This article covers common questions regarding the AI features in Arbor (Arbor AI) and is intended to help you understand how Arbor AI works. For more information on Arbor AI, see our dedicated section of the Help Centre on this subject here. You can also find our Arbor AI Subscription and Monthly Free Allowance Policy in this article.
At Arbor, we’re always striving to improve our products using leading technology. We may update this article from time to time as we develop our AI offerings. We encourage you to check this page regularly.
The Administering and Managing Arbor AI Series
This guide is a part of a series. The guides in this series are:
What is Arbor AI?
Arbor AI is a collection of AI-powered features. Our goal is to safely and seamlessly integrate AI throughout Arbor, using it to make your everyday tasks even easier.
You will know when a feature is powered by AI because you will see the Arbor AI icon next to it.
What can Arbor AI do?
You can find out more about the capabilities of each of our AI features in our dedicated Help Centre articles.
We are also testing a wide range of other AI features in Arbor Labs. To find out more about upcoming AI features in Arbor, see our dedicated page on this in the Arbor HQ!
How does Arbor AI work?
When you ask Arbor AI to perform a task, it uses a third-party LLM (Large Language Model) to determine how to find the information you need. Arbor currently makes use of OpenAI’s GPT family of models and Google’s Gemini family of models. These models have knowledge of a set of Arbor MIS tools, through which they can request data.
The Arbor MIS tools are hosted on the Arbor MIS tenant and include an Attendance Tool, Behavioural Tool, Demographics Tool, Student Information Tool and such other tools developed by Arbor. Based on your query, the LLM will request the data it requires via Arbor’s appropriate MIS tools.
Arbor’s MIS tools then check what information the user has permission to access, before retrieving the information needed to complete the task. Arbor’s MIS tools will only provide data that the user already has access to. The LLM will then look at the provided data and return a formatted result to the user.
We have provided an example information flow at the bottom of this article.
How do the AI features process data, including personal data such as student and staff data?
OpenAI and Google have been engaged as Arbor’s subprocessors. This means that, just as Arbor processes personal data on behalf of our customers as data processor under the UK GDPR, OpenAI and Google process that same personal data on Arbor's behalf (and, by extension, on our customers’ behalf as the data controller). They do not process customer data to train or otherwise improve or develop their AI models or other services.
Both OpenAI and Google are contractually required to process data in accordance with all applicable Data Protection Legislation.
We have entered into written Data Processing Addendums with OpenAI and Google that impose equivalent data protection obligations to those contained within Arbor’s contract with you. These include obligations:
to process customer data on Arbor’s documented instructions
regarding confidentiality and the implementation of technical and organisational measures to the standard required by applicable Data Protection Legislation
to assist Arbor in responding to data subject rights requests, security incidents and personal data breaches
regarding the further engagement of subprocessors
to make available information necessary to demonstrate compliance
Overseas transfers of personal data
Under each of Arbor’s Data Processing Addendums with OpenAI and Google, Arbor contractually requires that any transfers of personal data to a country or territory which is not recognised as ensuring adequate protection under the UK GDPR and Data Protection Act 2018 must only take place using a mechanism that enables the lawful transfer of personal data to a third country in accordance with Data Protection Legislation. This may include (a) the UK Extension to EU-US Data Privacy Framework or another valid data protection framework recognised as providing adequate protection under Data Protection Legislation; or (b) the standard contractual clauses for the transfer of personal data to third countries adopted by the EU Commission, as amended by the UK addendum to the EU SCCs issued under the Data Protection Act 2018. Where required by Data Protection Legislation, Arbor has undertaken Transfer Risk Assessment(s) in connection with overseas transfers of personal data processed on our customers’ behalf.
Does Arbor send customer data to any LLM platform to train its services?
No. Arbor contractually prevents OpenAI and Google from using your prompts (inputs) and responses (outputs) to train or otherwise improve or develop their AI models or other services.
Can Arbor AI access all categories of personal data on Arbor or is this limited to certain data fields?
For each user, Arbor AI can only access the data that the user is already permitted to see in Arbor’s MIS. If a user cannot view a particular record, field or category of data in Arbor’s MIS (for example, because their role excludes it) Arbor AI will not view, retrieve or include that data in a response for them either.
For each user, no data will be accessed and processed by the Arbor AI feature unless there is clear right of access for that user to the particular data set.
Who has access to Arbor AI?
Arbor AI is enabled for all school users by default. Users with the 'Head Teacher' business role can manage access to Arbor AI (See the Can Arbor's AI functionality be turned off? section at the bottom of this article for information on how to do this).
How do I control access to Arbor AI?
Please visit our dedicated Help Centre section here on how to manage your school's Arbor AI access.
Access to any data retrieved by Ask Arbor on MAT MIS is also subject to any data sharing agreements that are in place.
Do I have to use Arbor's AI functionality?
No, but having Arbor AI turned on should help you work smarter and more efficiently. Access is fully controllable by the school, and our Help Centre documentation here provides guidance on how to do this.
Does Arbor record my AI usage?
All Arbor AI requests are logged by Arbor. The logs include:
The user making the request
Details of the request, including prompts
The response given by Arbor AI
The "tools" invoked by the Ask Arbor chatbot
Any follow-up requests
Arbor uses these logs, which we will generally only do once they have been aggregated and anonymised, to provide, improve and develop Arbor’s own AI offerings and services, including to diagnose problems. It’s important to note that customer data is not used to train, adapt or develop third-party large language models (LLM), such as OpenAI and Gemini.
Can Arbor AI make decisions autonomously?
No, a user must always initiate any interactions with Arbor AI. Arbor AI will not take action on your behalf. All AI-generated content and/or actions must always be reviewed and checked by the user.
For example, when creating an Arbor AI 'Student Report':
The user must click on 'Student Report' to initiate the report generation.
After the content is generated, the user is shown a preview of the report. The report will not be shared with anyone unless the user decides to do so.
How much do Arbor AI features cost?
Depending on the AI feature, the feature will either be included in an Arbor tier’s pricing or will be an additional paid feature.
To the extent commercially viable, our aim is to offer our AI features at no additional cost, as we believe that they are directly aligned with our mission of transforming how schools work for the better. However, Arbor incurs third-party costs in connection with the use of its AI features, meaning we do need to charge for usage in some cases. Our usage policy for all Arbor AI features can be found here.
Why is US English often used in AI responses?
We use both OpenAI and Google Gemini models to generate responses. These models have been trained primarily in English content from the United States. This means that the responses tend to follow a US English writing style, even when "British English" is specified.
We will continue to monitor developments in LLM (Large Language Model) responses. Should it be possible to specify British English as a default language, we will adopt that.
Does Arbor address the use of AI features in its Terms of Use made available to users?
Yes, Arbor's AI features form part of our Services. You can read more about them in Section 6 of Arbor's Terms of Use.
Is there a significant benefit to using Arbor AI, as opposed to conducting tasks using Arbor's regular features?
The AI features are simply making common tasks easier and more efficient for users. There is nothing being processed or produced by AI that cannot be done using previously existing manual processes.
What mechanism does Arbor rely on to ensure the secure transfer of personal data to countries outside the UK?
Under each of Arbor’s Data Processing Addendums with OpenAI and Google, Arbor contractually requires that any transfers of personal data to a country or territory which is not recognised as ensuring adequate protection under the UK GDPR and the Data Protection Act 2018 must only take place using a solution that enables the lawful transfer of personal data to a third country in accordance with Data Protection Legislation. This may include (a) the UK Extension to EU-US Data Privacy Framework or another valid data protection framework recognised as providing adequate protection under Data Protection Legislation; or (b) the standard contractual clauses for the transfer of personal data to third countries adopted by the EU Commission, as amended by the UK addendum to the EU SCCs issued under the Data Protection Act 2018. Where required by Data Protection Legislation, Arbor has undertaken Transfer Risk Assessment(s) in connection with overseas transfers of personal data processed on customers’ behalf.
Who should we contact for further information or clarification regarding compliance with data privacy laws?
For additional information or clarification regarding GDPR compliance and Arbor's AI features, please contact your Arbor representative directly.
What is the accuracy and margin of error of Arbor AI?
We do not publish a single “accuracy percentage” or fixed margin of error for Ask Arbor. This is consistent with industry leaders such as OpenAI and Google Gemini, as the accuracy of generative and agentic AI systems depends heavily on the quality and clarity of the user’s input, or, in this case, the user’s question or prompt.
Instead, we’ve designed Arbor AI to be assistive, controlled, and auditable, with safeguards that reduce the risk of unfair outputs.
Two of the techniques that we use are RAG (Retrieval Augmented Generation) and having an “agentic” approach, whereby the agent has access to “tools” which have a predetermined set of parameters.
The AI determines which Arbor MIS tools it needs to use to retrieve the requested data. These tools are hosted within the Arbor MIS tenant. Examples of tools include (but are not limited to):
Attendance Tool
Behavioural Tool
Demographics Tool
Student Information Tool
School Data Report Creator (for generating reports)
When using an Arbor AI feature, the user always acts as the “human in the loop”. The user must check any of the generated content and choose whether to act on it. All Arbor AI-generated content can be reviewed and edited before it is applied.
As a principle, Arbor never takes autonomous actions based on LLM-generated content.
Does Arbor AI demonstrate any bias?
Every AI model exhibits certain biases. Ask Arbor is powered by OpenAI and Google Gemini’s models.
You can read more about OpenAI’s biases in this OpenAI help centre article.
You can read about Google Gemini’s biases here.
Arbor continually strives to minimise bias in its prompting and retrieval augmentation techniques through the continual improvement and development of its AI offerings.
Can you provide me with an example information flow?
In the example below, we have asked the Ask Arbor Assistant (the agent) for a student's (Abby Adams) phone number.
User Request:
A user (e.g., a teacher or Head Teacher) submits a query via the Ask Arbor Assistant. For example, they might request a report on pupil progress for a specific class or year group.
-
The LLM identifies the Required Data:
The LLM determines which Arbor MIS tools it needs to use to retrieve the requested data. These tools are hosted within the UK and include systems like:
Attendance Tool
Behavioural Tool
Demographics Tool
Student Information Tool
School Data Report Creator (for generating reports)
-
Querying the Arbor MIS Tools:
The LLM makes structured calls to the appropriate tools.
Only data that the user has permission to access is retrieved.
The LLM does not directly access raw databases; it interacts with Arbor’s standard API and tools.
-
Compiling the Data:
The LLM retrieves and formats the requested information into a readable response for the user, based on the user’s query.
For example, when generating reports, the School Data Report Creator is used to structure the results.
The LLM does not retain the inputs or outputs itself.
-
Delivering the Response:
The LLM presents the data in a user-friendly format with clear headings, structured tables, or bullet points.
The user is given options to refine their request, such as adding more details or generating a PDF.
This approach seeks to ensure a secure, controlled, and efficient way for users to access school data without compromising student privacy or compliance with Data Protection Legislation.
Comments
Article is closed for comments.