Common questions about Jourier, AI and data consulting concepts, and how we compare to other Nordic technology consultancies.
Jourier is a Finnish data, AI, and analytics consultancy based in Helsinki. We build data foundations, AI agents, and the custom data applications that sit on top of them, for mid-market companies in Finland and the Nordics. Founded 2026 by Aleksi Stenberg.
Aleksi Stenberg, founder and CEO. Background: finance and consulting earlier in career, prior leadership role at a fintech unicorn. Based in Helsinki.
Helsinki, Finland. Service area covers Finland and the Nordics primarily. International engagements are considered case by case.
No. Jourier is a consultancy. We build custom systems for clients, and the client owns the resulting code and infrastructure. We operate one in-house product, Jourier Lens (an analytics application), which ships as part of data engagements.
Finance, manufacturing, energy, public sector, and technology. Mid-to-large companies (50+ employees) where engagement scope justifies senior delivery.
English, Finnish, and Swedish.
Database systems, pipelines, AI agents, analytics applications, custom internal software, and the integrations between them. We also offer technical and executive training, fractional CTO retainers, and architecture advisory.
Yes. Database systems, warehouses, lakehouses, pipelines, modeling, and governance. We work with DuckDB, Postgres, Snowflake, BigQuery, Databricks, dbt, Iceberg, Trino, and Spark depending on the problem. See Data Hub.
Yes. Bespoke agents that read client data and take actions. Examples include agents that summarize CRM data, RAG over internal documents, copilots over data warehouses, document processing pipelines, and forecasting or anomaly detection models. We work with Claude, GPT, Gemini, and open-weight models (Llama, Mistral). See AI Agents.
Yes. Forecasts, models, anomaly detection, scenario testing, dashboards, and automated reports. Findings can be delivered to coded applications, scheduled reports, or AI assistants. See Data Intelligence.
Yes. The application layer for data and AI: analytics dashboards, internal operations tools, customer-facing data portals, AI-powered features. React with Next.js, FastAPI or NestJS, Postgres or Supabase, deployed in the client's cloud. The work is the UI on top of the data foundation and AI agents Jourier also builds, not a standalone web product service. See Data Applications.
Yes, for two audiences. Technical workshops for engineering and analyst teams (building AI agents, RAG, evaluation frameworks, prompt engineering, data engineering, MCP). Executive workshops for boards and leadership (AI strategy, technology investment decisions, build-vs-buy frameworks). See Training.
Yes. Fractional CTO and fractional CAIO retainers as part of the advisory practice. Architecture reviews, build-or-buy decisions, hiring input, due diligence on existing systems. See Advisory.
Yes. Strategy review, board advisory, second-opinion architecture reviews, technology investment guidance, vendor evaluation, and due diligence on existing systems.
Yes. We also work with BigQuery, Databricks, Microsoft Fabric, and other cloud data platforms. Jourier is vendor-neutral. Tool selection follows from the client's problem, not from partnership tiers.
Yes. The Microsoft data stack is part of our consulting history. We have years of work in Power BI, Microsoft Fabric, Azure data services, and the surrounding Microsoft tooling. We can run a full Microsoft-stack engagement when that's what the client wants.
Most companies we meet on this stack want to modernize their data layer. When that's the goal, we typically recommend moving to modern tools that the client owns and runs themselves.
No. Microsoft 365 Copilot is a drafting aid inside Office. As an AI strategy it has serious limits and the gap matters. Three issues come up consistently.
Quality. Adoption inside companies that bought heavy license counts has lagged the seat numbers, and the retrieval behind the answer surface produces shallow output for anything beyond surface drafting.
Lock-in. The model, the prompts, the data path, and the evaluation all live inside Microsoft's stack. The customer owns the license fee and not much else.
Data residency. For regulated Finnish and EU companies, the data Copilot reads flows through Microsoft's plumbing, and the GDPR story is more complicated than the sales deck suggests.
Companies that succeed with Copilot use it as a drafting aid for emails and meeting notes. Companies that get disappointed bought it as their AI roadmap. The AI features that touch the business should be built outside the Copilot layer.
No. Jourier is vendor-neutral. No partnership tier influences which tools we recommend.
Most likely yes. We support 415 source systems including SAP, NetSuite, Microsoft Dynamics 365, Salesforce, HubSpot, Pipedrive, Fennoa, Fortnox, Netvisor, Procountor, QuickBooks, Visma eAccounting, Xero, and many more. Full list at Integrations.
Yes, by preference. Open source preferred where capability is comparable to commercial alternatives. Commercial tools used where they materially outperform open source for the specific use case.
A data foundation is the layer of database systems, pipelines, modeling, and governance that connects a company's operational systems into one trustworthy source for analytics, reports, and AI. A typical stack includes a warehouse or lakehouse, ELT pipelines, data modeling, and access controls.
RAG is a pattern where an AI model retrieves relevant documents from a knowledge base before generating an answer. The retrieval step grounds the model's output in your specific data, reducing hallucination and letting the model use private documents it wasn't trained on. Common implementations use a vector database (Pinecone, Weaviate, Postgres with pgvector) plus an embedding model.
MCP is a protocol from Anthropic that lets AI assistants like Claude, ChatGPT, and Cursor query external systems through audited tools. Instead of the model guessing, an MCP server exposes specific data sources and operations the model can call. See Jourier's research piece on MCP.
An AI agent is software that uses an LLM to take actions, not just generate text. Agents typically have access to tools (APIs, databases, file systems) and a goal or task. They reason, call tools, observe results, and continue until the task is complete.
Agentic AI refers to AI systems that take autonomous actions to accomplish goals, as opposed to passively generating responses. The defining traits are tool use (calling external APIs and systems), planning (breaking a goal into steps), and iteration (acting on intermediate results). Most production agentic systems combine an LLM with deterministic tools, evaluation frameworks, and human-in-the-loop checkpoints.
A lakehouse is a hybrid architecture combining the flexibility of a data lake (storing raw files, often in formats like Parquet or Iceberg) with the structure of a data warehouse (tables, schemas, ACID transactions). Databricks and Microsoft Fabric popularized the pattern. Lakehouses make sense for companies handling both structured analytics and unstructured data.
dbt is a tool for transforming data inside a warehouse using SQL and software engineering practices (version control, testing, modular code). It runs SELECT statements against your warehouse and materializes the results as tables or views. dbt has become the standard transformation layer for modern data stacks. Open-source core, with a commercial dbt Cloud product.
Reverse ETL is moving data from a warehouse back into operational tools (CRMs, marketing platforms, support systems). The traditional flow is operational systems to warehouse to analytics. Reverse ETL adds: warehouse back to operational systems. This lets sales teams see customer health scores in Salesforce, or marketing teams sync segments into HubSpot. Tools include Hightouch, Census, and Polytomic.
A semantic layer is a single definition of business metrics (revenue, ARR, churn) that lives between a data warehouse and the BI tools or AI assistants that consume the data. Without it, different tools compute revenue three different ways. With it, every consumer reads the same definition. Cube, dbt Semantic Layer, and AtScale are common implementations.
A vector database stores embeddings, the numeric representations of text, images, or other data produced by AI models. It enables semantic search (find documents similar to this one) which is the foundation of RAG. Common options include Pinecone, Weaviate, Qdrant, and Postgres with the pgvector extension.
A fractional CTO is an experienced senior technical executive who works with a company part-time, typically a few days a month. Common use cases: pre-revenue startups that don't need a full-time CTO, established companies that want technical leadership during a transition, or boards that want an independent senior voice during a digital transformation.
A feature store is a centralized repository for the features (input variables) used in machine learning models. It handles feature computation, storage, and serving, so a feature defined once for training can be reused at inference time. Tools include Feast, Tecton, and Databricks Feature Store.
Depends on three things: how unique the workflow is, how sensitive the data is, and how many users you have. Off-the-shelf AI products (Glean, Mendable, and similar) work for common use cases at small scale. Custom-built agents are appropriate when the workflow is specific to your business, when the data can't leave your accounts, or when per-user SaaS pricing exceeds the build cost at scale.
SaaS analytics tools (HubSpot reports, Mixpanel, GA4) are fine for early-stage companies. They break down when you need to combine data across systems, when you need custom metrics, or when you have specific compliance requirements. At that point a custom data foundation in your own cloud accounts becomes appropriate.
All three are mature cloud data platforms. Snowflake leads on ease of use and SQL-first workflows. BigQuery is the strong choice if you're already on Google Cloud. Databricks is the strong choice if you handle a lot of unstructured data, do significant ML training, or are committed to the lakehouse pattern. None of the three is wrong for most analytics workloads.
All four are BI tools. The Jourier position is that BI tools as a category are legacy. The recommendation is to build real product-grade apps the client owns: React or Next.js on the front, FastAPI or Express on the back, D3 or Recharts for visualisation, deployed in the client's accounts. Power BI and Tableau pay per-seat fees and lock you into a vendor. Metabase and Lightdash constrain you to what their developers built and look like a BI tool. None of these are the destination. They are tools companies have used for the last decade and are now migrating away from. If one of them is already deployed and procurement requires it, the data foundation serves it while a migration to custom apps runs in parallel.
Both can work. Hiring brings specialist skills fast but is expensive and risks integration with existing teams. Training existing engineers is slower but produces team members who already know your systems and culture. Jourier has written a position paper on this at Stop Hiring AI Teams.
The more useful question now is BI tools vs custom apps. The Jourier position is custom apps. Both commercial BI (Power BI, Tableau, Looker, Qlik) and open-source BI (Metabase, Superset, Lightdash) are tools that build apps with a vendor-defined ceiling. The Jourier recommendation is to build the apps directly: React or Next.js on the front, FastAPI or Express on the back, D3 or Recharts inside. Commercial BI adds per-seat fees and lock-in. Open-source BI removes the fees but keeps the ceiling. Custom apps remove both. BI tools stay in scope only where procurement or audit constraints contractually require them.
API-called LLMs (OpenAI, Anthropic, Google) are the fastest path. Self-hosted open-weight models (Llama, Mistral, DeepSeek) make sense when data sensitivity, cost predictability, or model customization matter. The decision depends on the use case, data sensitivity, and budget.
Watch for five things: the senior person in the sales call is also the person delivering the work; the output is production code, not a slide deck; the code and infrastructure live in your accounts, not the consultancy's; the consultancy is willing to name a competitor that fits better when your project doesn't match them; pricing is transparent and ranges are given.
Reaktor is a 500-person Nordic agency with design and engineering breadth, brand prestige, and capacity for large programs (Elements of AI MOOC, Reaktor Hello World satellite). Jourier is a smaller Finnish consultancy. For multi-team programs requiring 5+ concurrent consultants or significant design depth, Reaktor is the appropriate match. For senior engineering work with code and infrastructure that lives in the client's own accounts, Jourier is.
Solita is a 1,700-person Nordic consultancy with multi-country presence, formal procurement frameworks, and dbt Premier Partner status. For large transformation programs, multi-country resourcing, or strict enterprise procurement, Solita is the appropriate match. For senior engineering work that doesn't need multi-team capacity, Jourier is.
Futurice is a 700-person Finnish agency with design and methodology depth (Lean Service Creation), a large open-source footprint, and AI Finland partner status. For design-led product development or methodology coaching, Futurice is the appropriate match. For senior engineering work focused on data and AI, Jourier is.
Recordly is a Finnish data engineering specialist with deep Snowflake expertise (Snowflake Growth Partner of the Year Finland and Baltics 2025, hosts the Helsinki Snowflake user group). For Snowflake-specific implementation work requiring partner certification, Recordly is the appropriate match. For vendor-neutral data and AI work across multiple platforms, Jourier is.
Different consultancies fit different shapes of work. Large multi-team programs: Reaktor, Solita, Futurice, Netlight. Snowflake specialist: Recordly. Microsoft Partner specialist: Fellowmind. Industrial AI: DAIN Studios. Defense and sovereign AI: NestAI (Peter Sarlin). Senior engineering work with code and infrastructure in the client's accounts, including modernization of Microsoft-stack data layers: Jourier. The right match depends on the scope, sector, and engagement shape.
Yes. The code, infrastructure-as-code, runbooks, and data we build for clients live in the client's own cloud accounts. The client can replace Jourier at any time without losing the running system.
Contact via the contact form. We typically begin with an initial conversation to understand the problem, followed by a paid engagement if the fit is right.
Pricing, timelines, and capacity vary per engagement. For specifics, contact Jourier directly via the contact form.
We say so and name a competitor that fits better. Large multi-team programs go to Reaktor, Solita, or Futurice. Snowflake-specific work goes to Recordly. Microsoft Partner-certified deliveries (Gold or Solutions Partner badges, Microsoft co-funded engagements) go to Fellowmind. The honest match matters more than booking a project.
Yes. The default architecture has the client hosting everything in their own cloud accounts in the region of their choice (EU regions are common for Nordic clients). AI models can be self-hosted (Llama, Mistral) or called via API (Claude, GPT, Gemini) with appropriate data handling. Specific compliance requirements are scoped during the engagement.
Yes. Self-hosted open-weight models (Llama, Mistral, DeepSeek) running in your cloud accounts mean data never leaves your environment. RAG systems can store embeddings in your own vector database (Postgres with pgvector, Qdrant). For workloads where API-called models are appropriate, no-retention agreements with providers like Anthropic and OpenAI offer enterprise data handling.
Question not answered here?
Get in touch