Ruparupa and Kawan Lama Group: weekly sales reports became hourly once online and store data shared one data lake
Solution case studies are scenarios based on common industry problem patterns. They do not refer to any client and are not MarkasDev project portfolio items. Our client portfolio is published separately.
Industry context
Kawan Lama Group was founded in 1955 and, according to SAP (29 July 2025), now runs more than 1,200 stores and more than 30 brands, including AZKO, INFORMA, Toys Kingdom, Pet Kingdom, Krisbow and Chatime. Fortune Indonesia (1 August 2025) adds that the network is served by 14 distribution centres. The group's Google Workspace customer page (undated, opened 16 October 2026) says it employs more than 35,000 people across six business pillars, with retail as its largest line of business.
Inside the group sits Ruparupa, Kawan Lama Group's first digital platform for household, furniture and lifestyle needs, built by PT Omni Digitama Internusa. In August 2021, Ruparupa CTO Ronny Winoto said the platform carried more than 67,000 product variants and used an omnichannel model: customers could buy fully online, order online and pick up at group stores such as Ace Hardware and Informa, or see an item in store and finish the purchase online (Gizmologi, 6 August 2021).
What makes the case worth studying is the order of the steps. Ruparupa first moved to the cloud to absorb pandemic traffic spikes, then built a data lake so online and store sales could be read from one place every hour. Four years later, the parent group announced a cloud ERP to give staff real-time stock visibility across outlets.
A note on how this page was put together: it draws on openly published material, namely an AWS technical blog co-written with Ruparupa's data team, an SAP announcement and media coverage. Kawan Lama Group and Ruparupa are not MarkasDev clients, and we were not involved in any of the projects described. We read the story through Diagnose, Deliver and Operate because the problem is common in Indonesian retail: sales data exists, but it arrives late, sits in many places and is not trusted. Every number on this page comes with its source and date.
Problem
The AWS Business Intelligence Blog post co-written with Olivia Michele and Dariswan Janweri P. of Ruparupa (22 February 2023) is fairly candid about the starting point. Ruparupa generated data from its e-commerce site, inventory systems, and distribution and finance applications, but had no infrastructure able to process that volume and variety of formats quickly.
As a result, data analysts processed data by hand. Ingestion alone took 1 to 1.5 hours for hundreds of thousands of rows. Cleansing was inconsistent: after data had been cleaned, steps were often found missing and everything had to be cleaned again. The analytics team could only produce weekly and monthly reports, and some important metrics only appeared in the monthly ones.
Reach was narrow too. According to the same post, these routine reports were read by at most 20 people in management. Other business units in Kawan Lama Group only received manually prepared weekly reports, and business users had no visibility of sales as they happened in stores or on the website.
The pressure came from the pandemic. At a virtual press event on 5 August 2021, Ronny Winoto said transactions for health products such as oximeters, oxygen concentrators, thermometers and masks rose 115 percent compared with before the pandemic, especially during Indonesia's emergency mobility restrictions (Media Indonesia, 6 August 2021). Spikes like these arrive without warning, so the infrastructure had to adjust capacity without letting costs balloon.
At group level the problem was similar but larger. SAP (29 July 2025) notes that Kawan Lama's legacy ERP had been in use for more than a decade, that opening a single new outlet often required new system configuration and long manual processes, and that customers were sometimes told to wait days because staff could not see stock in the system, even when the items were available.
Constraints
- Data came from many systems in different formats: structured transactions, inventory, distribution and finance data, plus semi-structured or unstructured customer interaction data (AWS, February 2023)
- Pandemic traffic spikes were unpredictable, with health product transactions up 115 percent compared with before the pandemic (Media Indonesia, August 2021)
- Costs had to stay under control, so the data platform was designed to scale to terabytes in a cost-effective way (AWS, February 2023)
- Business users had little trust in the numbers because manual reports often had to be cleaned again and the data lagged by a day to a week
- Online and offline had to be comparable, because Kawan Lama Group also runs physical stores outside the Ruparupa platform
- The group is large and varied: more than 1,200 stores, more than 30 brands and 14 distribution centres (SAP and Fortune Indonesia, 2025), running on a legacy ERP more than 10 years old
- The data team started very small, with a single analyst, according to the AWS post
Approach
Diagnose
This approach section is our reading of openly published documents. MarkasDev did not take part in the project.
The standout feature of Ruparupa's story is that it tested demand before building a large platform. From April 2021, the team used AWS Database Migration Service (DMS) to stream change data capture (CDC) updates into an Amazon Aurora MySQL database, then built an Amazon QuickSight dashboard to replace a static slide deck. The AWS post says the pilot was received extremely well because users could finally see current data.
The pilot also exposed its limits. It covered a single data domain, data beyond transactions had not been brought together, and the data in that first warehouse still lagged by one day. The team called it a temporary solution. In other words, the pilot worked as a diagnostic tool: it proved that people would use a dashboard and surfaced the real requirements, namely many data sources, hourly freshness and costs that stay reasonable as volumes grow.
For other retailers, the equivalent questions are: which decisions are late today because of the weekly report, who besides the board needs those numbers, and how fresh does the data really need to be. That is the mapping we do in Diagnose before choosing tools or vendors.
Deliver
This is also our reading of openly published documents.
The architecture described in the AWS post (February 2023) is simple enough to copy. AWS DMS reads the CDC log of the source databases, covering every insert, update and delete, and stores it in a raw layer on Amazon S3. An S3 lifecycle policy moves older data to Amazon S3 Glacier. An AWS Glue PySpark job with the Apache Hudi connector then merges those changes into a transformed layer in compressed, partitioned Parquet, every hour, using Glue's built-in scheduler.
Two technical choices stand out. First, Apache Hudi lets the data lake accept upserts and deletes, so the data in S3 stays consistent with the source systems. Second, the team did not write one job per table. They built a single generic job that takes the table name as a parameter, while each table's structure is stored in AWS Secrets Manager. That pattern lets hundreds of tables be handled without hundreds of jobs.
Metadata for every table lives in the AWS Glue Data Catalog, shared by the Glue jobs, the Amazon Athena query engine and QuickSight. The pilot dashboard was repointed from Aurora to Athena, and new dashboards were built as business units asked for them. QuickSight also sends email alerts when a value crosses a threshold.
The cloud foundation came first. Ronny Winoto said Ruparupa left its on-premises infrastructure behind, adopted a microservices architecture so a fault in one module would not bring down the others, used AWS Auto Scaling to adjust capacity within minutes and Amazon CloudWatch to spot anomalies (Gizmologi and Media Indonesia, August 2021). An instant delivery feature was reportedly built in under two months.
For other companies, the heart of this work is system integration & APIs: pulling data from many applications without disturbing transactional systems, then presenting it in a form people trust. We deliver that by milestone through Deliver.
Operate
This is also our reading of openly published documents.
Ruparupa's data lake was not declared ready straight away. The AWS post records that implementation took around 3 months, plus 1 more month just for data validation before production use. That validation month is often skipped in similar projects, yet it is where user trust is earned.
Once it was running, the team grew to match. The analytics team went from one person to seven to handle new use cases: merchandising, operations, store manager performance and trending SKUs. The number of dashboards grew to around 70 since the April 2021 pilot, all driven by requests from business users. Their content widened from transactions to other metrics such as new SKUs, operational escalations for free customer services and SLA monitoring.
Infrastructure cost was managed through long-term usage commitments. Ronny Winoto estimated that AWS Reserved Instances and Compute Savings Plans cut long-term infrastructure costs by 72 percent and 50 percent (Gizmologi, 6 August 2021). That is a CTO's estimate at a media event, not an audited figure, and the source does not spell out the baseline.
For other organisations, Operate here means watching the hourly data jobs, keeping data quality up, taking requests for new dashboards and reviewing cloud cost commitments regularly. That is the scope of Operate and server maintenance.
High-level architecture
This flow is our simplification of the AWS Business Intelligence Blog post of 22 February 2023, not an audit of Ruparupa's production system. The post does not give total data volume, monthly cost or sales uplift in numbers. We found no open source explaining whether this architecture is still in use in 2026 or how it relates to the RISE with SAP ERP the group announced in July 2025. No AI or machine learning component is central to the solution in the sources we opened, so the AI field is set to false.
Typical outcomes
The figures below are quoted from sources we opened, with their publication dates. Most of the data lake results come from a single post written jointly by AWS and the Ruparupa team, so they should be read as company claims rather than an independent evaluation.
Data freshness. Management can now see the current state up to the last hour, where previously they only had weekly reports. The AWS post (22 February 2023) calls this 168 times faster, a figure that is essentially one week (168 hours) divided by one hour.
Reach. Dashboard users grew from 20 people in management to around 180 users, a ninefold increase, including other business units in Kawan Lama Group (AWS, February 2023).
Number and scope of dashboards. Around 70 dashboards have been built since the QuickSight pilot in April 2021, covering transactions, new SKUs, operational escalations and SLA monitoring (AWS, February 2023).
Team and timeline. The analytics team grew from 1 to 7 people. The data lake took around 3 months to implement plus 1 month of data validation (AWS, February 2023).
Trust and sales. Other business units are said to trust the data more because it is fresher and can be drilled down to SKU level. Online and store sales are now easier to compare because they sit on the same platform. The post says the implementation helped increase sales revenue in several business units, but gives no figure.
Cloud foundation (2021). Estimated savings of 72 percent and 50 percent on long-term infrastructure costs through Reserved Instances and Compute Savings Plans, an instant delivery feature built in under two months, and delivery coverage of 600 points in 69 cities (Gizmologi, 6 August 2021).
Group step (2025). SAP (29 July 2025) reported that the RISE with SAP implementation is already helping teams see real-time inventory across outlets, so staff can serve customers faster even when an item is not on the shelf. Next on the list are better forecasting, fewer stock-outs and smarter promotions. The release gives no outcome figures.
Figures at a glance:
- Data freshness: weekly to hourly (AWS, 22 February 2023)
- Dashboard users: 20 to about 180 (AWS, 22 February 2023)
- Dashboards: about 70 since April 2021 (AWS, 22 February 2023)
- Analytics team: 1 to 7 people (AWS, 22 February 2023)
- Ingestion time before the data lake: 1 to 1.5 hours for hundreds of thousands of rows (AWS, 22 February 2023)
- Long-term infrastructure cost savings: 72 percent and 50 percent, CTO estimate (Gizmologi and Media Indonesia, 6 August 2021)
- Group scale: more than 1,200 stores, more than 30 brands (SAP, 29 July 2025), 14 distribution centres (Fortune Indonesia, 1 August 2025)
Why this case matters for retailers in Indonesia
Many Indonesian retailers have complete sales data in their point-of-sale systems, marketplaces and ERP, yet daily decisions still wait for a weekly recap built in a spreadsheet. Ruparupa's case shows the way out does not have to start with a big project. One pilot dashboard, one month of validation and reusable data jobs were enough to change who reads the data and how quickly they act.
The case also shows where analytics stops. Hourly dashboards help management read trends, but a customer waiting to hear about stock needs an operational system that unifies inventory across outlets. Kawan Lama Group tackled those two layers with separate projects, and other businesses would do well to plan them separately too.
If your sales reports are still weekly, or online and store data cannot yet be compared, start with Diagnose to map your data sources and the decisions they delay, move on to system integration & APIs through Deliver, then keep data quality on track with Operate. More case studies in this sector are on the Distribution & Retail page.
Sources
- AWS Business Intelligence Blog, 22 February 2023. How Ruparupa gained updated insights with an Amazon S3 data lake, AWS Glue, Apache Hudi, and Amazon QuickSight (co-written with Olivia Michele and Dariswan Janweri P., Ruparupa): https://aws.amazon.com/blogs/business-intelligence/how-ruparupa-gained-updated-insights-with-an-amazon-s3-data-lake-aws-glue-apache-hudi-and-amazon-quicksight/
- Gizmologi.id, 6 August 2021. Layanan Cloud AWS Bantu Ruparupa.com Lancarkan Strategi Omnichannel di 69 Kota (in Indonesian): https://gizmologi.id/news/cloud-aws-omnichannel-ruparupa/
- Media Indonesia, 6 August 2021. Cloud Solusi Hadapi Disrupsi Digital dan Pandemi (in Indonesian): https://mediaindonesia.com/teknologi/423711/cloud-solusi-hadapi-disrupsi-digital-dan-pandemi
- SAP Southeast Asia News Center, 29 July 2025. Kawan Lama Group Strengthens Operational Efficiency and Scalability with RISE with SAP: https://news.sap.com/sea/2025/07/kawan-lama-group-strengthens-operational-efficiency-and-scalability-with-rise-with-sap/
- Fortune Indonesia, 1 August 2025. Resep Kawan Lama Group Modernisasi Operasional dan Manufaktur (in Indonesian): https://www.fortuneidn.com/business/resep-kawan-lama-group-modernisasi-operasional-dan-manufaktur-00-3dhnm-phhjt8
- Google Workspace, Kawan Lama Group customer page (undated): https://workspace.google.com/customers/kawanlama/
Accessed 16 October 2026 (WIB).