How a Century-Old Retailer Cut the Hidden Tax of Fragmented Internal Tools

Every operations leader has seen the same quiet waste: an engineer opens three portals, checks two Slack channels, and waits half an hour for a colleague in another timezone to confirm which API version is actually in production. Multiply that by dozens of staff and hundreds of tasks, and you've built an invisible tax on every release, every support ticket, every new hire's first month.

HEMA, a Dutch retailer with a century of operation behind it, reached the breaking point. Their teams were "portal-hopping"—constantly switching between internal tools, documentation sites, and messaging apps to find answers that should have been immediate. Rather than adding another interface to the pile, they built something different: an internal AI assistant called HAL that meets employees where they already work, with security architecture designed for a regulated enterprise rather than a startup's convenience.

The Real Cost Isn't the Tools—It's the Switching

Fragmentation feels like a minor annoyance until you measure it. A developer checks a deployment dashboard, switches to a wiki for context, opens a ticketing system for status, then messages a teammate when the first three sources conflict. Each switch carries cognitive load; the context lost between them often exceeds the answer gained.

HEMA's teams faced exactly this. The company runs substantial technical infrastructure behind its retail operations, with distributed engineering and support staff who need fast, accurate information to keep systems running. Portal-hopping wasn't a quirky habit—it was a structural bottleneck that compounded into delayed releases and burned-out teams who spent more time finding information than using it.

The insight from HEMA's experience, documented by AWS, is that AI assistants only deliver ROI when they eliminate this friction rather than adding to it. HAL doesn't live on a separate website or require another login. It operates inside the tools teams already use, pulling from governed knowledge sources without creating yet another place to check.

Why Architecture Matters More Than Model Hype

Most AI discussions focus on capabilities—reasoning, coding, multimodal understanding. HEMA's project suggests a different priority: integration architecture determines whether those capabilities ever reach your employees productively.

HAL is built on Amazon Bedrock AgentCore, but the critical design choice isn't the underlying model. It's the Model Context Protocol (MCP), an open standard that lets the AI assistant connect to multiple data sources and tools through a consistent, governed interface. MCP acts as a universal adapter that lets HAL query documentation, check system status, and surface answers without custom integration work for each new source.

The practical consequence: HEMA can add or change backend systems without rebuilding the assistant. The AI layer stays stable even as underlying tools evolve. For a company with accumulated technical debt, this decoupling is essential. Startups can afford to rebuild; established operations need integration paths that respect existing investments.

Security Designed for Compliance, Not Convenience

Here's where HEMA's implementation diverges from typical AI pilots. Many internal assistants handle security as an afterthought—users authenticate once, credentials sit on devices, and access controls are coarse at best. This becomes a compliance liability at scale, especially in regulated industries or any company with audit requirements.

HAL's security is anchored in Microsoft Entra ID, with a critical constraint: no AWS credentials reside on client devices. Authentication flows through existing corporate identity infrastructure, and permissions follow the same role-based access that governs other systems. The AI assistant can only see what the user is already authorized to see.

This matters operationally because it lets HEMA deploy HAL broadly without creating parallel access management or shadow credential stores. It matters strategically because it turns AI from a governance risk into a governance extension—enforcing existing policies rather than circumventing them.

What "100 Years Old" Actually Means for Modernization

The "100-year retailer" framing carries weight beyond marketing. HEMA's challenge isn't unique to retail or to century-old companies—it's the universal tension between established process and necessary change. Legacy process debt accumulates invisibly: workflows that made sense for 2015's team size, documentation spread across three platform migrations, tribal knowledge held by people who've since moved on.

Modernization pressure hits these organizations just as hard as startups, but the path is steeper. You can't simply replace everything. You need deliberate integration design that connects new capabilities to existing systems without requiring wholesale migration.

HAL's architecture reflects this reality. MCP provides the integration flexibility. Entra ID anchoring respects existing security investments. The assistant layer sits between users and fragmented sources rather than replacing them prematurely. It's a pragmatic approach that acknowledges real organizational constraints.

A Practical Checklist for Operations Leaders

If HEMA's situation resonates, the next step isn't calling your AWS rep—it's assessing whether your current fragmentation justifies the investment and whether your organization can support governance-first design.

Audit the invisible tax

• Survey engineers and support staff: how many portals do they check for a typical task? How long do they wait for answers from colleagues?
• Map your top five recurring questions that require multiple sources to resolve
• Calculate hours per week spent on information-finding rather than value-creating work

Evaluate existing AI pilots honestly

• Do they integrate into current workflows or require another interface?
• Is access governed through your existing identity system, or do they create parallel credentials?
• Can they query multiple backend sources without custom integration per source?

Assess security architecture readiness

• Is your identity provider (Entra ID or equivalent) already the authoritative access control for internal tools?
• Can you enforce that AI assistants inherit these permissions without credential distribution?
• Would your security team approve an assistant that sees everything, or one that respects existing boundaries?

Determine integration feasibility

• Do your critical knowledge sources expose APIs or structured access that an MCP-based assistant could consume?
• Is your AWS infrastructure mature enough to support Bedrock AgentCore deployment, or would you need significant foundation work?

The Decision Beneath the Technology

HEMA's case, as presented by AWS, is a single-company story with self-reported outcomes—not independently verified benchmarks. The specific security architecture reflects HEMA's Microsoft-centric environment, not universal applicability. Results will vary based on your existing identity infrastructure and cloud maturity.

But the underlying decision is broadly relevant. Most mid-to-large companies with multiple internal platforms face the same choice: continue tolerating the productivity drag of portal-hopping and tribal knowledge, or consolidate access into a governed layer that meets users inside their existing workflows.

The technology to do this exists today. The harder question is whether your organization will prioritize integration architecture and security governance over chasing model capabilities that never reach your employees productively. HEMA's century of operation suggests that longevity comes from deliberate, constraint-respecting modernization—not from being first to adopt, but from being right about what actually changes how work gets done.