logo
mobile-logo

FAQ

Explore our FAQ to learn more about how Fisent supports organizations with compliance, automation, and innovation.

Videos

Case Studies

White Papers

Use Cases

Frequently asked questions

Fisent BizAI provides unapparelled accuracy or hallucinations by applying a variety of implementation techniques and design controls when using BizAI to digitize and interpret content.  Some of these techniques include:

  • Actions – Fisent’s proprietary content processing protocols that are specifically designed to achieve the desired content processing outcomes (e.g., what responses/outcomes you want from BizAI)
  • Want/Need/Do – Our implementation methodology that is specifically focused on delivering the exact business outcome needed for each and every use case. This unique approach takes advantage of the transformative power of AI and Automation capabilities to narrowly focus on consistent, accurate, and scalable results.
  • Proprietary research techniques used to strictly guide the LLM’s for the purposes of your exact use case
  • Data Enrichment – (Only if selected by customer) – we can apply “memory” which is used to enhance the LLM outputs at runtime
  • GEF Testing – Fisent’s proprietary testing harness allows us to validate the efficacy (e.g., accuracy, consistency, performance, etc.) of the BizAI outputs during the implementation phase to ensure we have the best model and parameter configuration set-up for each use case. Additionally, we apply GEF to newly released LLM’s to test their performance on each use case and bring recommendations to our customers when/if we see a material improvement in use case efficacy.
  • Finally, BizAI outputs have a high degree of explainability through the rationale that is provided with each BizAI output that displays a brief explanation of the logic behind the generated outputs.

BizAI is a pioneer in the “Intelligent Content Processing” (ICP) solution category. ICP and OCR and IDP are often mistaken as “the same” or similar. While they do share a common purpose – to digitize content processing and enable automation – they are quite different. Here are some specifics as to the key differences business owners should be interested in.

OCR and IDP solutions:
  • Specific Applicability
  1. Fundamentally – OCR & IDP are software models trained to process specific content patterns – to identify and extract (“lift”) targeted content within a limited set of documents – this means:
          • These solutions require static (or near static) format, verbiage, layouts, etc.
          • They are used to identify targeted data/content and digitally extract that verbiage and input it into another software system or wherever you instruct the software to distribute that exact content that was lifted from the original data source
  2. Longer Time-To-Value: these solutions require significant upfront investment for use case application including model training – which is accomplished by storing and using your data/documents to create patterns (e.g., coordinates and key value pairs) allowing the OCR/IDP solution to process data or documents consistently
          • Limited to no adaptability: OCR & IDP rely on consistency – if the format, data, language, structure, etc. change – they most likely need to re-train the software for this new variation (e.g., “123 Main Street” vs. “123 Main St.”)
          • Point/Limited Use Case: Enterprise organizations of all sizes DO NOT control the file type, language, format, level of detail, structure, etc. of the content sent to them by customers, vendors, partners, examiners, regulators, etc.  OCR & IDP, by nature being a trained software solution, cannot natively ingest that content and process it – they will either fail (reject) or send it as an exception for human to review and decision 
          • Reliability: OCR & IDP solutions typically provide a “confidence” score which is used to demonstrate how confident the OCR or IDP solution is that it “identified the right content and lifted the specific data you have trained it to extract”
  • Specific Applicability
  1. Unlimited Applicability: Fundamentally – BizAI is a GenAI-based intelligent content processing solution – it reads and interprets content – understanding not just what the content says, but more importantly, what it means – like a human operator would do when reading or listening to various content sources (e.g., a contract, new account application, dispute letter, Power of Attorney, etc.)
  2. Zero Retention: BizAI DOES NOT train models – any format/any file type/any language can be natively processed – out of the box
  3. AnyModel/Any Host/Any Content: 
    1. BizAI can natively process various modalities – data, documents, images, and audio content sources
    2. BizAI can natively process 40+ languages – fluently – not just the words, but the meaning of the words/phrases/content – like a human operator would do
  4. Time to Value: The fact that BizAI can natively process any structure (e.g., email, handwriting, different languages, different layouts, images, audio, etc.) – means we focus solely on digitizing the content (e.g., ingesting, processing, interpreting, etc.) to drive the exact business outcome you are seeking – we don’t need to train models, train for variations, etc.
  5. Reliability: 
    1. GEF – Fisent’s proprietary “GenAI Efficacy Framework” is a testing harnass that allows us to empirically test each and every use case for accuracy, consistency, performance and other factors – ensuring you know you are applying the best model with the right configuration for your use case(s)
    2. Measurement – BizAI offers a “confidence rating” feature – that assesses the BizAI outputs – both did the service identify the exact content you are requesting – but also (like a human) the validity of the content – for example -“is this the right answer?”

BizAI’s full-optionality capabilities allow for Customers to easily select and apply any model / any host of their choice. This means Customers can utilize the best performant LLM, or one that they are most comfortable with for their use – and BizAI takes care of security, error handling, upgrades, etc. for you. Critical for all companies to consider is how to “future proof” their application of AI given the rapid evolution we are experiencing now. Fisent BizAI leverages, tests, and applies the latest and greatest Large Language Models (LLM’s) as new more capable options are released or as your use case needs evolve. This holds true even if it means using a customer's own model (BizAI supports Bring Your Own Model).

BizAI provides an entirely new level of transparency or auditability transforming your risk posture. As our customers apply BizAI to enable digital content processing – BizAI outputs a “rationale” with all AI generated responses. This means – anyone wanting to know “why did I get this response/answer?” can see what the AI reviewed/considered and ultimately concluded for every AI generated output. This is ground breaking, think about knowledge-based processing today – humans perform it – they don’t write down what they were thinking, what information they compared, what other sources they used to come to a conclusion – BizAI does!! This rationale is stored within a Customer’s application layer, allowing for easy review and auditing. Furthermore, historically, AI technology has been a “black box” where understanding responses required a highly specialized skill set, with BizAI we’ve made sure this is no longer the case.

BizAI has a zero-retention architecture, which means no customer data beyond simple API usage metrics (e.g., processing time, volumes, etc.) are persisted or retained. Fisent does not store customer parameters, outputs or inputs and does not train or fine tune models with customer data. Additionally, the enterprise LLMs that are utilized within BizAI do not train or persist any customer data. In other words, customers can process content without concern of their inputs or outputs being retained in any form. Customer data is only available or stored within their own application layer where BizAI returns the outputs/responses. Adding a little wrinkle in the above FAQ is what we call “memory”. Fisent (ONLY UPON CUSTOMER’S REQUEST) can apply a feature called “memory” to enable “enhanced content processing & retrieval”. BizAI for a specific use case will utilize previously processed outputs to enhance new use case processing at runtime – think of it like actual human memory – “I’ve seen this before – the answer looks something like this…” – again – ONLY AT CUSTOMER’S REQUEST.

Yes, BizAI undergoes regular penetration testing by a third party specialized security organization. Additionally, Fisent invests in multiple layers of test automation that is run daily and includes endpoint testing with a security focus. In addition, the surface area is restricted to a specified set of API endpoints supported by AWS API Gateway, a mature best-in-class security service.

  • BizAI provides model optionality, which means our customers can choose to use models deployed by Fisent or models deployed within their own infrastructure
  • With Fisent deployed models, we are only using enterprise LLM’s that are hosted on major cloud providers such as Amazon, Google, and MS Azure. All of these models are “pre-trained” – allowing us to use them out-of-the-box with no additional training required
  • All of these providers publish similar statements about the security and privacy of the data processed through their models, specifically that no data is retained or used for training purposes

Yes. While BizAI is built to enable full straight-through processing (STP), we understand that additional human validation may be required for some period of time. In fact, we strongly recommend that all of our customers start with HITL prior to moving to STP to ensure the outputs are meeting the needs of the business. HITL is typically built into the application layer where the operators are performing their day-to-day work. This allows a more integrated experience for the user and allows Fisent to gather feedback on the service which we use to tune the AI outputs where appropriate. Once the reliability of BizAI has been sufficiently validated in production, customers have the option to enable full straight-through processing.

Typical BizAI implementations range from 5-7 weeks in length. With accelerated implementation timelines available depending on the customer’s SDLC processes and resource availability. Typical implementation schedule:

  • Planning – 1-2 weeks
  • Design & Build – 2-3 weeks
  • Testing – 2 weeks
  • Go-live – Dependent on customer’s SDLC processes/governance protocols

What can BizAI do for your organization?

Enterprise-ready in days—no downtime. See your exact workflow in action with a free, customized BizAI demo.

CONTACT US