Table of Contents
Background
“We’re at their mercy almost,” the client said about its own technology. “If we had control over some of those things, that would be great.”
This is the central challenge that drives every retail system consolidation engagement we take on.
Roughly twenty people run this premium consumer products brand. They sell direct to consumer, across a dozen-odd international marketplaces in several currencies, through a handful of regional storefronts, and to B2B buyers over EDI. It is a real business at real scale, and it was being operated out of spreadsheets.
Not by choice. More than ten systems held pieces of the picture, and none of them talked to the others. The ERP knew what had been invoiced. The marketplace connector knew most of what had been sold. Nobody owned the gap between the two, so every month someone in finance closed the gap by hand, reconciling hundreds of thousands of transactions across markets and currencies, and every month the answer arrived too late to act on.
Key Outcomes at a Glance
- $10,000 a year in connector fees retired
- $25,000 to $30,000 in replacement costs avoided, plus 150 to 200 hours of migration work never spent
- $458 a month in infrastructure, $300 to $350 a month in AI usage, with no per-user licensing anywhere in the stack
- 10-plus fragmented systems consolidated into one warehouse, one CRM and one orchestrator
- 12 of 49 discovery gaps closed the moment the warehouse existed, before any AI was pointed at anything
Challenge
The constraints were structural, and each one blocked the path to retail system consolidation.
The Data Was Fragmented by Design
The connector between the marketplaces and the ERP delivered about 80 percent of transaction data. The missing 20 percent was not noise, it was the money: international taxes, storage fees, selling fees by country, refund administration, promotional rebates, inventory credits. Worse, every direct-to-consumer order landed against a single generic customer record, which meant the company could see what it sold but never to whom. Per-customer economics simply did not exist.
An Expensive Stack, Coming and Going
The connector cost about $10,000 a year to deliver its 80 percent. The iPaaS linking the third-party logistics provider was scheduled for end of life, and the replacement path quoted a new-provider charge of $25,000 to $30,000 plus 150 to 200 hours of migration work. The EDI provider carrying about half of B2B orders kept its mapping on its own infrastructure, out of reach. NetSuite, the ERP, sat on a premium tier while its CRM went unused and reporting happened in spreadsheets. Every one of those was a bill for something the company could not change.
The Last Attempt Had Already Failed
A CRM roll-out the sales team never adopted, because it asked them for more administration rather than less. That failure set the real bar: whatever replaced it had to survive contact with people who had already rejected one.
A discovery pass across the business found 49 gaps between where the company was and where it needed to be. Twelve of them turned out to be the same gap wearing different clothes: there was nowhere for the data to go.
The complexity of the old stack was hidden in the “spaghetti” of manual reconciliations and fragmented data feeds. The new architecture replaces that disorder with a central, sovereign hub.
The following diagram illustrates the shift from a siloed, vendor-dependent ecosystem to a unified architecture where all components, from the data warehouse to the CRM and AI layer, reside entirely within the client’s own AWS environment.
Solution
So that was built first, and it was built inside the client’s own AWS account.
The Data Foundation
The foundation is a PostgreSQL warehouse on RDS, layered from raw ingestion through to reporting, with the marketplaces and storefronts feeding it directly. The database sits in private subnets, reachable only from the application tier. The monthly AWS bill goes to the client, not through HyScaler.
CRM, Orchestration, and AI on Top
Odoo serves as the CRM, synced bidirectionally with NetSuite so the ledger stays the ledger and the sales team gets a tool that fills itself in rather than one more place to type. Kestra runs the workflows, including the logistics integration rebuilt in-house before the iPaaS end-of-life date arrived, which is how the migration quote stopped being relevant. Metabase serves the dashboards. Claude on AWS Bedrock answers questions against the warehouse and drafts the follow-ups, with role-based access so people see what they should.
Ownership Built Into the Architecture
The ownership terms were contractual, and they shaped the architecture rather than decorating it. No business data leaves the client’s environment for a public model or a third-party service. The account, the data and every component belong to the client. HyScaler’s own access runs through a role the client can revoke without asking anyone. Access levels, territories and who sees what are set by the client’s team, not by a vendor’s support queue. The whole stack is open source, so there is no per-user licence anywhere in it to renegotiate later.
Methodology
The client’s own words set the design brief. “We’re at their mercy almost,” as the opening quote put it, “if we had control over some of those things, that would be great.” Every vendor relationship in the old stack extracted a fee for something the client could not change: the connector, the iPaaS renewal, the EDI provider holding its own mapping on infrastructure the client could not reach. A rebuild that solved the data problem while introducing a new set of vendors the client still did not control would have traded one version of the same complaint for another.
So the build followed two rules from the outset, not as terms added once the architecture already existed. Every component had to be open source, so there was no per-user licence to renegotiate later and no vendor positioned to raise a price simply because the client had become dependent on it. And everything had to run inside the client’s own AWS account rather than HyScaler’s, so the account, the data, and every credential belonged to the client from day one.
That is a different starting point than most platform rebuilds, which treat ownership as a contract clause negotiated once the system is already built. Here, ownership was a constraint the architecture had to satisfy before anything else, the same way the compliance requirements shaped the choice of database, orchestrator, and hosting model above. It is also why the previous CRM rollout’s failure mattered to this decision: a client already burned by a system built around a vendor’s convenience was not going to accept a second one, however capable it looked in a demo.
Results
The Arithmetic
This part comes first because it is checkable. The $10,000-a-year connector is retired. The $25,000 to $30,000 new-provider charge never got paid, and the 150 to 200 hours behind it never got spent, because the integration was rebuilt before the deadline that would have forced them. The replacement runs at about $458 a month of infrastructure, with AI usage at roughly $300 to $350 a month, and no per-user licensing anywhere in it.
Systems Consolidated
Ten-plus systems became one warehouse with one CRM, one orchestrator and a dashboard layer above them. Twelve of the 49 discovery gaps closed the moment the warehouse existed, before any AI was pointed at anything.
What’s Harder to Quantify
The sales team uses the CRM daily rather than working around it, which is the test the previous rollout failed. The engagement is still early, so no savings percentage and no revenue uplift is claimed here. What is claimed is the line-item arithmetic above.
Conclusion
The pattern repeats across industries: the problem is rarely the software a client runs; it is the space between the things they run, and the bills that accumulate in that space. This approach to retail system consolidation ensures that you own the data, the stack, and the long-term roadmap. The same discipline applies to every data engineering engagement HyScaler ships.
Glossary
- iPaaS (Integration Platform as a Service): third-party software that connects other systems together so they can exchange data automatically, without someone moving it by hand
- EDI (Electronic Data Interchange): the standardized way businesses exchange documents like purchase orders electronically, common across B2B and retail supply chains
- ERP (Enterprise Resource Planning): the core business system that tracks finances, inventory, and orders; NetSuite in this case
- CRM (Customer Relationship Management): the system a sales team uses to track customers, prospects, and every interaction with them
- B2B (Business-to-Business): sales made to other businesses, as opposed to individual consumers
- AWS (Amazon Web Services): the cloud provider hosting the client’s entire stack, inside their own account
- VPC (Virtual Private Cloud): an isolated, private section of a cloud account; nothing inside it is reachable from the open internet unless someone explicitly allows it
- RDS (Relational Database Service): Amazon’s managed database service, used here to run the PostgreSQL warehouse
- PostgreSQL: an open-source database system, the foundation the client’s data warehouse is built on
- Kestra, Metabase, Odoo: the named open-source tools running the workflows, dashboards, and CRM
- Claude on AWS Bedrock: an AI model running inside the client’s own AWS account rather than a third-party service, so no business data leaves the client’s environment
- NetSuite: the client’s existing ERP system, kept in place as the ledger of record