event stream capturing every scan, receipt, dispatch and GPS ping, processing more than 40 million events a day.
Handheld scanner app for warehouse staff covering receiving, put-away, picking and cycle counting, with full offline support.
Driver app with optimised routes and proof of delivery by photo and e-signature.
Shipper tracking portal plus open APIs so each customer can plug the data into its own systems.
Standardised integration layer connecting existing warehouse systems rather than forcing a big-bang replacement.
Cluster-by-cluster rollout, with on-site support during the first week and training delivered shift by shift.

Results

Across the 18 warehouses running on the new platform, the client measured the following after six months:

Inventory accuracy of 99.7%.
Order status latency cut from six to twelve hours to under one minute.
Picking productivity up 23%.
On-time delivery up from 91% to 97.5%.
Where-is-my-order calls down 60%.

The remaining 14 warehouses and the full vehicle fleet are scheduled to join the platform during 2026. Because data now flows through a single event stream, the client is also preparing new services on top of it, including dynamic delivery slots and carbon reporting per shipment for its larger customers. Zentrova's team in Ho Chi Minh City supports the platform around the clock throughout the rollout.

Shippers no longer call to ask where their goods are, because they can see it on screen themselves. — Regional Operations Director, client company

Provincial online public service portal

gioithieucongty08.mauwebdep.com.vn — Sat, 24 Oct 2026 13:12:23 +0000

The government of a coastal province with 1.4 million residents set out to handle most administrative procedures online, from business registration to construction permits. Its legacy system supported only around 200 procedures, worked poorly on mobile phones and could not exchange data with the software used by individual departments. Zentrova is now delivering a new public service portal together with the case-processing platform behind it.

The challenge

Residents often visited the one-stop service centre two or three times for a single application: once to ask, again to submit missing documents, and finally to collect the result. Officials re-keyed information from paper forms, which led to errors and made reporting difficult. Because each department used its own software, provincial leaders had no consolidated view of how applications were progressing. Many residents in fishing and farming communities also worked irregular hours, making office visits especially costly for them.

Our solution

Unified public service portal on web and mobile, designed around residents' needs, with forms pre-filled from data the government already holds.
Electronic identity verification (eKYC) and OCR for documents, reducing data entry for residents and officials alike.
Microservices case-processing platform where approval workflows are configured per procedure, without rewriting software.
Data integration hub connecting departmental systems and shared government databases using national data exchange standards.
Leadership dashboard tracking overdue cases, satisfaction scores and workload by department.
Security by information system classification, with independent penetration testing before every expansion.

Results

Phase one, covering the 480 most frequently used procedures, is live at provincial level and in nine districts and towns:

72% of phase-one applications are now filed online.
Average processing time reduced by 35%.
Overdue cases down from 8.4% to 1.6%.
Resident satisfaction at 93%.

Phase two will bring all 1,250 procedures onto the portal and extend it to commune level ahead of formal acceptance in 2027. Alongside the technology, Zentrova is training more than 1,500 officials and supporting community help desks so that older residents are not left behind as services move online. Residents can also follow every application by text message, and phase-two procedures are being redesigned using feedback from more than 3,000 users collected during phase one.

People can now complete their paperwork in the evening, on their phones, without taking time off work. — Senior official, Office of the Provincial People's Committee

24/7 security operations centre for an insurance group

gioithieucongty08.mauwebdep.com.vn — Sat, 24 Oct 2026 13:12:23 +0000

An insurance and financial services group with 22,000 employees and agents holds sensitive data on millions of customers, from health information to payment accounts. As ransomware attacks on the region's financial sector increased, the board asked for continuous, measurable monitoring and response. Zentrova was selected to build and run a security operations centre (SOC) as a managed service.

The challenge

The in-house security team had just six people working office hours, juggling alerts from more than ten disconnected tools. Thousands of alerts arrived every day, most of them false positives, so genuinely dangerous signals were easy to miss. The average time from detection to resolution of an incident was around six hours. Regulators and reinsurance partners were also asking for clearer evidence that risks were being monitored and managed around the clock, not just during business hours.

Our solution

Centralised SIEM platform collecting logs from 18,000 endpoints, core insurance systems, agent applications and cloud infrastructure.
EDR on every workstation and server, able to isolate an infected device within seconds.
SOAR-driven response automation with more than 60 playbooks, from locking compromised accounts to blocking phishing domains.
Threat intelligence continuously updated with indicators of attacks targeting financial institutions in Southeast Asia.
Three-tier analyst team on duty 24/7 in Hanoi and Da Nang, with response times committed by severity level.
Regular attack simulations and quarterly ISO 27001 compliance reporting to the executive committee.

Results

The SOC went live after eight months and has operated without interruption since 2023.

Mean incident response time cut from six hours to 38 minutes.
Mean time to detect now under 10 minutes.
Alerts requiring human review down 85% through tuning and automation.
Zero customer data breaches since go-live.
Staff click rate on simulated phishing emails down from 14% to 3%.

The client's own security team has shifted from firefighting to governance, architecture reviews and risk planning, while Zentrova handles day-to-day monitoring and response. Every month both teams review the incidents handled, the playbooks triggered and the detection rules added, so the service keeps improving as the threat landscape changes. Detection coverage is mapped against well-known attack techniques and reported to the board every quarter.

We know someone is watching our systems at three in the morning, and we know exactly what they will do if something happens. — Chief Information Security Officer, client group

AI customer-service assistant for an express delivery company

gioithieucongty08.mauwebdep.com.vn — Sat, 24 Oct 2026 13:12:23 +0000

An express delivery company handling more than 900,000 parcels a day receives around 45,000 support requests daily through its contact centre, chat and social media. Most questions are repetitive: where is my parcel, when will it arrive, can I change the address. The company wanted to test an AI assistant that could resolve these requests on its own, freeing agents to focus on complex cases. Zentrova is now running the pilot with 200,000 customers.

The challenge

During peak shopping seasons, contact centre waiting times sometimes exceeded 15 minutes and up to 30% of callers hung up. The previous chatbot relied on rigid scripts that recognised only a few dozen phrasings and constantly handed customers over to agents. Any new solution had to guarantee that it would never give wrong answers about fees, compensation policy or a recipient's personal details. It also had to cope with the informal, abbreviated Vietnamese customers actually type, including regional expressions and missing diacritics.

Our solution

Large language model with retrieval-augmented generation (RAG): the assistant answers only from approved policies, price lists and procedures, citing its internal sources.
Direct integration with operational systems via APIs to track shipments, reschedule deliveries and open claims.
Identity verification before disclosure, with personal data masked in conversation logs.
Safety guardrails that detect out-of-scope questions, frustrated customers or compensation requests and hand them straight to an agent, together with a conversation summary.
Automated evaluation suite of more than 4,000 real customer questions, re-run whenever the model or documentation changes.
Controlled pilot on web chat and the mobile app, measured against a control group of customers.

Results

After three months of the pilot, early indicators show:

58% of requests fully resolved by the assistant.
First response time under five seconds.
Answer accuracy of 96% in sampled quality reviews.
Post-conversation satisfaction of 4.5/5.
Agent handling time for escalated cases down 40%, thanks to ready-made summaries.

The next step is to bring the assistant to the voice channel of the contact centre and open it to all customers. Before that, the team will extend the evaluation suite to voice transcripts and agree clear service thresholds with the client that must be met before each expansion.

We want AI to do the repetitive work well, and people to do the work that needs empathy. This pilot shows that balance is achievable. — Head of Customer Experience, client company