What's new in November 2019

Search, Sticky Messages, Typo corrections, Model A/B testing, Analytics 2.0 and Scalability.


Type one word, send a whole message.

The new search feature is as straightforward as you would expect. Works on any word that occurs in the Recommendation Studio. No more guessing what that sentence started with, just type the keyword, select and send.

Sticky messages 🔖

Promo, disruptions, or news


Sticky Messages allows you to create a few messages that are always visible for the agent. You can think of announcements for new products, service messages, or meaningful events or incidents.. One-click and sent to the customer.
Sticky Messages can also be scheduled to become visible at a given time.

Sticky messages are managed via de Deepdesk Recommendation Studio. The designated content editor can manage what messages are available to the agents, including a title for quick reading, the message and ordering.

Typo corrections

Autocomplete + Autocorrect = Autoperfect ✨

Autocompletions can now suggest text completions that correct up to 3 typos. Less deleting characters and rewriting your message. Faster typing.

Model A/B testing

We now deploy and serve multiple models in production. This allows us to compare new ML models to confirm improvement and run live experiments.

Analytics 2.0

We’ve completely overhauled the tracking and reporting of our analytics. This allows to accurately measure what is happening and how we perform.

It's all about Characters Assisted

We now track all characters typed, assisted, deleted, cut or any other change on a very detailed level with an accuracy of ~98%. This allows to report our main KPI Characters Assisted on agent level, source, channel or any of the other 15 dimensions.

Make data driven decisions

This also gives insight on a more detailed level like the performance of different recommendation types, or the impact on typing speed. This helps to understand the usage, performance and quality of various features or activities and allows to make data driven decisions for our roadmap.

Scalability 🛠

With 50 million requests per month and hundreds of requests per second at peak hours, a lot of effort is being done to improve server performance and scalability, providing a continued real time experience for agents.

  • Scalability: Central caching using Redis
  • Server performance and stability improvements → Auto scale at a busy monday afternoon, down scale in slow weekends.
  • In-depth monitoring using Sentry and StackDriver Trace Logging in GCP → Insights in what’s happening at server and application level to understand and improve performance
  • Recommendation Studio redactions now load live and faster, with less impact on production server.
  • Reduce server load by 'debouncing' autocomplete requests. Autocomplete skips irrelevant characters to reduce unnecessary requests to the recommendation engine.


Read more in our January 2020 post 🤓

Header image by Jude Beck on Unsplash