FAQ
What is the AI Mesh?
AI Mesh is a network of different types of AI models, regexes, and filters.
AI models vary in their strengths—some excel at sentiment detection (understanding the tone or emotional context of a document), while others specialize in topic detection (identifying the subject matter of the content).
The decision-making process within an AI mesh follows an if-this-then-that logic, but instead of rigid rules, it operates on probabilities and statistical patterns. For example, if a document contains Personally Identifiable Information (PII), the model may classify it as confidential. However, this classification is not based on simple rules; rather, it emerges from a combination of learned patterns and probabilities derived from vast amounts of data.
How can the AI Mesh be updated/changed?
To accomplish tailoring, to achieve specific use cases, we don't re-train models instead we swap models because we have a large network of smaller models. This modular approach allows us to replace specific components as needed, ensuring adaptability without requiring a full retraining process.
If a model underperforms, we don’t need to ingest user data to retrain it. Instead, we refine the system by introducing alternative models or adjusting filters and topic detectors.
This approach keeps our AI system dynamic while maintaining user privacy—we never use user data to train our AI. Our training is done entirely on in-house data, and we do not engage in AI trading.
Any updates or modifications to the ML model are handled internally by adjusting the mesh of models rather than collecting user inputs. For any customization or changes to the ML model, please reach out to Support.
How can a document or email be manually modified with GV Classification?
The agent software gives a list of the available compliance, classification, and possible categorization tags active within an organization.
When the relevant ones are chosen for the document or email in creation, the software will modify and track that document or email in the future.
If users remove the automatic visual tags, the document continues to keep the metadata tags and the centralized audit log keeps a record of the classification.
How does the ML model help with manual classification?
A staff member classifies a document or email with the classification agent by selecting tags for compliance and classification.
A suggested set of tags is provided based on the actual content of the document or email.
Users can choose to use suggested tags or set completely different ones.
If they select different tags the GV Classification software evaluates its own knowledge of the document and decides to either learn from the new tags or to generate an audit log if it looks like the user has made a mistake.
This allows for training and identifies user errors, but crucially enables the Classification software to learn from expert users.
How does GV Classification teach staff about better data hygiene?
GV Classification interacts with staff while they work on documents and emails. When classifying these, the user is prompted with suggestions and also blocks certain risky actions such as sending internal documents outside the company.
Whenever a staff member is blocked or warned a text summary can be requested that explains why this action was blocked or why the classification is needed. An explanation is also presented on where to find more information and even indicate the relevant internal policy and procedure.
Effectively, data security is reinforced through training as a user works.
Can customers add their own PII definitions from the dashboard/ wizard (more like regex
Yes that will be possible for the customers to configure their regexes.
Why is GV Classification superior to other vendors?
There are a number of areas where this new classification solution is even better. The main advantage is that GV Classification effectively leverages artificial intelligence (AI) and machine learning (ML) to provide superior classification that is industry and business-specific.
Is the ML system trained with customer data, public data or is it just initial base data?
The ML model is based on real business data from organizations. Deployments start with a master model and then it adapts to the customer’ environment.
Does the user have to add the tags each time manually? How much automation can be configured?
AI suggestions are provided to help end-users to apply the correct classification tag. Default auto- labelling is a also an option that user can leverage.
Does Getvisibility save any customer data?
No, file content is never saved. The classification server maintains a registry of file names and their properties but not the content. There is also an anonymization mechanism built into the Classification software that reduces file content to a mathematical number that is used throughout the platform.
Will this affect network or file server performance?
No, the software runs in a throttled manner that controls the rate at which files are scanned. It appears like a normal staff workstation. On the staff laptops or desktops, very lightweight plugins interact with the staff member, and suggest document classifications, and alert the staff member to risky actions.
Does sample data need to be provided?
Usually no data is needed. If he AI Model reports new document types it has not seen before, a small sub-set of sample data (a few hundred files) of these new files maybe required, which are immediately converted to an anonymous descriptive number then used to train and update the model. This will ultimately help in improving the accuracy of classification results. This process uses none of the actual document data.
Can Classification use special keywords to classify documents and email?
Yes, although the ML capability of the platform is used for the majority of decision making in terms of classification and suggestions, Classification can be configured to detect certain important keywords that have a significant impact on the classification of the document.
How does Classification know about file permissions and access rights?
Classification scans the central registry of permissions, users, groups and access rights. It links this information to the files that are found during a scan or that are accessed by staff using their laptops or desktops.
Can the tags that are used in the default UI be adjusted?
Yes, all of the tags in the UI and results screen can be modified, altered or deleted based on what is required. Classification has an exceptional standard model that does not need large modifications. However if significant changes are needed, large sets of documents will need to be shared to enable a very customised ML Model to be built.
How to see who is misclassifying documents?
A high-level report on misclassification is generated. This outlines which document was misclassified, when, on which device and who made the misclassification. It is also possible to explore this information on the management console provided you have the correct access rights and a valid login.
Does the ML Model help to control emails from being sent outside our organization?
Yes, the ML Model features allow correct classification of emails before sending and even adapt for attachments if they are present. Knowing exactly how sensitive an email or attachment is allows the correct level of warning or blocking to happen before the data is sent externally. All of this happens without any need for specialized network hardware such as proxy servers.
Can the Agent work on a laptop that is offline?
Yes, users can classify and tag documents using the same rules as when they are online with the same warnings, blocking of risky activities and help assistance to explain the reasoning for the restrictions.
While the ML classification suggestions are only available when online, the pattern-based suggestions are available offline all the time.
How does the ML improve over time?
Machine Learning is at the core of the platform and can benefit from corrections and normal day-to-day operations of staff to learn and improve over time. Any corrections made by staff that address new types of documents are used to improve the Machine Learning models in a totally anonymized way.
Do the agents and the GV Classification server need to be rolled out at the same time?
Yes, the agent software that runs on laptops and desktops is dependent on the GV Classification server. They can operate for extended periods in offline mode without the server being present but do need to synchronize back to the server. This supports a centralized version, configuration control and comprehensive reporting.
What reports are available, and can custom ones be created?
On the Analytics page of the UI there are out of the box informational widgets and the data from these can be exported in PDF format.
How long do file scans take and how many must I do?
The time it takes to carry out a scan varies depending on the size of the share(s) the software is pointed at. The larger the share the longer the scan will take. 125,000 documents can be classified daily - based on average file size and a single scanning virtual machine with GV Classification software installed. It is recommend that 2-3 scans of a selected file share or file repository be done and checked with the Account Team to ensure the tagging is using in-house standards and the documents are well covered. A useful exercise is to check for any documents that are reported by Classification as low confidence classifications, as these may highlight proprietary documents in the organisation. The Classification system can learn to find them with minimal effort.
How long does it take to install?
The server image can be installed in under 1 hour. Including configuration, the system shouldbe ready to scan file shares within 4 hours. An agent installation takes 1 minute per machine. Once installed the agent connects to the server and becomes active. Note, architecture and sizing vary depending on number of users and file repository size.
Do files need to be scanned if the manual classification agent is used?
No, the automated scanning and classification of files is optional, and the GV Classification agent only needs to be used for newly created documents or emails. However, it is included in the base license and can automatically scan and classify documents that may not be opened again for a long time. But that might cause issues with passing a compliance audit. It is best practice to set it up and let it run.
Can Classification classify inbound emails?
When users reply or forward emails, they can be forced to classify emails.
Can a custom compliance configuration be applied?
Yes, Classification supports custom compliance configuration.
Is down-grading classification activities by users tracked?
A widget can be configured to display this in the Analytics Dashboard.
Can users be forced to classify documents?
On the dashboard configuration can be set to force users to classify documents.
Can compliance names be customised?
Yes, compliance names can be easily customised in line with company policies and needs.
Can classification level be modified?
Yes, The admin can modify the classification level on the GV Classification management console or on the dashboard.
Can Getvisibility add visual labels/marking to the documents?
Visual labelling/marking is added by GV Classification to emails and documents. This can be in the form of a header/footer or watermark.
Can unclassified emails be blocked before a user sends out the emails?
Yes, unclassified emails can be blocked from being sent out or warn users if they try to send an unclassified email. All activities are recorded in the audit log.
Last updated
Was this helpful?