Guide

SAP generative AI consultant versus architect: who does what

As of 2026-10-10

Edited by Cédric Mary, sole editor and publication director — 27 years in the SAP ecosystem. How these pages are produced, and where automation is used.

An SAP generative AI consultant, also called an SAP GenAI consultant, gets language models working on SAP data and processes under control: model access through the Generative AI Hub, an orchestration pipeline, grounding on your data, filtering and masking. An SAP GenAI architect decides beforehand which layer, which model, what cost and what compliance posture. This page is about the programme role; the product itself is covered in our guide to the Generative AI Hub consultant. To test a candidate, ask which orchestration modules they have configured and for a case where a filter stopped a request.

Looking for a SAP AI consultant? Cédric Mary, SAP AI & Analytics Architect & Team Lead, takes SAP Business AI, Joule and Business Data Cloud missions.

What the Hub is, and what it is not

SAP documents it as a capability of AI Core and AI Launchpad that gives access to generative models and their orchestration in a governed runtime. It does not replace AI Core, it extends it: execution, quotas and billing stay AI Core mechanisms, and it needs the extended service plan.

Two routes lead to a model. Under the foundation-models scenario each model is deployed per resource group and called through its provider's API. Under the orchestration scenario a single deployment exposes a harmonised API modelled on OpenAI's, so switching model is a configuration change, at the cost of provider-specific features the mapping does not cover. That trade-off is an architect's decision; the consultant has to live with it.

The orchestration pipeline, in the order SAP fixes

One call chains modules in an order SAP sets centrally: grounding, templating, translation, masking of personal data, input filtering, the model call, output filtering, then unmasking. Only templating and the model call are mandatory.

Three failures recur: an input rejected by a filter, in which case the model is never called; an output rejected after the call; and an answer that is well formed but wrong because grounding returned nothing useful. Someone who has really built a pipeline reads the intermediate results and names the stage that failed.

Grounding: documents, structured data, graph

Documents suit vector search; figures and business objects do not. They come from governed data products, Datasphere models or HANA tables, and the knowledge graph supplies the semantics of which object a question refers to. A question such as which customers are affected if a work centre goes down follows a chain of relations that no similarity search reconstructs.

An analytics background counts for a great deal here. A clean semantic model and accessible, permissioned data weigh more on project success than prompt skill, so a Datasphere or SAC consultant does not start from zero: what is missing is the mechanics of the generative layer.

Guardrails and model choice

Content filtering can run on input and output with two services, one of which also detects prompt attacks. Masking of personal data has two modes: anonymisation, irreversible and safer, and pseudonymisation, reversible in the response and needed when the output must name a person. Coverage varies by language and entity type, so a test on your own data is not optional.

On model choice, SAP gives access to third-party and its own models, including sovereign options on its infrastructure. SAP does not publish which model serves which Joule capability; you choose a model only for AI you build on the platform. The architect should present a selection criterion covering execution region, cost and portability, not a model name.

Decision table: consultant, architect, or both

Question over internal documents — Roles: consultant · Modules: document grounding, template · Check: indexed volume, refresh, access rights.

Question over figures — Roles: consultant with Datasphere background · Modules: structured grounding · Check: how the models are described.

Choice of model provider — Roles: architect · Modules: harmonised orchestration · Check: dependence on provider-specific features, lock-in.

Personal data in the prompt — Roles: both · Modules: masking, filtering · Check: test in your languages, transfer assessment.

Use touching people — Roles: architect with legal support · Modules: all · Check: classification under the EU AI Act, before any pilot.

Production, skills and evidence

A proof of concept often runs in a default resource group on one developer's credentials, because that is the fastest route to a demo. Production tolerates neither. AI Core organises each asset through scenario, executable, configuration and deployment, and the deployment is what turns a model call into an operated, versioned and monitorable object. Resource groups carry isolation, cost allocation and access control.

The consultant should name that gap early: where the application will run, who operates it, what it costs under SAP's commercial model, and what the exit is if it does not pay. Ask for one use case they took into operation, not a count of prototypes.

SAP's generative AI developer certification is taken hands-on in an AI Launchpad tenant: prompt templates, orchestration, structured outputs. It rewards a different profile from a memory test, yet it does not replace a project reference.

SAP's own analysis locates the shortage in people who link models to governed data, processes and regulation. That suits an analytics consultant who masters the semantic layer and has added the generative mechanics; have the candidate describe exactly how.

SAP generative AI consultant and SAP GenAI consultant: one job, two spellings

Postings and CVs use SAP generative AI consultant and SAP GenAI consultant for the same work: getting language models running on SAP data under control. The spelling tells you nothing; the evidence does. Ask which orchestration modules the person configured, for a case where a filter stopped a request, and for one use case taken into operation.

The market uses generic wording. On 7 October 2026 only 4 of 1,657 live postings named AI Core or the generative AI hub, 3 of them freelance and all in Germany, against 93 naming generative AI or GenAI, 54 of them freelance and 61 in France.

SAP generative AI architect and SAP GenAI architect: what they decide first

An SAP generative AI architect, also written SAP GenAI architect, settles four things before the consultant configures anything: the layer (a feature SAP ships, a Joule agent, or a custom application on the hub), the model selection criterion covering execution region, cost and portability, the grounding channel, and the compliance posture.

The hub trade-off is theirs. The orchestration scenario makes switching model a configuration change but gives up provider-specific features. Any use touching people is classified under the EU AI Act before a pilot starts, with legal support.

SAP GenAI expert, SAP GenAI developer and SAP BTP GenAI consultant: reading the labels

SAP GenAI expert is a self-description, not a qualification. SAP GenAI developer points at build work: prompt templates, orchestration configurations and structured outputs. SAP's generative AI developer certification is taken hands-on in an AI Launchpad tenant and rewards that profile, though it does not replace a project reference.

SAP BTP GenAI consultant adds the platform layer: an AI Core tenant on the extended plan, a resource group per use case and environment, and credentials someone owns. Whatever the label, ask for the same proof: a pipeline built, a failure read at the right stage, a production hand-over.

What we cannot assert

SAP does not publish which model serves which Joule capability, and no consultant rate is cited. Provider and model names change by release.

Frequently asked

Is an SAP GenAI consultant the same as an SAP GenAI architect?

No. The consultant configures and delivers the pipeline; the architect settles layer, model, cost and compliance first. On a small scope one person can hold both.

Do I need the Generative AI Hub to use Joule?

No. Joule and features shipped inside applications do not require you to operate the hub. It serves applications and agents you build on models yourself.

Can the model be changed without rewriting the application?

Under orchestration, yes, as long as you stay within the harmonised API. Provider-specific features sit outside it.

Does the data sent to the model stay in Europe?

That depends on the execution region of the chosen model, which orchestration does not decide for you. Read the data-usage clauses of your contract as well.

Can I hire a freelance SAP generative AI consultant?

Yes, mostly in France: on 7 October 2026, 54 of the 93 postings naming generative AI or GenAI were freelance, 61 in France. We hold no published day rate for the role.

What should I ask a freelance SAP GenAI consultant before engaging?

Three things: the orchestration modules they configured, a call that failed because a filter fired, and a use case taken into operation rather than a count of prototypes. Then settle where the work runs, who holds the credentials, who operates it at the end, and keep SAP's consumption as a pass-through line.

Where do I find an SAP generative AI freelancer or an SAP GenAI freelancer?

Under generic wording rather than SAP product names: search generative AI or GenAI with SAP. A common profile is a Datasphere or SAC consultant who has added the generative mechanics; have the candidate describe exactly how.

Is an SAP generative AI contractor or SAP GenAI contractor billed differently?

The wording does not change the commercial model. Contractor and freelancer are engagement forms; SAP bills model consumption separately, in AI Units for delivered AI and in capacity units for custom AI. Add a re-pricing clause, given how often SAP changed its model in 2025 and 2026.

Engage the editor

This guide is written by Cédric Mary, SAP AI & Analytics Architect & Team Lead. 27 years of SAP — Datasphere, Business Data Cloud, SAC Planning, Joule. Freelance or permanent · Hybrid / remote · Anywhere in EMEA. Available from 19 October 2026.