Your operations generate data. Are you forecasting with it?

GenAI makes timeseries forecasting accessible for every industrial company. Most industrial companies already have the data they need for reliable forecasting. Machines log energy consumption by the second. Sensors track asset behaviour around the clock. Systems record volumes, transactions and operational events continuously.
Yet for most companies, that data sits unused. Not because the use cases are missing, but because timeseries forecasting has traditionally been too complex and too expensive to implement.

Sirris has been researching a new approach based on GenAI time series foundation models. The results show that reliable forecasting is now possible without a data science team. 

What GenAI changes 

Think about what better forecasts would mean for your operations. Across industries, the same challenges appear again and again. Predictive maintenance that flags a failure before it causes downtime. Energy demand forecasting that anticipates consumption peaks before they hit your bill, from CNC machines to wind farms.  

Production or service schedules built on real demand signals instead of gut feeling. Sales and cost estimates grounded in historical patterns rather than spreadsheet guesswork. These are not edge cases. They are daily operational challenges for industrial companies of every size and sector.

Until recently, building the models to address them required months of specialist work that most companies simply could not justify. 

The cost of not forecasting 

A new category of AI models, known as time series foundation models, is changing the cost equation.

Models such as Chronos-bolt and TimesFM are freely available, run on a standard laptop and require no training on your specific data. You provide historical data and receive forecasts directly.

Tasks that traditionally required an ML engineer, such as algorithm selection, feature engineering and model training are no longer needed when using pre-trained foundation models.

Sirris evaluated these models through benchmark experiments and compared them with established industrial forecasting methods. The conclusion is striking: they match, and in several cases outperform, traditional specialist forecasting solutions that require far more development effort.

The barrier is no longer modelling expertise. The main requirement is simply having historical data. 

The second barrier: development effort 

Even with a strong forecasting model, someone still needs to build the software around it. Sirris tested whether Generative AI coding assistants could remove this barrier as well. Using a GenAI coding assistant, we built a complete forecasting proof-of-concept in just seven minutes. 

A task that previously required days of development work was completed almost instantly. 

• No specialist team.
• No ML expertise.
• No expensive licences.
• No months of development. 

What this means in practice 

In practice, the approach applies to many operational challenges across industry.

• Predictive maintenance 
  Anticipate failures before they cause downtime in manufacturing, energy or utilities.
• Energy forecasting 
  Predict consumption peaks and optimise energy procurement or usage.
• Production and capacity planning 
  Align production output with real demand signals.
• Sales and demand forecasting 
  Improve the accuracy of commercial and operational projections.
• Cost estimation 
  Build more reliable operational and project cost models. 

For each of these use cases, the barrier to building a working forecasting baseline has dropped dramatically.

What once required a data science team can now be prototyped by someone with domain knowledge and the right tools, in a fraction of the time and cost.