Service: Agentic Analytics

Agentic analytics: self-service with numbers you can trust

Our Agentic Analytics solutions are for teams that need answers from their data faster than a dashboard can give them. We build BI agents tailored to your business, that query and model your data, then talk you through the answer within minutes. The agents work where your team already is, for example right in Slack.

Semantic layerAI agents on your dataForecasting & machine learning
“Behind every trusted answer is a shared definition. A good data agent carries that definition into every question, for every team, because it runs on a unified foundation: one semantic layer, governed business logic and clean data.”
Elena, Lead Analytics & AI Solutions at Gemma Analytics
Ask the data
Your expert

Elena

Lead Analytics & AI Solutions
What we do

What we deliver

The three building blocks build on each other. The semantic layer defines what a metric means, agents answer questions on top of it, and forecasting models use the same definitions to look ahead.

01

Semantic layer

We define your metrics once, centrally in dbt, with fixed metrics, dimensions and filters. From then on dashboards, analyses and AI agents all calculate the same way.

Without this foundation an AI would only amplify differing definitions, so we build it first if it is missing. You get a documented semantic layer that every AI query builds on.

That way the same question always gets the same answer. The agent writes no free SQL; it only picks metric, dimension and filter, and the semantic layer turns that into the same query every time. A fixed set of test questions checks that it stays that way.

02

AI agents on your data

Your team asks questions in natural language. AI agents query the data through the semantic layer, model it and explain the answer within minutes, so nobody waits for the next report. Every answer names the metric it used and its definition from the semantic layer, so you can check it.

You get an agent, for example a Slack bot built on Claude, that reaches approved metrics through the MCP server of your BI tool, with a system prompt tailored to your terms. On top come access rules per user group and a set of test questions with expected answers your team can use to keep checking quality.

03

Forecasting & machine learning

Machine learning models predict what comes next, such as demand, conversions or churn. They learn from the same metrics that are defined in the semantic layer, so a forecast of net revenue means exactly the net revenue in your dashboards.

Scores and forecasts go back to where your team already works, for example live in the dashboard. The agent can query them too, for instance when someone asks what revenue is likely to look like next quarter.

Data agent What an answer looks like
How did revenue per channel develop last quarter?

Revenue grew in three of four channels in Q3, most strongly in the online shop (+15%). Amazon fell from 260 to 240 thousand euros.

Metric: net revenueSource: dbt model
Why did Amazon drop?

Almost the entire drop comes from supplements: 22 thousand euros less than in Q2. The other product groups on Amazon are stable.

Filter: channel AmazonMetric: net revenue
How did supplements on Amazon develop week by week?

Up to week 31 weekly revenue was stable at around 6 thousand euros. Since week 32 it has stayed at about 3.5 thousand. Shall I check prices and stock levels for that period?

Filter: Amazon, supplementsDefinition from the semantic layer
Ask your data...

Who may query what

Analysts may go deeper into the raw data, everyone else asks through the defined metrics.

Semantic layerMCP serverAccess per user groupTest questions

With conversational BI you ask your data a question in plain language, for example in Slack, and get an answer with the metric and a short explanation.

A dashboard shows views someone defined in advance. An agent also answers the questions nobody thought of when the dashboard was built.

For the answers to be right, it needs a semantic layer underneath. Metrics, dimensions and filters are defined there once, and the agent queries only these definitions, never raw tables. Without that foundation every answer calculates a little differently.

It does not replace dashboards. They stay the right place for recurring reports, and the agent takes the ad hoc questions in between.

A semantic layer sits between your data warehouse and everything that shows or queries numbers. It defines once what a metric means, for example how revenue, margin or an active customer is calculated, and which dimensions and filters you can analyse it by.

Dashboards, analyses and AI agents use these definitions and write no SQL of their own. Every analysis calculates the same way, and a change to a definition applies everywhere.

We usually build the semantic layer in dbt and make it available to dashboards and agents through the BI tool, for example Lightdash.

Case studies

Agentic analytics in practice

Three clients, three ways AI works on their data.

Glow25

A scalable modern data stack

A semantic layer plus an AI interface: teams ask questions in natural language and get accurate answers from governed data.

Full transparency, faster decisions
Explore the case study
VALUE AG

Accurately predicting future demand and optimizing operations

A machine-learning forecast takes trends, seasonality, product mix and external factors into account to right-size staffing.

90% forecast accuracy, up to 50% less mis-staffing
Explore the case study
Smoobu

AI-powered conversion predictions for faster, smarter growth

The model learns from historical and behavioural data and finds new patterns without static rules.

Forecast error cut by 50%
Explore the case study
Typical problems

Typical problems we solve

Does any of this sound familiar? Then a conversation about AI on your data is worth it.

Every question waits for an analyst

Your team is the bottleneck: more requests than capacity, and decisions left waiting on dashboards.

Business teams cannot analyse on their own

Business teams cannot analyse data independently, so fast, independent decisions are out of reach.

Every team asks the AI on its own

Teams try AI on different data sources, and in the end different answers to the same question are going around.

The built-in BI agent falls short

The chat in the BI tool returns empty or wrong answers, for example because it assumes the wrong year.

One metric, several definitions

Without a semantic layer every dashboard calculates differently, and an AI would only amplify that.

Forecasts run on rules of thumb

Planning with manual estimates and static rules misses seasonality and real-time trends.

How we work

From discovery call to agents in production

We work hands-on and focus on results, in four steps.

01

Discovery call

We spend 30 to 60 minutes on your challenges and on the questions your teams want answered fast.

02

Check the data foundation

Agents need governed, reliable data. We check what is there and what has to be modelled first. If there is no semantic layer, we build it first. The AI only queries defined metrics, never the raw tables.

03

Build agents & models

We build the AI interface, the agents and the machine-learning models, and align each step with your team. We test with a fixed set of typical questions and expected answers and, where useful, compare with the built-in agent of your BI tool.

04

Production

Answers and scores are available in daily work, for example live in the dashboard.

Tools & technologies

Technologies behind agentic analytics

We work with these tools, among others. Which ones fit depends on your setup.

dbt logoLightdash logoSnowflake logoGoogle BigQuery logoPython logoClaude Code logo
Trusted by

70+ successful projects in six years, a senior team of 20 based in Berlin.

Viessmann logoVaillant logoyfood logoFlaschenpost logoglow25 logoheyData logoaudibene logoHomeToGo logoManiko logoWikinger Reisen logoasphaltgold logoPets Deli logoVALUE AG logomotel a mio logoju:it logoInno Nature logoMocMakler logon. logoSmoobu logoTasty Urban logoAUXEUM logohellomed logoNatsana logoSwappie logowholey logoFOND OF logoCleanwatts logoViessmann logoVaillant logoyfood logoFlaschenpost logoglow25 logoheyData logoaudibene logoHomeToGo logoManiko logoWikinger Reisen logoasphaltgold logoPets Deli logoVALUE AG logomotel a mio logoju:it logoInno Nature logoMocMakler logon. logoSmoobu logoTasty Urban logoAUXEUM logohellomed logoNatsana logoSwappie logowholey logoFOND OF logoCleanwatts logo
FAQ

Frequently asked questions about agentic analytics

It depends on the scope. The main cost drivers are the number of data sources, the questions the agents should answer, whether a semantic layer and clean data models already exist, whether your BI tool already ships an MCP server, like Lightdash, whether the agent should run in Slack or in the BI tool, and whether machine-learning models are part of the project. In a free discovery call we clarify your needs and outline what a project could look like. For eligible SMEs, BAFA funding can cover part of the consulting costs.

No. You ask in plain language, and the agent works through the defined metrics in the semantic layer. You do not need to write SQL or build dashboards for it.

How long it takes depends mainly on how ready your data is (existing models and a semantic layer), how many data sources and questions the agents cover, and whether forecasting models are included. A typical agentic analytics project with us takes six to ten weeks. We work out how long your project takes in the discovery call.

Yes. Our clients have between 30 and 15,000 employees, mostly 50 to 200. Agentic analytics fits wherever teams ask more questions than your analysts can answer, whatever the company size. For SMEs, BAFA funding can apply on top.

The agents query through a semantic layer with predefined metrics, dimensions and filters. They do not write raw SQL themselves. In our experience, how good the answers are depends mostly on the instructions given to the agent and less on the model. That is why we follow up with fixed test questions.

BI & Data Analytics delivers dashboards, reporting and deeper analysis on a unified foundation. Agentic Analytics builds on that: AI agents query your data and talk you through the answer, within minutes.

To BI & Data Analytics

Agentic AI means an AI agent works through a task in several steps on its own instead of giving a single answer. In data analytics the agent queries your data through the semantic layer, models it and explains the result with the metric and its definition. That is what we mean by agentic analytics.

Yes. Forecasting and machine learning are part of agentic analytics for us, which is what many call advanced analytics or data science consulting. Our models predict demand, conversions or churn, for example, and use the same metrics as your dashboards. The forecasts for VALUE AG and Smoobu are two examples.

Possibly, for the consulting part of an agentic analytics project. Gemma Analytics is registered with BAFA as a consulting company. For eligible SMEs, the programme covers up to 50% (up to 80% in the new federal states) of eligible consulting costs as a non-repayable grant, up to a maximum amount. The application has to be approved before the project starts, and we walk you through it at no charge.

More about BAFA funding
Competencies

More ways we can help

BI & Data Analytics KPIs, metric definitions and analysis structure on one unified foundation, so every team works from the same numbers. View service
Data Apps & Infrastructure The infrastructure, pipelines and integrations that analytics and AI workloads run reliably on. View service AI Consulting Your route to AI-first. We work out where you stand, back the use cases that pay off fastest, then set up governance for you to scale. View service Intelligent Automation AI agents and automations that take manual work off your team and run entire workflows. View service AI Adoption We make AI part of how your organisation works. Picking the tools is quick; we help you use these tools confidently and sustainably. View service
Contact

Ready for answers in minutes?

Book a free discovery call: 30 to 60 minutes, focused on your challenges.

Bijan, Founder & Managing Director at Gemma Analytics
Your contact

Bijan Soltani

Founder & Managing Director

Bijan and the Gemma team look at your data and your questions with you. In the call we also check whether your project qualifies for BAFA funding.