Data Enrichment: The Missing Ingredient in Time Series Modeling
Explore why data enrichment is crucial for improving time series model accuracy and how to implement it effectively.
For decades, time series modeling was stuck in a narrow paradigm.
Most models were univariate — they focused on one signal at a time, in isolation. They ignored metadata, and they rarely incorporated exogenous variables. This made sense: the modeling tools of the time simply couldn't handle more complexity. Why collect more data when the models couldn't use it?
Today, that's changing — and fast.
At Synthefy, we've developed state-of-the-art models that are multivariate and metadata-aware. That means our models don't just look at one signal at a time. They learn across many variables simultaneously, and can condition on external context — everything from weather patterns to macroeconomic indicators to supply chain events.
But to unlock the power of these models, you need the right data. And that's where data enrichment comes in.
What Is Data Enrichment for Time Series?
Data enrichment is the process of adding valuable, correlated information to your time series datasets — transforming simple signals into richly contextualized data that enables stronger, smarter models.
In the past, a utility company modeling electricity demand might include temperature and time-of-day. Today, enriched models might include:
- Local traffic patterns
- Interest rates and inflation expectations
- Major event calendars or holidays
- Real-time economic indicators or policy changes
These aren't just "nice-to-have" extras. For modern, metadata-aware models, they can be essential features that determine whether a forecast captures subtle patterns — or misses entirely.
Dataset enrichment joins exogenous data with customer data, before being fed into Synthefy's models. The dataset enrichment allows downstream forecasting tasks to perform better.
From Models to Platform
At Synthefy, we believe that modeling and data enrichment must go hand-in-hand. Powerful models are only as good as the data they see. That's why we're building a platform — not just a model.
We're starting with a growing library of public data sources — from financial markets to weather feeds to social signals — and enabling clean, structured integrations with our forecasting stack. Customers will also be able to bring in their own data, or connect their own API keys to services they already rely on.
We're also exploring agentic enrichment tools: systems that can ingest unstructured data like earnings reports or news headlines, and automatically convert them into usable, timestamped metadata for model conditioning.
In short: we're building the connective tissue between the real world and our models.
Enrich your data with social media, ERPs, databases, or even MCP servers, with no code, all inside Synthefy.
Real Results
A customer asked us to model the demand for a core packaged food product.
We first modeled demand using only its own historical sales data — no dataset enrichment. We got an RMSE value of 0.24.
We then used Synthefy to enrich the dataset with prices of related pantry staples often bought together — like flour, rice, sugar, and butter. And when we used Synthefy to model this enriched dataset, we got an RMSE of 0.21.
Why does this matter? This improvement in demand forecasting allowed the customer to optimize their inventory, saving time and money.
The Way Forward
Of course, the appropriate enrichments depend on the vertical being modeled. There's still work to do in identifying where enrichment delivers meaningful gains, and in which domains.
That's why our strategy is to start narrow:
- Focus on high-impact verticals (e.g. energy, finance, healthcare)
- Identify a small set of high-leverage external signals for each
- Build integrations to those signals first — whether through APIs, scrapers, or MCP
- Show our customers how enrichment works by demonstrating it on real POCs
We expect some data to yield big wins. Others may be noisy or redundant. But our early results make it clear: the path to more powerful forecasts runs directly through data enrichment.
From Forecasting to Foresight
With enriched data, Synthefy's models become more than forecasters. They become interpreters — able to contextualize behavior, simulate scenarios, and surface unexpected drivers of change.
Enrichment is not just a technical add-on. It's the foundation of our approach. If time series modeling was once limited by the data it ignored, Synthefy is built to make sure we don't ignore it anymore.
— Team Synthefy
Originally published on Medium
