What to Look for When Choosing a Forecasting Provider

Whether you're a direct marketer, a utility or a grid operator: when you buy solar, wind or load forecasts, you decide on something that shows up directly in balancing energy costs and trading results. These seven criteria help you compare providers objectively, and at the end we work out when building your own makes sense.

1. Verifiable forecast quality

Every provider quotes good numbers. What matters is whether you can check them yourself: can you retrieve earlier forecasts per calculation time? Is there a trial on your own assets over several weeks? Watch the metric: MAPE is unsuitable for solar and wind because it explodes for values close to zero. Metrics normalised to installed capacity, such as nRMSE and MAE, plus the bias, are more meaningful. Our article on the pitfalls of MAPE explains why.

2. One model per asset, not per region

Regional upscaling is cheap but doesn't reflect your asset. Ask whether the provider takes turbine type, orientation, hub height or customer structure into account and retrains the model on your measurements. Just as important: are curtailments and unavailabilities kept separate so the model doesn't learn from distorted data?

3. Several weather models and continuous updates

No weather model is the best in every situation. Robust forecasts combine several, for example ICON from the German Weather Service and the ECMWF IFS, and align with the latest measurements intraday. Ask about update frequency and horizon: from quarter-hours for intraday trading to several days for procurement.

4. API, portfolio and integration

A forecast only helps if it flows into your systems without manual work. Check for:

• a documented REST API that uses your own asset IDs,
• requests for single sites and the whole portfolio, ideally with a sum column,
• time zone, unit and time resolution of your choice,
• alternative delivery as files via sFTP, email or S3,
• access to historical forecasts and weather data, for example for your data science team's own analyses.

5. One platform for solar, wind, load and heat

If generation and demand forecasts come from different providers, you maintain several interfaces and compare forecasts built on different logic. A single platform simplifies integration, billing and balancing group management.

6. Transparent pricing and a real trial

Reputable providers state their price and let you test before you sign. Check what you are billed for, for example installed capacity, and whether weather models, portfolio requests or updates cost extra.

7. People who understand your business

Forecasting isn't something you set up once. Portfolios change, and so do markets. Find out who helps you when a forecast looks off, and whether the provider takes your requirements into its development.

Build or buy: develop in-house or purchase?

Many energy companies have their own data science teams and wonder whether to build forecasts themselves. A simple total cost of ownership (TCO) calculation over three years for a portfolio of 100 MW solar and wind shows the order of magnitude. The figures are assumptions you should replace with your own. For buying they are deliberately conservative: we use the list price, although the price per MW falls as the portfolio grows.

On top of that comes a factor no table shows: an in-house system usually needs six to twelve months before it is in production and on par with proven systems. Deviation costs keep accruing during that time, and they are often larger than the licence. Our balancing cost calculator shows how much a better forecast saves you.

Building your own can pay off if forecasting is core to your business model, you have an experienced team and a large portfolio: with the assumptions above, running costs only break even at around 260 MW, and that is still at list price. Because the price per MW falls for larger portfolios, the threshold is considerably higher in practice. A middle way is to buy the forecast and focus your team on what only you can do: trading strategy, portfolio management and your own analyses.

Checklist for your provider request

• Can I retrieve earlier forecasts and check them against my measurements?
• Which metrics do you report, and relative to what?
• Is there one model per asset, and how are curtailments handled?
• Which weather models do you use, and how often do you update?
• What do the API, portfolio requests and file delivery look like?
• What does it cost, and what is included?
• Can I test on my own assets beforehand?

At alitiq we answer these questions openly: forecasts for solar, wind and load and heat from one platform, published benchmarks, €18 per MW and month for solar and wind, and a free 30-day trial. Get in touch, and we'll gladly compare against your current system.