> For the complete documentation index, see [llms.txt](https://docs.flowlity.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.flowlity.com/concepts/product-similarities.md).

# Product similarities

When a product is new, has sparse demand history, or is replacing another product, forecasting from its own data alone may not be enough. **Product similarities** let you link related products together so that their demand history feeds into the forecast of the target product — giving the engine more data to work with and producing better predictions.

### Why it matters

Accurate forecasting depends on having enough historical demand. In practice, many products lack this:

* **New product launches** — No past demand exists yet.
* **Product replacements** — The old product's history is highly relevant to the new one.
* **Seasonal or slow-moving items** — Limited data points make patterns hard to detect.

By linking a similar product, you transfer its demand history to the target product. The Forecast Agent then uses this enriched history alongside the product's own data to generate a more reliable probabilistic forecast.

### How similar products are linked

There are two ways to establish product similarities in Flowlity:

#### 1. AI recommendations (Product Similarity Agent)

The **Product Similarity Agent** automatically identifies the most similar products for each item in your catalog. It uses three complementary distance metrics, configured at the site level:

* **Category-based similarity** — Compares categorical attributes such as product name, description, or tag values. Products that share the same attributes are considered similar.
* **Text similarity** — Measures semantic closeness between text fields (name, description) using language models. For example, "red sports car" and "crimson racing vehicle" would score highly.
* **Numerical similarity** — Compares numerical values (e.g. average price) within a configurable threshold. A **price penalty threshold** defines the acceptable range — products outside this range receive a lower similarity score.

The agent combines these metrics and returns up to **5 recommended products** per item, each with a **similarity coefficient** (displayed as a percentage). Only recommendations above 30% similarity are shown.

The agent runs weekly and updates recommendations automatically as your catalog and data evolve.

#### 2. Manual linking

You can also add similar products manually from the **Demand module** using the **Product strategy** sidebar. This gives you full control and is especially useful for product replacements.

When linking products manually, you can configure:

* **Source product** — The product whose demand history will be used.
* **Coefficient** — A percentage that controls how much of the source's demand is transferred. For example, if two products are being consolidated into one, you might set 50% for each.
* **End date** (optional) — Makes the link active only until a specific date. Useful when a replacement is temporary or when the new product has accumulated enough of its own history.
* **Customer mapping** (optional) — If your site uses the customer axis, you can map source and target customers to control which demand streams are transferred.

### Viewing similar products demand

In the **Demand module**, you can visualize the impact of product similarities directly in the chart:

* Open **Table settings** (right-hand side of the screen).
* Enable the **Similar products past demand** line.

This line shows the demand history that is being contributed by linked similar products. It appears alongside **Cleaned past demand** (what the model uses) and **Raw past demand** (the unmodified signal from your data), giving you full transparency into what feeds the forecast.

### Current product ratio

Each product has a **current product ratio** setting (default 100%) that controls the balance between the product's own demand and the demand contributed by its similar products. Lowering this value gives more weight to the similar products' history.

### Where to find it

| Location                                     | What you can do                                                                                                                                                                                              |
| -------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ |
| **Demand module > Product strategy sidebar** | Add, edit, or remove similar products for a specific item. View AI recommendations.                                                                                                                          |
| **Agents module > Product Similarity Agent** | See the agent's configuration: which similarity metrics are active, which attributes are analyzed, and the price penalty threshold. Use the "Try it out" feature to preview recommendations for any product. |
| **Demand module > Table settings**           | Toggle the "Similar products past demand" line in the chart.                                                                                                                                                 |

### Related pages

* [Demand forecasting explained](/concepts/demand-forecasting.md) — How the probabilistic forecasting engine uses similar products alongside other inputs.
* [Forecast events and demand adjustments](/concepts/forecast-events.md) — Other ways to refine the demand signal.
* [Demand](/modules/demand.md) — Where you view forecasts and access the Product strategy sidebar.
* [Agents](/modules/agents.md) — Monitor the Product Similarity Agent and its configuration.