Product Pulse Update: June 2023

Major update to Mavenoid’s proprietary AI model: now even more fine-tuned for hardware support than before

Since the early days of Mavenoid, we have always invested significant time and effort in perfecting our proprietary Machine Learning model.  It is the “heart” of our product, powering innovative functionalities like:

• Intent Recognition: an intelligent search field that understands customer queries and guides them to the best solution for their problem;
• Gap Analysis: a tool that detects and helps fix the most common and critical search gaps in self-service flows;
• Impact Analysis: packaged insights allowing to identify topics to prioritize for coverage with self-service support flows;
• ...and more areas where understanding what customers say is crucial for successful self-service support request resolution.

In June, we released a significant upgrade to our model. What is new?

1. We retrained it on a cleaner and more extensive dataset of hardware-specific conversations, improving its understanding of terminology and phrases commonly used to describe issues with consumer electronics, home appliances, EV chargers, smart home devices, industrial machinery, micromobility, and other verticals we’re serving.
2. Now semantic search works with exact matches in one fine-tuned LLM enabling accurate handling of both vague (i.e. “my coffee is too weak”) and specific (i.e. “error code A4563 for product MX678”) requests, which is especially relevant for hardware self-service support efficiency.
3. The model provides more relevant results for medium- and high-complexity searches (e.g. things technicians would search for).
4. Ranking of search matches was improved so the most relevant suggestion is more likely to be offered to a customer, and ranks higher on the list.

The first results are very promising for all the key metrics:

• self-service resolution rate increased platform-wide, with some brands seeing a +6.3% increase,
• search usage increased, indicating that customers noticing the increased relevance and performance of the search field  and are using it more frequently than before
• search match selection rate increased as well, meaning that intent recognition became more accurate and the results provided by Mavenoid search are more relevant leading to faster and more efficient resolutions

Mavenoid’s Machine Learning team measured the new model’s search performance against OpenAI’s LLM for hardware queries, and saw a relative Recall@5 gain (proportion of relevant items in the top-5 results) of +25% compared against OpenAI.

This new model will soon power even more Mavenoid features, bringing you faster and smarter self-service support, so stay tuned for updates!

Advanced Canvas Search: helping you manage high volumes of self-service content efficiently

We’re introducing more AI and automation to generate and update self-service content to ensure all hardware-related customer questions can be answered. This means the volume of content brands have on their average Mavenoid canvas is growing too, and it needs to be manageable; otherwise, content updates will become unreasonably time-consuming. That's why we enhanced our Flow Canvas search, enabling content creators to locate the required nodes swiftly and immediately edit them. It supports:

• Filters by node type, e.g. specifically finding symptoms or solutions that match a search word or phrase;
• Filters by node settings, e.g. searching by nodes containing a form or a step-by-step guide;
• Filters for reviewing recent updates or changes made by a specific colleague(s)—great for efficient collaboration:

• "last modified", allowing you to find nodes edited in the last 24 hours, 7 days, 30 days, 90 days or 12 months
• "modified by", where you can choose to see nodes edited by one or multiple colleagues

• These filters can be used together, and they can even be used outisde of search context, for example, to find all nodes matching a specific criterion their combination

Honorable mentions

Improved editing of the Choice List forms

Creating and editing static and dynamic choice list form fields has become much easier. We introduced a simple form for managing various choice options. You'll have - no pun intended - a choice of:

• "Static choice list": simple form that uses pre-defined options;
• "Dynamic choice list": advanced form, where options dynamically change based on previous customer steps or data fetched from 3rd party systems like your ticketing platform or CRM.

This enables an elegant and flexible way to present customers with form options in the Product Assistant.

AI Copyeditor is now multi-lingual

Teams that manage self-service flows in their native language (e.g. German or Italian), can now benefit from the smart AI rewrite too! It ‘speaks’ the most popular languages that we support, and guarantees good writing quality.

Did you know that...

Mavenoid Assistant can be used to capture, qualify, and store sales leads! Notable examples include:

‍1. An industrial equipment manufacturer uses their Product Assistant to run a detailed "services advisor" program. Once a customer has narrowed down their needs, they can submit detailed specs and request a quote (e.g., for safety modernization specific to their equipment). The lead is then emailed to the sales team's inbox and automatically saved to their proprietary CRM.Previously, this process would have been carried out through a series of back-and-forth emails with the sales team, which was much slower and required significant effort from both employees and the customer.

‍2. An EV charger manufacturer uses the Product Assistant to help customers select a charger from options available in their region, matching their specific use case. It then saves a structured lead in Salesforce and aids the team in providing a personalized quote.All of the above was implemented live with zero coding effort, requiring just a few hours of planning, configuration, and testing.

‍3. A manufacturer of 3D printers for dentistry uses Mavenoid’s Product Assistant to schedule sales demos and additional training. These requests are labeled, saved in Salesforce, and routed to the most relevant team (determined by region and skills), that can best assist the customer with their needs.Interested in setting up a similar scenario? Speak with your Mavenoid Customer Success Manager, and we'll be happy to assist you!

Fixes and polishes

• There was an issue where all translations were shown to the translator if they went directly to the component content in our Translation Management System. Now it's fixed.
• We have fixed a glitch when the Product Assistant wasn’t loading on old Safari browsers. Now everything works as expected.
• There was an issue with automated Zendesk import, when {{ syntax clashed with Mavenoid's own syntax. This is now fixed.
• The issue of panel values not refreshing when toggling between two different Live Support nodes has also been resolved.
• Previously, product manuals would not load while navigating Salesforce-powered apps. This issue has now been rectified, and they load properly!
• Previously, the performance of translation key filters was suboptimal when filtering by multiple translation statuses simultaneously. We enhanced their performance, ensuring they now operate much faster.

We welcome your ideas on how to improve Mavenoid!

Use this form to submit your pain points, ideas, and feedback straight to our product team. As always, we are all ears and want to hear your thoughts!

Thank you and stay tuned for more updates ;)