Dr SWOOP - Responsible AI and Security
What is Dr SWOOP?
Dr SWOOP is a Generative AI bot that helps people use SWOOP better and improve communication and collaboration. It focuses on content that relates to SWOOP Analytics such as our support articles and content posted on our website such as blog posts, case studies and benchmarking reports.
While Dr SWOOP will strive to provide accurate and helpful information, it is AI and not infallible. Dr SWOOP's responses are based on patterns in data and might not always be correct or applicable to every situation. It's always a good idea to cross-check important facts and consider human judgment when making decisions based on the information it provides.
How does it work?
The underlying AI model used in our chatbot is currently GPT-4o-mini, developed and pre-trained by Microsoft Azure OpenAI. As users of this model, we do not train or retrain it ourselves as the training was conducted by Azure OpenAI on a broad dataset that includes publicly available and licensed information up to a certain cutoff date. However, Dr SWOOP is enhanced through a technique called Retrieval-Augmented Generation (RAG). This means when you ask a question, the Dr SWOOP retrieves relevant documents or data from our custom knowledge base we maintain (such as internal files, FAQs, or structured content we’ve provided), and uses that context to generate more accurate and specific answers.
So to clarify:
• The AI model itself is not trained or modified by us.
• We continuously update the content used by the RAG system, which allows the chatbot to answer questions using our latest information without changing the base model.
There are two versions of Dr SWOOP:
If you see the 'insights icon' next to a report, then you have the pro version. The insights icon shows two small sparkly stars and will appear in the report header like this:
Dr SWOOP Pro
The pro version of Dr SWOOP is built with Microsoft Azure OpenAI and you choose which regional data center you would like it to run. Microsoft currently only offers the GPT model Dr SWOOP is using in the US and Sweden. We add more regional hosting options as they become available from Microsoft.
Dr SWOOP Pro is initiated in one of the following ways:
- A person launches Dr SWOOP by clicking on the owl icon, and enters a chat prompt, for example "What is the best time to post a news article on our intranet?"
The SWOOP Analytics Engine AI Endpoint sends the chat prompt to the Azure OpenAI endpoint. - A person clicks the Dr SWOOP insights icon which triggers a chat prompt to be sent to Dr SWOOP without the need for the person to type anything.
The SWOOP Analytics Engine AI Endpoint retrieves the customer data included in the relevant SWOOP report, and a custom chat prompt is sent to the Azure OpenAI endpoint with pre-defined instructions that are unique for the each insights icon.
Answers generated by Dr SWOOP Pro
In both instances, the Azure OpenAI endpoint passes the chat prompt to a search index (RAG), and a Large Language Model returns to the Azure OpenAI Endpoint the relevant answer. The Analytics Engine AI Endpoint delivers the reply to the person in the Dr SWOOP interface.
Dr SWOOP - Standard
In the standard version Dr SWOOP doesn’t know who you are and doesn’t have access to any of your SWOOP Analytics data.
You can use it like you use a search engine and ask general questions like "What is the Mention Index in SWOOP?". If you give it some clues, Dr SWOOP could also recommend ideas for improvement. For example, you could ask:
- “My community’s Curiosity Index is 5%. Is that good? What can I do to improve?”.
- “What are your top recommendations for creating a vibrant community?”
- “What insights from SWOOP can I use to make my intranet better?”
Data Protection and Privacy for Dr SWOOP (Standard and Pro)
Dr SWOOP is hosted in the Azure region selected by the customer (currently either US or Sweden, but as Microsoft extends the regional availability, additional regions will be able to be selected).
For privacy and security refer to Microsoft's support documentation that highlights the following:
"Your prompts (inputs) and completions (outputs), your embeddings, and your training data:
- are NOT available to other customers.
- are NOT available to OpenAI.
- are NOT used to improve OpenAI models.
- are NOT used to train, retrain, or improve Azure OpenAI Service foundation models.
- are NOT used to improve any Microsoft or 3rd party products or services without your permission or instruction.
- Your fine-tuned Azure OpenAI models are available exclusively for your use.
The Azure OpenAI Service is operated by Microsoft as an Azure service; Microsoft hosts the OpenAI models in Microsoft’s Azure environment and the Service does NOT interact with any services operated by OpenAI (e.g. ChatGPT, or the OpenAI API)."
This means that no other SWOOP customers, or Microsoft, or Microsoft's customers, will become aware of the data Dr SWOOP is analysing.
As a customer you own every prompt submitted to Dr SWOOP and the full text of every response it receives. SWOOP retains any pre-existing intellectual-property rights in portions of a response that reproduce SWOOP-authored material and we grant you as the customer a perpetual, worldwide, royalty-free licence to use those portions as part of the conversation content for your internal business purposes or as required by law or regulation.
For quality assurance, security monitoring, and to provide customers with visibility into end-user usage, SWOOP will store an encrypted copy of each prompt submitted to Dr SWOOP and the corresponding response. SWOOP will:
- retain them for up to two years;
- restrict access to authorised SWOOP support staff and Customer-designated tenant administrators;
- use them solely for (i) troubleshooting, (ii) measuring service accuracy and performance, and (iii) providing Customer with usage analytics; and
- delete them earlier upon Customer’s written request or upon termination of the subscription.
Download the Dr SWOOP Technical Overview from our Trust Center to learn about the architectural design of Dr SWOOP.
Considerations and actions taken in the development of Dr SWOOP
Privacy
Objective: Ensure that the content you use for training respects the privacy and consent of individuals mentioned in the data.
Actions taken: All the data that is made available for the bot to be trained on has been approved by authorised employees in SWOOP Analytics, or is already included in the customer's SWOOP dashboard.
Bias
Objective: Be mindful of potential biases in your content. If your data reflects biases, the model may inherit and perpetuate them. Strive for fairness and inclusivity in your training data.
Actions taken: Our content is a mix of benchmarking results and commentary on insights derived from them. The benchmarking results make use of large datasets inclusive of all employees that have been active on the platforms we are analysing. The results are statistically based and therefore less prone to bias. Also, the benchmarking data does not include data about sex, gender, religion, age or other attributes that in other applications of AI could involve a bias risk.
Our commentary and case studies reflect our collective experience of “what good looks like”. It will reflect our own biases for what “good practice” looks like, and our selected case studies will reflect this. Given the intent of our benchmarking, we don’t believe such a bias is harmful; in fact quite the opposite.
Transparency
Objective: Clearly communicate to users that they are interacting with a bot and explain the limitations of its knowledge and capabilities.
Actions taken: The bot introduces itself as a bot, and it explains what material it has been trained on. Some of the help desk articles Dr SWOOP draws upon have been written by the help of AI, but all content has undergone human review and approval prior to publication.
Reliability
Objective: Ensure that the information provided by your bot is accurate and reliable. Misinformation can have serious consequences, so be cautious about the quality of the content you use for training.
Actions taken: Only published material is provided to the bot. This material has undergone peer and editing reviews. We have created a separate standalone ‘what good looks like’ table for all products, so this can be updated independently and instructed the bot to always look at this source when referring to or using goals/targets. Dr SWOOP is instructed to only answer questions based on the available material, and will therefore not use general content sources from the internet in its responses.
Temperature settings have been set to be deterministic rather than creative.
K/P settings ('Top K' and 'Top P') have been set such what we avoid expansion of sources. These settings control accuracy in the selection of sources upon which the prompt outputs are based.
Through meta-tags we ensure that product-specific sources are used only in prompts outputs that are relevant for the SWOOP product in which Dr SWOOP is used.
Semantic re-ranking is used to guide the AI towards the most relevant sources.
SWOOP employees are regularly reviewing content sources to ensure relevancy and accuracy.
User well-being
Objective: Prioritize the well-being of users. Avoid engaging in activities that may harm users emotionally or psychologically. Monitor the interactions and be ready to intervene if the bot is being used inappropriately.
Actions taken: Bot is instructed to only answer questions about the material it is trained on. We review the reports to see which questions are being asked.
Legal
Objective: Ensure that your bot complies with relevant laws and regulations, especially those related to data protection, privacy, and online behavior.
Actions taken: The use of SWOOP's products is tied to our contractal commitments which include data protection and privacy.
Monitoring and maintenance
Objective: Regularly monitor and evaluate the performance of your bot. Implement mechanisms for users to provide feedback, and use that feedback to improve the bot's behavior.
Actions taken: Enable tusers to submit feedback via our support portal available in the window where the chatbot is embedded. We review reports to see which questions are being asked.
Openness and Accountability
Objective: Be transparent about your development process, and be ready to take responsibility for the actions of your bot. Provide clear channels for users to seek clarification or express concerns.
Actions taken: We publish a support article (this page) for how the bot is developed and used. This is linked to the bot interface. Also, we provide a link to our support portal where people can provide feedback or raise issues.
Your Responsibilities
You, as the user, are responsible for ensuring that your use of Dr SWOOP complies with the policies and guidelines set forth by your organisation, including but not limited to those related to data privacy, security, and the use of artificial intelligence tools. It is your duty to be informed about and adhere to any restrictions your organisation may have on the use of external AI tools.
You must not input, upload, or otherwise provide access to data that is confidential or proprietary to your organisation unless you are authorised to do so and such actions comply with your organisation's policies. You are responsible for ensuring that the information you provide does not infringe upon the privacy rights or intellectual property rights of others.
SWOOP Analytics owns the intellectual property rights for the answers generated by Dr SWOOP. As per our contractual terms, SWOOP Analytics will defend our customers against all claims made against a customer by any third party alleging that the customer’s use of the software under the terms of the agreement infringes the intellectual property rights of that third party
Enabling and Disabling Dr SWOOP
Given that organisations are still formulating their policies for the use of AI we have decided that Dr SWOOP will be disabled as a default. That means that until a SWOOP admin enables it, it will not be visible.
Steps to enable Dr SWOOP (must be performed by a person who is a SWOOP admin)
- Log on to SWOOP.
- Click Admin / Dashboard and locate the Dr SWOOP settings.
- Select "Show" to enable everyone who can access SWOOP to also access Dr SWOOP. Alternatively, select "Admin" to enable Dr SWOOP only to people who are SWOOP admins.
When Dr SWOOP is enabled you will see this floating icon in the lower right corner:
If you have enabled Dr SWOOP, but the chat interface doesn't appear when you click on the icon it is most likely due to an ad-blocker or similar technology that is preventing Dr SWOOP from loading. Please contact your local IT support to have this resolved.