If you want to make the most of Answer, here's a set of best practices and actions you can implement on your platform.
Before deployment
Browse documentation
Before embarking on your adventure, we advise you to take a quick look at Answer's documentation. This will give you a better idea of the possibilities.
Testing a set of questions
The first thing to do before deploying Answer to all your users is to test the functionality on a subset of content. To do this, nothing could be simpler: simply apply to join the Answer beta program.
You will be asked to select two or three spaces and a few users who will have access to the preview feature. We advise you to select users who are willing and have a good knowledge of the various contents available on the platform.
Once the beta is activated, you'll be able to run a whole series of tests on your existing content.
💡 Good to know: Types of questions
There are many different types of question, and Answer only answers some of them at the moment. So it's useful to know what does and doesn't work yet.
Answer uses up to 5 blocks of text extracted from your knowledge base of around 250 words to answer users' questions. From here, it's easy to understand the current limitations.
Answer will easily answer a question such as "What is Elium?" if the answer is in your knowledge base (i.e. in one of the blocks extracted by Answer).
Whereas aggregation questions that would require you to gather dozens or hundreds of pieces of information to generate an answer still won't work (example: "Make me a list of all the products that have feature X"), nor will filter or order-type searches (example: "Give me the last 5 articles on topic X").
Note: In our roadmap, we plan to teach Answer to answer this type of question by teaching it to use "classic" search.
A good starting point is to use the keywords used in classic search to generate questions your users might be asking. These are available in the search statistics section of the administration panel.
✍🏻 Note
Please feel free to use the Answer feedback function, even in beta phase. This feedback is fundamental to enable us to improve Answer's search and response algorithm.
Setting up integrations
Answer is available in the Web application, but not only. Slack and Teams are also available. Don't hesitate to activate and test these integrations.
⚠️ Attention
Although we try as far as possible to have functional parity between the integrations and the functionality on our own application, some of them may not be available in all integrations. We're still working on implementing feedback in the Teams integration, for example.
In general
Focus on native content
Overall, native content offers greater flexibility in extracting information, but also in managing feedback afterward.
Although Answer is based on your Elium content and Word, Excel or PDF files (and even videos!), native content still offers significant advantages.
Annotations, for example, are only available on native content. The same applies to expiration dates.
🎯 Roadmap
In the future, it will also be possible to provide feedback at block level for native content. These feedbacks will be assigned directly to the content owners to distribute the workload.
✍🏻 Note
If you have a lot of knowledge in files, don't worry. Answer is indeed designed to extract information from files and use it to generate answers.
Create continuous text
To generate answers, we supply Answer with content blocks. These blocks are represented (for the moment) as continuous text. For example, a list of bullet points will be reformatted as continuous text.
Counter-intuitively, Answer may find it more difficult to extract information if the document is (visually) over-structured, as might sometimes be the case in PowerPoint or Excel. It's worth reformatting complex content into continuous text.
💡 Hints and tips: The importance of titles
Technically, content is indexed in blocks of around 200 words. Each block is accompanied by the content title at the time of indexing. This means that the title is very important to Answer's ability to find the right content blocks to generate a relevant response.
Do it regularly
Handling feedback
Feedback is the basis for improving Answer's ability to answer users' questions. From the administration panel, you have access to all feedback left by your users.
Each piece of feedback contains a range of information, such as the question asked, Answer's response and the sources used to answer the question. This information is at your disposal to help you understand why Answer doesn't answer correctly.
For the moment, improving Answer's ability to respond depends exclusively on your content. So it's your content that you'll be able to modify Answer's responses.
🎯 Roadmap
In the future, the answers themselves will be used to enhance Answer's capabilities.
We're working on the possibility of posting feedback at content block level. This will enable contributors to understand which parts of their content need to be improved.
Finally, in the longer term, feedback will be used as the basis for Answer's learning process.
Export and review all questions
All the questions asked by your users can be exported from the administration panel. This CSV file contains the same information as the feedback.
The difference with feedback is that you have access to the very questions that users haven't given feedback on. We believe it's vital to review all the questions in order to better understand how Answer is used on the platform by your users.
Things you need to know
About tags
For the moment, tags are not taken into account for answer generation, nor for upstream search. Of course, they remain useful for advanced search.
Videos are also taken into account
The videos are transcribed and the text is used by Answer to generate responses. You can therefore upload your onboarding or training videos, for example.
Expired content are used
Whether your content has expired or not, it will be used as a source to generate a response.
🎯 Roadmap
In the future, contributors will be notified when their expired content has been used as a source by Answer.
Other resources
Documentation
Documentation was also available in the Help Center, describing each Answer feature in greater detail 👇🏻
For more technical information
Technical information on how Answer works is available on our blog. Here's a link to the article: https://elium.com/blog/how-we-gave-superpowers-to-our-search-with-chatgpt/