Why the EU Mittelstand needs cloud architects, not cloud salespeople

The pitch every Mittelstand CIO has heard: “Migrate to the cloud, and your unit economics improve.” The reality, two years in: monthly invoices balloon, no one knows which workloads drive the spend, and the original promise looks oversold.

It’s not the cloud’s fault. It’s that the firms making the migration decision were sold by people who got paid the day the SKU shipped, not the day the workload ran cheaper.

The reseller incentive problem

A cloud reseller’s economics are simple. They earn a margin on the consumption their accounts drive: sometimes a fixed percentage, sometimes a tiered rebate from the hyperscaler. Their best year happens when their clients spend more. Their worst year happens when their clients spend less.

Architecture is the opposite economics. An architect’s reputation is built on the workload running smaller next quarter than it does this quarter. Same throughput, same SLOs, lower bill. Recommending a smaller instance, killing a region, deleting an idle account, refusing a “modernization” that adds cost without value: those moves are how an architect proves competence.

The reseller and the architect can sit on the same call. They will give different advice.

Three patterns we see in week one

Mittelstand engagements share a recognizable shape. Within the first week of a review we typically surface three categories of waste.

1. Over-provisioned non-production environments

Engineering teams spin up production-grade instances for sandboxes “just in case.” Tooling deploys mirror prod sizes because the original template was prod’s. Nobody scales test environments down at night or on weekends because the script was never written.

The fix is rarely glamorous. It’s a Tags taxonomy (environment, owner, cost-center), an Environment != prod autoscale-to-zero rule overnight and weekends, and a quarterly review of orphan-looking resources. In a typical mid-market AWS or Azure tenancy this lever alone returns 10-18% of monthly spend.

2. Forgotten egress

The Mittelstand data-stack pattern is classic: a SAP source, a Snowflake or Synapse warehouse, a BI tool sitting in front, and a dozen analyst exports each week. Each hop crosses a network boundary. Each crossing meters egress.

The bills are insidious because individual hops are small. The aggregate, for a 50-person analytics team running 200 dashboards across continents, can comfortably hit five figures per month, none of it forecasted.

The fix is part architecture (collocate the warehouse with the BI tool’s region; cache results closer to the user; use VPC endpoints for hyperscaler-internal traffic) and part discipline (alert on the top-N egress endpoints monthly).

3. Reserved-instance gaps

Most Mittelstand firms don’t qualify for the enterprise-tier discount programs the hyperscaler reps quote. What they often don’t realize is that 1-year Reserved Instance and Savings Plan commitments at the smaller account scope still produce 30-50% net discount on steady-state compute, without needing the enterprise paperwork.

The blocker isn’t access. The blocker is that the modelling work hasn’t been done. Which baseline of compute is steady-state? Which is bursty? Which is dev-only? An architect-led 2-week portfolio model usually justifies one or two strategic commitments and explicitly recommends against others.

What an architect-led review actually delivers

A typical 3-week architecture review for a Mittelstand firm produces four deliverables:

1. Cost-driver report. The top 10 line items, broken out by service, account, and workload owner. Includes the question “do you know this was running?”, which alone often kills 5-8% of monthly spend.
2. Right-sizing recommendations. Per-workload, with the SLO impact stated. Not a generic “size down everything by 30%” claim; specific instance families and sizes with a defensible rationale.
3. Commitment portfolio model. Which workloads to lock for 1-year, which for 3-year, and which to leave on-demand. Includes the risk model: what happens if forecast is wrong.
4. Operating model recommendation. Who runs FinOps from here. Is there a quarterly business review, or just an invoice that lands in a CFO inbox once a month.

The deliverables aren’t novel. The discipline around them is. The reason Mittelstand engagements over-deliver on the headline savings number is that the firm has rarely had any one person looking at all four of these axes together.

The Mittelstand-specific issues

Three things make Mittelstand engagements different from F500 engagements:

Decision velocity. A Mittelstand firm typically has one technology decision-maker per major axis (one for cloud, one for data, one for security). That person is busy. The review needs to deliver a concrete shortlist of decisions, not options to debate.

Audit posture. German firms in particular treat the IT general controls audit seriously. Any cost-reduction recommendation has to come with the change-management story. Killing a region is a 30-minute conversation; killing a region without breaking the audit trail is a 3-week project.

Regulator alignment. DORA, GDPR, and the upcoming AI Act don’t reward cost optimization that breaks data-residency or third-party-risk posture. EU Mittelstand cloud reviews have to bake those constraints into the recommendation, not retrofit them at the end.

A reseller pitch generally ignores all three. An architect-led review treats them as the primary input.

Where this lands

The honest summary of two years of Mittelstand FinOps engagements: 25-55% of monthly cloud spend was avoidable. The variance is workload-dependent. ML training and large data pipelines have more headroom; lean web stacks have less.

The work is not glamorous. It’s data-driven, slightly boring, and almost always defensible to an auditor. It also happens to compound: a portfolio with right-sized non-prod, committed steady-state, and a working FinOps practice is a portfolio whose next 12 months of cost are predictable.

The pitch you should be wary of is the one that promises 60-70% in 30 days. The pitch you should engage with is the one that opens with “show me your top 10 line items, by service and account, and let’s talk.”

A reseller will close the deal either way. Only one of those conversations leaves you with a smaller invoice.