Tamber Stats
Useful metrics for tracking recommendation performance.
Track how recommendations are really working
Tamber Stats enables measurements that are real time, high resolution, and tightly aligned with the objectives of a recommender system. Where backtesting on historical data can provide an indication of real-world performance, Tamber Stats provides access to the raw results.
To start using Tamber Stats, you can now attach contextual information to the events you track. This information includes 1) the context in which the event occurred, and 2) whether or not this event should be counted as a successful recommendation (i.e. was the item in the event presented to the user as a recommendation).
These values tie the user experience to the abstract interaction data.
Useful metrics
The second step is extracting metrics and analysis. Tamber Stats shows you immediately useful metrics without requiring you to parse and crunch numbers yourself. Metrics like the proportion of events driven by Tamber, the percentage change in events-per-session for users who interact with Tamber suggestions, and where in your app those changes are happening.
Tamber Stats also has filterable charts and graphs for you to explore, as well as visualizations for things that are uniquely useful to recommendations. For example, it is often important to know how the engagement-rate of content tends to change over time. Do items quickly fade from relevance, or does engagement tend to increase gradually and then plateau? This can unilaterally determine a number of major design decisions, and is a case where visualization is especially explanatory.
We are still building these output features, and for now they will be available to customers by request, and as they are developed.
Our objective is to make the world’s content discoverable. We think that the systems for discovery can be made dramatically more useful to people, and that building better measurement tools are an important step to making them work better. Tamber Stats is our solution to this problem, and we are eager to show you what good measurements look like.
Email us at team@tamber.com if you would like to start getting insights from Tamber Stats.
