Intergrid

The Challenge

As Intergrid prepared to expand beyond Finland, Jaakko Hyppönen, Head of Analytics, faced a sharp rise in forecasting demands across new geographies, each with its own market dynamics, weather patterns, and operational dependencies. Forecasts shape every major decision Intergrid makes: how much heat a customer site will need, how competitors operate their electric boilers, and how power prices will evolve each hour. Errors can directly affect real-world infrastructure, from factory heating systems to large greenhouse operations and municipal networks.

Before Nixtla, Intergrid relied on traditional time series forecasting models that required extensive manual setup, tuning, and validation for each new customer site and market. This approach worked when the company operated in Finland alone, but it became increasingly difficult to sustain as Intergrid expanded into additional European markets.

• · Each new geography required adapting forecasts to different consumption patterns, weather conditions, and market behaviors, creating significant engineering overhead. As a result, before Intergrid could onboard a new customer site, teams often spent several days preparing and validating forecasts to ensure the level of accuracy required to safely control industrial heating systems.
• · To support rapid expansion while maintaining strict safety and reliability standards, Intergrid needed a forecasting solution that was fast to deploy, accurate out of the box, and able to scale seamlessly across new markets with minimal engineering effort.
• · Building a single global forecasting model internally was not a viable option. The data volume, compute requirements, and specialized expertise required far exceeded what a fast-growing company could support. Open-source foundation models offered partial capabilities, but none delivered the combination Intergrid needed, including configurability, ease of integration, and operational reliability across thousands of time series and diverse European markets.

The Solution

TimeGPT delivers fast, accurate, and scalable forecasts across all operations

Intergrid adopted Nixtla's TimeGPT as its unified forecasting engine, replacing traditional models with a scalable, foundation-model-powered workflow. Integration took less than an hour, an immediate improvement over the multiple days typically required to build and tune manual forecasts.

Today, TimeGPT runs continuously across Intergrid's customer network, forecasting heat demand and power market movements on an hourly basis. These forecasts feed directly into Intergrid's optimization platform, guiding decisions on when to run electric boilers, when to draw from thermal storage, and how to position bids across Nordic power markets. With reliable API performance and consistent response times, Intergrid meets hourly market submission deadlines without interruption, ensuring safe and continuous operation of real-world heating infrastructure.

Nixtla's clear documentation and copy-and-paste prompts made it straightforward for the team to get started. Intergrid set up the initial forecasting pipeline by feeding historical data into TimeGPT, generating API calls using guided examples, and connecting the output directly into its existing analytics stack. Once configured, the team sends data to TimeGPT through a simple API request and receives ready-to-use forecasts without maintaining training pipelines, compute infrastructure, or model-serving systems.

As Jaakko puts it, “You can just send a request with your data, and TimeGPT returns a ready-made forecast. It's super painless, and we don't need to set anything up on our end.”

TimeGPT's ability to incorporate exogenous variables, including weather forecasts, historical temperatures, and power market data, produces more context-aware predictions. Its configurability allows Intergrid to adjust forecasting horizons, inputs, and confidence levels, see the impact, and fine-tune performance without rebuilding tailored models for each new geography or customer site. This flexibility replaces the per-market re-engineering that previously slowed expansion.

TimeGPT also improves over time without additional engineering effort. As Nixtla releases new model generations, Intergrid upgrades by changing a configuration parameter, immediately benefiting from accuracy improvements without retraining or re-architecting systems. This reliability and automation give the team the bandwidth to explore new power market products, such as intraday, balancing, and reserve markets, and to evaluate unfamiliar European geographies with minimal operational lift.