# EnateAI - FAQs ## Email Classification **Smarter triage for smoother workflows** Manually sorting a high-volume inbox is slow, inconsistent, and not the best use of time. Email Classification uses AI to automatically categorize incoming emails as they arrive in Enate. It analyses the subject and body of each email against your pre‑defined categories and assigns the most appropriate classification, along with a confidence score. When the confidence score is above your configured threshold, the email is automatically triaged to the correct category so the right team can action it. When the confidence score is lower, the system presents the recommended category to a user, who can review and confirm or change it. This ensures triage remains fast whilst keeping agents in control of edge cases. By reducing manual sorting and minimizing misrouting, Email Classification speeds up responses, improves consistency and removes hours of repetitive admin work from busy teams. **Proven results:** ✅ 900 hours saved per 30,000 emails ✅ 76.6% classification accuracy, delivering results from day 1 The detailed explainer article for EnateAI Email Classification can be found [**here**](/enate-help/enateai/enateai/enateai-for-email/enate-ai-email-classification.md). #### **FAQs** **What does Enate Email Classification actually do?** Email Classification uses AI to read the subject line and body of each incoming email (excluding signatures and footers) and determine which predefined category it best fits into. It then recommends or automatically applies that category based on your chosen confidence threshold. **How does the AI decide which category to use?** The AI receives: * The **email content** (subject + body) * Your **list of categories** The AI is then prompted to determine: *“Which category does this email belong to?”* It then selects the best match and assigns a **confidence score** indicating how sure it is about its decision. **What is the confidence threshold and how does it work?** The confidence threshold determines whether Email Classification auto categorizes an email: * If the confidence is above your threshold, the email is automatically triaged. * If the confidence is below your threshold, the user reviews and confirms the suggested category manually. You can set the threshold anywhere from 0% to 110% and can easily change it at any time. **Can I stop Email Classification from auto categorizing emails at first?** Yes. Setting the confidence threshold to 110% means Email Classification will *never* auto triage. Instead, it will simply recommend a category for each email. This is a great way to build trust before gradually allowing auto triage. **Does the AI use keyword matching?** No — Email Classification is **not** a keyword-based system. For example, seeing the word *“EXPRESS”* in an email does **not** automatically assign it to an EXPRESS category. It analyses the meaning of the entire message, not isolated terms. **Does Email Classification learn from user feedback or retrain on client data?** No. Email Classification does **not** retrain on your data or learn from manual re‑categorizations. This is to ensure full compliance with the EU AI Act and maintain complete data safety for clients. **Why doesn’t Email Classification improve accuracy based on user corrections?** Because we do **not** create a feedback loop or store user emails for training. However, we **do** capture when emails are re‑categorized so you can: * track AI accuracy * identify category naming issues * spot trends that may need process updates **How important are category names for Email Classification?** Very important. If a new human member of your team would struggle to understand the category name, Email Classification will too. To get the best results: * avoid acronyms * remove ambiguity * use clear and descriptive names Our Professional Services team can help you create effective naming conventions. **Does Email Classification use metadata or context about the categories?** Not currently. At the moment, Email Classification only receives the **category names**. We are working on adding richer category context metadata to improve accuracy. **What reporting is available for Email Classification?** We are developing a dedicated Email Classification performance report that will include: * Email volume triaged * Average turnaround time * Correct categorization rates * Re-categorization count * AI performance by confidence bucket * AI processing volume vs. accuracy * Top AI predicted categories * Top user changed categories These insights will help you to fine‑tune category design and confidence thresholds, and to validate ROI for Email Classification in real-time. **What data does Email Classification process?** Email Classification processes only the subject and body of the email — not signatures, footers, or historical data — ensuring the highest level of data privacy and compliance. *** ## **Email Data Extraction** Email Data Extraction uses AI to read incoming emails and automatically pulls out the key information needed to populate custom fields in Enate. It analyses the email body, subject line and the names of the fields in the work item, and extracts the best fit values. The user reviews and confirms the data before creating the ticket, reducing manual effort while maintaining accuracy. Email Data Extraction turns unstructured emails into structured work instantly, helping teams reduce manual data entry, increase speed, minimize errors, and ensure inbound work is picked up and processed quickly. The detailed explainer article for EnateAI Email Data Extraction can be found [**here**](/enate-help/enateai/enateai/enateai-for-email/enateai-email-data-extraction.md). #### **FAQs** **What does Email Data Extraction do?** Email Data Extraction identifies and extracts important information from incoming emails and uses it to prepopulate custom fields in Enate work items. This helps users create structured work without having to manually copy and paste all of the data. **What content does Email Data Extraction analyze?** Email Data Extraction looks at the email body, the subject line and the names of the custom fields that need to be filled. Using this context, it selects the most relevant values to propose and populates the fields. **How does the AI decide what information belongs in each field?** It interprets the intent behind each field name and searches the email for the most relevant matching value. For example, if a field is called Account Number, the AI looks for values in the email that resemble an account number or appear near related context. **Does the user get to review the extracted data?** Yes. Users always review the extracted values before the ticket is created. They can change or correct any field as needed. Human-in-the-loop validation ensures quality and reduces errors. **Can Email Data Extraction read attachments?** No, attachments cannot be read. Email Data extraction works purely on email subject and email body. **What types of work is this useful for?** Email driven work such as service requests, inbound claims, HR enquiries, onboarding documentation, customer updates, invoice-related queries and any operational emails that require structured data. **Does this reduce manual work?** Yes. It eliminates the need for agents to manually read emails and type values into fields. Tickets are created faster and with more complete information. **Is Email Data Extraction a no-code tool?** Yes. It requires no development work. Once enabled, it works automatically with your existing field names and workflows. **How does this help with speed?** By automatically extracting key values, Email Data Extraction speeds up ticket creation and helps teams respond to inbound work faster. **Does it help reduce errors?** Yes. It reduces risks associated with missed or misread emails by pre‑populating fields accurately and ensuring users validate information before submission. **Does the AI rely on keyword matching?** No. It uses natural language processing to understand meaning and context, not just keywords. This allows it to handle varied wording or formats. **Does Email Data Extraction work out of the box?** Yes. There is no training required. Clear and descriptive field names improve extraction accuracy. **Why are good field names important?** Clear field names help the AI understand what information you expect it to find. Fields such as Policy Start Date or Customer Reference Number produce better results than ambiguous or abbreviated labels. **How does Email Data Extraction support SLA performance?** By structuring work automatically and reducing delays caused by manual data entry, inbound work is processed sooner, reducing the likelihood of SLA breaches. *** ## **Sentiment A**nalysis **Instantly understand the tone of every customer email.** Sentiment Analysis uses AI to analyze the **body of each incoming email** and classify it as **positive, neutral, or negative**, along with a confidence score. This helps teams quickly identify frustrated customers, prioritize sensitive cases, and respond with the right level of urgency and care, without the need for manual effort. **Proven results:** ✅ **96% accuracy** across **20,000 emails processed daily** by larger Clients The detailed explainer article for EnateAI Sentiment Analysis can be found [**here**](/enate-help/enateai/enateai/enateai-for-email/sentiment-analysis.md). #### **FAQs** **What does Sentiment Analysis do?** Sentiment Analysis uses AI to read the **body** of an email and determine whether the overall sentiment is **positive**, **neutral**, or **negative**, along with how confident the AI is in that judgement. This gives agents quick insight into customer tone and helps your team to prioritise appropriately. **What parts of the email are analyzed?** The AI analyses only the **body** of the email. It does not use: * Subject lines * Footers or signatures * Attachments * Images or files **How does the AI determine sentiment?** The model evaluates the overall meaning and tone of the message — not just keywords. * Analyses only the latest message (ignores threads, signatures, disclaimers) * Defaults to **Neutral** unless clear emotion is present * **Positive** = explicit praise, satisfaction, or strong gratitude * **Negative** = clear frustration, criticism, or dissatisfaction * Politeness, urgency, and routine business language are Neutral unless paired with emotion **What AI model does Sentiment Analysis use?** Sentiment Analysis is powered by GPT‑4.0 mini, selected because it proved to be: * The most accurate in internal testing * The most cost-effective model to run at scale **Is this a keyword matching system?** No - Sentiment Analysis is **not** keyword based. It interprets the full meaning and tone of the message instead of reacting to individual words. **Does Sentiment Analysis learn from our emails or user corrections?** No. There is no feedback loop and no model retraining on client data. **How accurate is Sentiment Analysis in a real workload?** An example larger client uses Sentiment Analysis on 15,000–20,000 emails per day and achieves 96% accuracy. **Are there any limits on email volume?** No. Sentiment Analysis can analyze emails at any scale, making it suitable for high volume operations. **How does Sentiment Analysis help agents day-to-day, and where does it add the most value?** Sentiment Analysis can be useful on a single email basis, helping agents quickly understand tone and respond appropriately. However, the *real* value emerges when you analyze sentiment trends over time. Sentiment Analysis powers a dedicated report that helps you identify patterns, hotspots and opportunities for service improvement across your operation. This includes: * Sentiment over time - track whether overall customer tone is improving or declining * Sentiment comparison by context - compare sentiment across customers, services and processes * Agents receiving the most positive or negative emails * Email sentiment by ticket categories * Top positive and negative senders * Email sentiment by the length of time a work item is open * Email sentiment by number of work items with defects This level of insight helps teams spot emerging issues early, validate process improvements, and understand the customer experience at scale - not just email by email. *** ## Thank you Email Evaluation **Stop thank you emails from creating unnecessary work** Thank You Email Evaluation prevents simple thank‑you replies from creating unnecessary work for your team When a new email arrives, Enate automatically re‑opens the associated resolved work item. Thank You Email Evaluation then analyses the body of the latest two emails in the chain to determine whether the new message is a genuine thank‑you with no further intent. The AI assigns a confidence score to this assessment. If the confidence is above the configured threshold, the work item is automatically closed again so it does not clutter the queue. If the confidence is below the threshold, the work item remains open for an agent to review and decide on the next steps. Save time, reduce queue noise, prevent unnecessary reopenings. The detailed explainer article for EnateAI Thankyou Email Evaluation can be found [**here**](/enate-help/enateai/enateai/enateai-for-email/enateai-thank-you-email-evaluation.md). **FAQs** **What does Thank you Email Evaluation do?** Thank you Email Evaluation identifies when a new incoming email is a simple thank‑you message. Because Enate always creates a new work item when an email arrives, Thank you Email Evaluation prevents these thank‑you emails from creating unnecessary reopened tickets by automatically closing them when appropriate. **Why is this useful?** Many customers send thank you messages after an issue has been resolved. Without Thank You Email Evaluation, every one of these messages triggers a new work item and forces agents to manually close them again. Thank you Email Evaluation automates this behaviour, keeping queues cleaner and reducing repetitive admin work. **\*\*What content does the AI analyze?\*\*** The AI reads the body of the latest email along with the body of the previous email in the chain. This helps it determine whether the new message is a pure thank‑you or contains a new request. **How does the AI decide whether an email is a genuine thankyou?** The AI checks whether the new email contains only closing language, such as gratitude or sign-off. It also checks the previous email to see whether there was any outstanding request or instruction. If the new email includes new intent such as a question or a follow up action, it will not be treated as a thankyou and will remain open. **What happens when the AI confidence is above my threshold?** When the confidence score meets or exceeds your chosen threshold, Thank you Email Evaluation automatically closes the newly created work item and prevents it from appearing as reopened work in agent queues. **What happens when the confidence score is below the threshold?** If the confidence score falls below the threshold, the work item reopens and moves to the To Do state so that an agent can review the message and decide whether follow up is needed. **Can I start with a high threshold first?** Yes. Many teams begin with a 110% threshold to see which emails the AI would have closed, without allowing it to take action. This helps build trust before lowering the threshold to enable auto closure. **Is the AI keyword based?** No. The AI evaluates intention and meaning, not just keywords. For example, a message that says “Thanks” but includes a request such as “Could you also check X?” will not be treated as a thankyou email. **Is any training required?** No. Thank you Email Evaluation is ready to use immediately with no setup, tuning or training required. *** ## Intelligent Document Processing **Automate document classification and data extraction with AI.** Manually processing documents and data is a slow task that's prone to errors and hard to scale. Enate AI Intelligent Document Processing automates document extraction using pre-built Azure Document Intelligence models. It automatically extracts key fields from incoming documents across multiple file types, languages, and even handwritten text.Specific models offer higher accuracy, while the general model provides flexibility for unstructured documents. High-confidence results are processed instantly whilst lower-confidence results are flagged for a human to review and confirm. Users have full visibility of where the extracted data came from so that they can correct values and submit the correct results. Activate IDP in Marketplace, choose your model in Builder, and you’re ready to go. **Proven results:** ✅ **98.5% accuracy** across **8,000 invoices per month** for one example client The detailed explainer article for EnateAI IDP - Document Extraction can be found [**here**](/enate-help/enateai/enateai/enateai-for-idp-document-extraction.md). #### **FAQs** **How do I set up Intelligent Document Processing?** Start by enabling the IDP adapter in Marketplace. Then in Builder, select the model you want to use. This defines which fields will be extracted. **What models are available?** You can choose from a general model or models for processing specific documents: invoices, identity documents, insurance surrender forms, purchase orders and share structure charts. Each of the specific models includes a set of predefined fields based on its template. **How does extraction work?** Documents are initially processed by Azure Document Intelligence service to extract structured data. This output is then passed to OpenAI to generate a contextual summary of the document. Using both the generated summary and the original document images, OpenAI performs a second pass to extract and refine the required fields, resulting in higher accuracy and better data quality. **How is the confidence score calculated?** Each field receives an individual confidence value. The overall score is the average across all fields. Empty fields receive zero, so they reduce the overall confidence. You set a threshold to determine when a human needs to review the document. **What happens when confidence is below the threshold?** The document is sent to a human reviewer. They can click on any field to see exactly where the data came from on the document, view all pages, update values or add missing details. They can save work as a draft or submit it. **How fast is Intelligent Document Processing?** Intelligent Document Processing typically processes each document in under a minute and can handle multiple documents at the same time. This is faster than many competing IDP tools that take several minutes per document. **What file types and languages are supported?** Intelligent Document Processing supports PDFs and all common image types. It works with documents up to ten pages long, supports twelve languages and can process handwritten content. **How do you create a new model for a client?** If a client needs a custom template, Enate builds a new model, defines the fields and tests it using 50–100 client documents and our AI-generated test data. This can be completed in as little as half a day. **How accurate are the different models?** Specific models are highly accurate because they use fixed templates. The general model handles any document type using dynamic fields, which offers more flexibility but lower accuracy. **Can Intelligent Document Processing extract tables?** Yes, table extraction is supported and works particularly well for our invoice model. However, since table structures can vary significantly across document types, any new document requiring table extraction may need some configuration or code adjustments to ensure optimal accuracy. **What happens after the reviewer submits the output?** The system generates an output file, tags it according to the configuration and populates the extracted values into Enate to continue the workflow. **Is document classification part of Intelligent Document Processing?** Document classification is available - it is a separate adapter that can be switched on independently in Marketplace. Intelligent Document Processing focuses on extraction, while classification handles tagging files based on their content. *** ## AI Analyst High‑volume, logic‑heavy work is time‑consuming, repetitive and prone to human error. AI Analyst automates these manual, logic‑based steps inside Enate workflows by reviewing your structured data files, applying your defined business rules and producing clear, structured outcomes that would normally require manual judgement. It evaluates the input data files against your defined prompt, applies your business rules and provides the outcome automatically where the rules are met. If information is missing, conflicting or unclear, the task is flagged and handed back to a human for review, correction or escalation. AI Analyst is well suited to operational scenarios such as payment reconciliation, invoice matching, supplier checks and other high‑volume activities where consistent interpretation of data is essential. By handling routine analysis and escalating only exceptions, AI Analyst reduces manual effort, improves consistency and allows teams to focus their time on the work that genuinely needs human judgement. The detailed explainer article for EnateAI AI Analyst can be found [**here**](/enate-help/enateai/enateai/enateai-ai-analyst.md). #### **FAQs** **What does AI analyst do?** AI analyst automates tasks that require reviewing data, checking logic or applying business rules. It interprets the information in a work item and returns an outcome that moves the workflow forward without human intervention. **How does AI analyst work?** It reads your prompt and the input data, applies the business policy you define and produces a result as an output file. The output can be a decision, a validation, a match or mismatch, or a structured set of values used to drive the next step in the workflow. **What use cases does AI analyst support?** Common use cases include payment reconciliation, invoice matching, supplier payments and upcoming scenarios such as change of directors processing. In each case, AI analyst compares data, checks accuracy and flags discrepancies for human review. **How does it handle payment reconciliation?** It compares incoming payment data with the relevant invoices or records in the workflow. It identifies matches, mismatches or missing information and only escalates the exceptions that require human investigation. **How does it support invoice matching?** AI analyst checks invoice details, totals, references and supporting information to determine whether an invoice should be matched, approved or escalated. It follows your business rules consistently and highlights discrepancies. **How does it help with supplier payments?** It reviews supplier payment information against your defined rules to confirm accuracy. Any incorrect values, missing details or policy violations are flagged and routed to the right user to resolve. **How do business policies work?** You start with a template policy and adapt it to your own requirements. The policy describes the rules AI analyst should follow and the conditions that determine each outcome. No coding is required. **What happens if the AI cannot complete a task?** If information is missing, inconsistent or unclear, AI analyst assigns the task to a human user. The user can update the data or escalate an issue to the admin team. This ensures the workflow continues without inaccurate decisions. **Does AI analyst require coding or integration work?** No. AI analyst is fully no-code. Users configure the inputs, rules and outputs using standard Enate action settings. **What does the output look like?** AI analyst will provide the output file that you request - when you build the policy, you set rules for it to perform and request the output file. For example, if you create a policy for analyst to perform the merging of two excels and set the output as a single excel file, then the output file will be an excel. Or, if you create a policy instructing AI Analyst to extract 5 fields from a set document and give the result in the txt file then the output file will be txt. **Can AI analyst be used in multiple processes?** Yes. It is process agnostic and can be used in any workflow where decisions depend on data validation, interpretation or logical comparison. **How does AI analyst improve efficiency?** It removes manual interpretation work, ensures consistent rule application, reduces the number of human touchpoints and speeds up end-to-end processing. Humans only handle exceptions rather than routine checks. **How does it support accuracy?** AI analyst applies rules consistently and flags anything unclear. It reduces human error and ensures tasks progress only when data is complete and aligned with the defined logic. **How does AI analyst fit into wider automation strategies?** It serves as a reasoning layer inside your workflows, complementing tools like Data Extraction, Intelligent Document Processing and Email Classification. While other tools capture and classify data, AI analyst interprets that data to drive the workflow intelligently. --- # Agent Instructions: Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter: ``` GET https://docs.enate.net/enate-help/enateai/enateai/enateai-faqs.md?ask= ``` The question should be specific, self-contained, and written in natural language. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.