Skip to main content

​
Video Walkthrough

​
What are ‘AI Answers’ and ‘Digital Twins’?

When Responsibly conducts a ‘Due Diligence’ on a supplier, all relevant information (including webpages, documents, reporting, etc.) on that supplier gets stored in a ‘Digital Twin’ - i.e. a digital representation of that supplier.
  • Read more about the due diligence conducted here
Using that Digital Twin, we can then try to answer the Questions you have asked. This feature draws on the same functionality as human answers, meaning you can score them, extract them to .csv, put them on a dashboard and more.
  • Read more about the ‘Answers’ here.
That means ‘AI answers’ can be used as an alternative to supplier engagement, or in coordination with supplier engagement as an effort to reduce the amount of questions asked, by trying to extract an answer first through AI. Using AI Answers as an engagement alternative: Since the ‘Digital Twin’ stores most of the supplier’s website, reporting and more, it can be a useful way to hunt information from which the supplier often responds ‘Please have a look at our website’. This is especially helpful for larger suppliers, where both public documentation is significant, while at the same time, ability to respond to questionnaires is limited when your spend with that supplier is relatively low. Using AI Answers as an engagement complement: In general, we always recommend you to reduce the amount of questions you ask the supplier directly, and one of the great tools to do so is ‘AI Answers’. To make this work, we recommend asking structued question with specific answer categories. This allows you to attach a score to the answer, and thereby to dynamically avoid applicable questions in data request. Read more about tactics to reduce the supplier burden here. Read more about scoring answers here. Read more about Dynamic Questionnaires here.

​
Getting ad-hoc AI answers on a specific supplier

On any supplier, click the blue AI logo in the bottom right corner of the screen. This will open up the co-pilot drawer where you can interact directly with the supplier’s Digital Twin, ask questions, see answes and the citations they draw upon. This is a great way to experiment with how the functionality works, and how to phrase a question to get the best answers when asking it in bulk across your suppliers. It’s also effective if you want to know more about a specific supplier, get input for an email exchange, or prepare before stepping a supplier meeting. Note that ad-hoc answers are not saved and cannot be scored. Therefore, use the method descriped below:

​
Getting AI-answers in bulk across suppliers

To store, score, download, and analyze AI answers, you need to create the relevant Questions in the ‘Questions’ database.
  • Read more about Questions here.
Once the Questions have been created, you go to the ‘Questions’ page, and click ‘Ask AI’ in the top right corner. When you click ‘Ask AI’ you will need to select the question you want to ask and which supplier to ask it to use the Segments feature.
  • Read more about supplier segments here.
Note that there are rare limits to the amount of AI Answers you can extract, so if you have a lot of suppliers, you can quickly exchaust your allowance. Therefore, we always recommended you test out the AI Answers on a small smaple of test supplier first, before asking it across bigger segments, to ensure you get the answers you want. If in doubt, reach out to your Responsibly representative for more guidenace.
  • Read more about rate limits in the next section.

​
Rate limits

Whenever you get an AI Answer, Responsibly uses Large Language Models (LLMs) that have a small cost component to its usage. Therefore, for bulk AI-answers, we have to limit how many you can ask per year. Your allowance is part of your subscription and can be found under Settings -> Quotas. Note: If a supplier’s digital twin is empty, we do not charge for trying. If you can need more AI Answers, you can easily buy packages of answers at cost effective prices. Reach out to your Responsibly representative and we will help you deign the most effective strategy to extract AI answers in bulk and advise you on how many AI answers you need + associated pricing. Not that the AI Answer functionality also has a cost component related to when the supplier’s ‘Digital Twin’ is preared and upserted. Responsibly covers this cost as part of your main subscription in most cases, however, in certain enterprise contracts with very large number of suppliers, the decision maker on your side may have opted-out in including this in the pricing model. If ‘AI Answers’ are unavailable on your account, it will be shown under Settings -> Quotes when AI Answers are disabled.

​
How it works

‘AI Answers’ use a technology called ‘RAG’ (Retrieve, Augment, Generate). For generating the answer based on the ‘chunks’ extracted, Responsibly selects the best cost-performance performing language model for the purpose that follows our general LLM usage guidelines. If a specific model is desired, get in touch with your Responsibly representative.

​
Trustworthiness of AI Answers

Generally, the AI Answer technology can be considered very accurate, but there are two limitations to be aware of:
  1. Finding the right resources in the digital twin: to navigate the digital effectively, we automatically optimize your ‘Question’ into a ‘Search query’. The Retrieval technology uses top indexing techniques and is effective in navigating a very large body of information. However, it is also probabilistic in nature and is operating within the exact boundaries of the lanugage used in the question and in the source material. If a supplier references the thing you are asking about in the completely unrelated terms, it is not always able to make the connection. That being said, it easily makes connections in related terms and across languages. Therefore, there are better and worse ways to ask a question, and a little experimentation and testing is always recommended before launching it in extremely high volumes. Also note that we only surface the top x results when querying a digital twin, prioritizing recency and fit for the query. For very general questions and very large digital twins, the answers will not be exchaustive.
  2. Generating the right answer based on the resources found: Once the source material has been navigated and citations extracted, RAG uses generative technology (i.e., large language models) to generate the answers based on the sources surfaced. While LLMs have improved massively on this front, there still a small percentage risk of hallucinating responses or being overly generous in connecting information between vage points in the source material.

​
Tips for using AI Answers

Ideas
  • Structured response = analysis
  • Ambiguous questions = ambiguous answers
  • Provide guidance and examples