ACP bought. Data to Vertex? Not necessarily.
PIM automation, AI search, Copilot — optionally with a local model.
Artificial Intelligence (AI) as an extension for stable Spryker setups in B2B industry. Spryker ACP bought but unused? Customer data should not go to Vertex AI? Spryker Glue becomes the model-independent connector — reorder models, AI search, Copilot optionally on local inference or with DLP filter in front of the cloud API. Data stays in-house, default. Cloud only when controlled.
Three developments.
All ready to deploy now.
Spryker has shipped three AI layers in the last 18 months. Most DACH customers have licensed or have access to parts of them — but have not fully activated them. The stack audit clarifies what you already have and what is missing.
The Spryker App Composition Platform is live in production with AI-relevant apps: Algolia for AI Search and recommendations, Akeneo Productlogic for PIM enrichment, Google Vertex AI for personalisation. Many installations have ACP apps in the contract that are not yet active in production — the most common entry trigger for this engagement. Sovereignty caveat: Vertex AI, Algolia, and Akeneo Productlogic are cloud services with no on-prem option. Glue API as the connector layer allows provider switching.
Source: Spryker ACP documentation · Algolia, Akeneo, Google Vertex AI partner status (2024/2025)From Spryker Cloud Commerce OS 202410 and 202501, AI Search and merchandising is natively available: semantic search, AI-driven merchandising rules, product recommendations based on behavioural data. Particularly relevant for B2B industry with technically described catalogues: product identification via natural-language query instead of exact part number. Spare parts identification for MRO / C-parts catalogues is the primary industrial use case.
Source: Spryker Cloud Commerce OS Release Notes 202410 · 202501 (Spryker Systems, 2024/2025)Spryker Copilot launched in Q4 2024 as Early Access and is rolling out gradually as module GA in 2025/2026. Back-office workflows: product content generation, merchandising configuration, order routing suggestions. Compatibility depends on the hosting model. DACH customers — Hilti, Ricoh, Siemens, Toyota Material Handling — run on Spryker. Industrial B2B catalogue is the core context.
Source: Spryker Copilot Early Access Announcement (Q4 2024) · Spryker DACH reference customers
Spryker module known.
Sovereignty option per row.
For each use case: which Spryker module provides the foundation, where the industrial lever sits, and which sovereignty option is available — cloud provider or local inference via Glue.
| Use case | Spryker module / ACP app | Industrial lever | Sovereignty option |
|---|---|---|---|
| Reorder-Predictions | Order Management + ACP Vertex AI / own model via Glue | MRO / C-parts: automatic reorder suggestions before stockout. Customer retention through proactive basket preparation. | Reorder model trained and inferred locally on your own GPU instead of Vertex AI. Transaction data never leaves the network. |
| RFQ-Automation | Quote Request Module + n8n via Glue as workflow layer | Automatically pre-structure complex enquiries, propose terms, accelerate routing. Sales capacity freed for high-value cases. | n8n on-prem + local Llama 3.3 70B for request classification and terms suggestion. Full data sovereignty over quote data. |
| Account-Manager-Copilot | Spryker Copilot (module GA 2025/2026) + back office | Customer context on demand: order history, open enquiries, product gaps. Account manager prepared, not post-researching. | Copilot hosting model determines egress. Self-managed Spryker + local inference via Glue as fully sovereign option. |
| Spare-Parts-Identifikation | AI Search (202410 native) + image upload via Spryker Glue | Identify technical spare parts by photo instead of part number. Reduces order errors, relieves support, increases self-service rate. | Image-matching model trained locally on your own product catalogue via vLLM. No image upload to external vision APIs by default. |
Three patterns blocking Spryker AI potential.
Three situations from active engagements — what prevents companies with a live Spryker installation from getting the AI layer into production.
-
Pattern 01 · Dormant ACP licence
ACP apps are licensed. None is active in production.
Algolia, Akeneo Productlogic, or Vertex AI are in the contract, but activation has been deferred — missing resources, unresolved data questions, conflicting priorities. Licence costs are running; value is not being generated. The stack audit establishes in Phase 1 which ACP app is activatable with what effort, and which sovereignty constraints are relevant in each case.
- Consequence
- Licence costs without ROI, AI potential untapped
- Frame reset
- Stack audit with ACP activation status and prioritisation by lever
-
Pattern 02 · Search stagnation
Search runs on keyword matching. Semantics are completely absent.
B2B catalogues with thousands of spare parts, technical specifications, and variant-heavy products are poorly suited to classic keyword matching. Customers search by function, not by part number — and end up nowhere. AI Search from 202410 is available in many installations and has not been activated. Spare parts identification by image is the next step, improving self-service rate and order quality simultaneously.
- Consequence
- High support load, order errors from manual part number searches
- Frame reset
- Activate semantic AI Search, image matching as the next level
-
Pattern 03 · Reorder blindspot
Reorders happen reactively. The stockout comes first.
MRO / C-parts customers order when stock is empty — not before. Reorder prediction based on order history and consumption rhythm is a solved ML problem that is integrable in Spryker via Order Management and Glue API. Proactive basket preparation retains customers and reduces emergency orders. The model can be trained locally — transaction data never leaves the network.
- Consequence
- Stockouts, reactive order cycles, missed upsell moments
- Frame reset
- Reorder model on-prem, proactive basket suggestion via Spryker Order Management
Spryker has the layers. What is missing is the lever plan — which use case first, with which sovereignty option, in which phase.
Five questions, five straight answers.
- "We have ACP — but nothing is running. What is holding us back?"
- Usually a combination of missing activation resources, unresolved data questions, and absent use case prioritisation. The stack audit establishes in Phase 1 which app is activatable with what effort — and which sovereignty constraints matter here.
- "Google Vertex AI is US cloud. Do we have an alternative for personalisation?"
- Yes. Spryker Glue API is the model-agnostic connector layer — you swap the AI provider without core changes. Personalisation and reorder models can be trained locally and inferred at DACH hosting providers (AD IT Systems, IONOS AI Model Hub, Hetzner GPU, StackIT). Vertex AI remains available as an option for non-sensitive data.
- "What does AI Search actually do for us — we have a 50,000-SKU catalogue?"
- Semantic search enables product identification via natural-language query instead of exact part number. For technical catalogues with spare parts, variants, and specifications, that is a direct lever on self-service rate and support load. From 202410 that is natively available — activation effort depends on your Spryker version.
- "Can we run Reorder-Predictions on-prem?"
- Yes. The reorder model can be trained on your order history locally and inferred on DACH GPU infrastructure. Transaction data never leaves the network, no Vertex AI required. Spryker Glue provides the connection to Order Management.
- "How do we get started?"
- Phase 1 is a stack audit: ACP activation status, AI Search version, Copilot compatibility, use case map with sovereignty options per use case. Terms are mandate-specific. Extensions are decided after Phase 1 — no lock-in.
Workbench, loops, handover. Three principles.
We build with you, not for you. Engagements are workbenches, not theatre. Delivery in short loops. Clean handover, then your team carries it alone.
Method in detailDeep platform expertise.
25 years in IT, 14 of them in B2B commerce. Architecture mandates with large enterprises, building and steering distributed expert teams, vendor-neutral project rescue. Focus areas: platform architecture, project rescue, team operations.
Chris Zepernick
Spryker has the layers. What is missing is the sovereignty plan.
ACP apps, AI Search, Copilot — these are ready-made building blocks. What most installations still lack is a use case plan that resolves sovereignty per application. Vertex AI for personalisation is legitimate when the data allows it. For transaction and customer data, a local model via Glue is the safer choice. The difference is not in the model — it is in the lever plan.
How we workWhich AI lever pulls first for you?
Spryker is running. The AI layer is ready. A few minutes on the phone — we sort out which use case pulls first at your installation.
