Readers in Europe and the UK may well remember a popular Saturday evening TV game show, which ran from the mid 1970s until only about a decade ago. ‘The Generation Game’ used a format whereby different generations of the same family would compete against other families to win prizes.
Grandmothers and granddaughters would go head-to-head against, say, sons and fathers of different families. After 45 minutes of aptitude tests and general knowledge questions, in the final moments of the show, the contestant pair with the most points would win any prize they could recall from a conveyor belt of items running past them at a rapid speed.
Anything from coffee makers via giant cuddly toys to Hi-Fi systems would rush before their eyes, and the winning couple would have 30 seconds to remember any items that they saw, which they could then claim.
The noticeable fact was that not many people could remember more than six or seven objects out of, say, thirty or more – a cheap way of the program’s producers keeping hold of their prize budgets.
But it’s this failing of human memory that leads to the gist of this article, because when people are left to their own devices to create documentation for almost any purpose, they make simple human errors.
From forgetting to add crucial elements to a legal contract to leaving a bottle of milk off a shopping list, most people just aren’t very good at producing documentation that is 100% accurate and fit for purpose.
Say goodbye to slow manual data entry.
This is where Artificial Intelligence (AI) comes to the rescue in the form of a document generation system of software. A document generation platform, or add-on to any Customer Relationship Management (CRM) package, automates the process of creating documents from raw data fed into its system. In the ‘bad old days’ before scanning and consequent Optical Character Recognition (OCR) this would have necessitated the manual creation of a template and typing individual lines or words into appropriate places.
The process would have been glacially slow. Imagine an insurance company with the daily details of hundreds of motorists buying policies, and having to manually add the model of car, its color, engine size, date of registration and an interminable list of specifications and driver criteria.
Fortunately, AI and OCR now enable such processes to be done in seconds rather than hours – mainly thanks to the use of Large Language Models (LLMs) trained via machine learning. First of all, OCR means that computers can now ‘read’. The textual content of scanned documents is ‘understood’ by the AI to be a list of words.
Large Language Models
LLMs draw upon vast amounts of Big Data. Platforms like Chat GPT and Google’s Bard simply digest reams and reams of pages of text from anywhere it can be sourced; e-books, the internet, scanned text, to produce a colossally huge number of words and sentences. The AI within the LLM then ‘learns’ the relationship between placement and words and ‘understands’ their meaning within a human and grammatical context.
This has now enabled platforms like Chat GPT to pass the Turing Test, an activity created by the genius computer scientist Alan Turing, whereby a person in an isolated room would have a typed conversation with another party on a screen. The test was deemed to be passed if the person turned out to be having a conversation with a computer, but thought the conversation was with another human being.
As a result of this understanding of context, document generation software can create anything from invoices, letters, reports, contracts and just about any form of business communication you might think of. The software starts the process by using predefined templates.
Then, using a rules-based architecture, the AI would put the appropriate words in the right places. In the example above of a scanned or inputted insurance policy proposal, the software would ‘know’ that BMW was a marque of car or motorcycle.
It would then scan the remainder of the input for that customer, and within milliseconds would recognize that the characters 325i meant a BMW Three Series car with a 2.5 liter petrol engine. All the specifications of the vehicle would be pre-programmed so the document template for that section of the insurance proposal could be filled out faster than a person could blink.
That’s obviously a very simplistic example. Document generation software goes well beyond such basic facilities, and can extract such things as legal terms, contractual clauses and the like, then place them into the correct order to make a legal contract from a scanned letter or email sent by a potential customer.
Features and Benefits
Now we have a basic understanding of how the document generation software (DGS) sources its material and content, let’s briefly examine some potential and actual facilities that can be adapted to almost any industry sector usage:
· Collaboration: the software may offer collaboration facilities, enabling teams and multiple employees to edit and review documents based on permission levels.
· Templates: Not only do pre-constructed templates come with document generation systems, but they can be adapted, and even brand new templates can be created from scratch. Formatting and layout can be pre-determined or specified, and placeholder ‘Lorem Ipsum’ text left in place of dynamic content. Placeholder text is very useful to be able to judge the final length of a finished document.
· Customization: Users can specify the content of documents based on desired criteria. For example, rules can be specified that determine whether or not sections are to be included or excluded based on determined rules.
· Data Integration: DGS should be able to employ various data sources, such as spreadsheets, databases and scanned documents, which can then populate templates accordingly, as in the car insurance proposal example above.
· Version Control: Ideally, DGS should include version control features, especially when collaboration is allowed, to be able to track who edited which documents, when, and the content they added or removed.
· Automation: Whilst DGS automates any document creation process, not only does this reduce manual data entry but mitigates human input error.
· Security and Compliance: DGS packages should include facilities to recognize such content that might be placed under ‘special categories’ of the EU General Data Protection Regulation (GDPR) and in the USA, such legislation as the California Privacy Rights Act (CPRA). In short, this means that information pertaining to any identifiable person’s sensitive data – perhaps such as sexual orientation, trade union membership, political affiliations and financial data. Such categories can be recognized by the DGS and either blanked out from view without appropriate permissions, or completely erased – depending upon the use-case context.
· Merge Fields: Merge fields can be used in templates to indicate where given data should be placed. DGS swaps those fields with actual data during the document generation process.
· Output Formats: Any DGS should be able to export its output into many popular formats such as PDF files, MS Word documents, Mac Pages format, Microsoft Excel, Comma Separated Values (CSV) files and even as HTML web content, depending on the final destination of the content.
· Platform Integration: DGS should ideally employ any number of Associated Program Interfaces (APIs) so that it can integrate and export to other business software and collaboration packages such as Slack, HubSpot, Salesforce et al.
Industry Sectors
DGS can be used across various industries, including financial, legal, customer service (especially call centers and outsourced offshore service centers), healthcare, utilities, insurance, e-commerce, government departments and many others.
For online marketers, even search engine optimization is now often performed by AI – from the discovery of keywords to their placement in content of web pages.
Commonly, DGS is used to draw upon other documentation to pull out necessary data for the creation of contracts, proposals and customer communications. When used correctly, DGS can speed up operations, improve efficiency and reduce errors in textual-intensive workflow.
Some challenges
No software is perfect. However, as DGS software runs on computers, once the machine is programmed, it does exactly the same thing over and over until the programming changes. In short, unlike humans, computers don’t make the same mistake twice, once they have been corrected.
But programming procedures and determining the rules to make DGS work smoothly can be challenging. Here are a few of the more common hurdles:
· Garbage in, garbage out. GIGO – if any source data is inconsistent, inaccurate or plain wrong, it can obviously lead to mistakes in any generated documents. Data quality control is essential, and programs to double check data against known parameters can be written. Simple human common sense could avoid such errors – for example, a simple typo error on a scanned document could make the price of a $10 product into a $100 item.
However, AI can now be programmed with parameter limits, for example, the DGS could stop its run and highlight the fact that a printer cartridge appears to be priced at $250 rather than $25 and wait for manual confirmation or denial. In the case of denial, the operator would be forced to change the parameter before the DGS would proceed in its task.
· Complex Templates and Document Formatting: One of the trickiest activities in creating documents is appearance and layout. The textual content, spelling and grammar might be perfect, but getting images, infographics, page breaks, font sizes and colors to all integrate correctly can be like herding cats – nowhere near easy!
Branding and design standards are crucial, so when exporting documents to any word processing packages, rigorous quality control needs to be put in place. This part of the DGS process can often be the most problematic. Aesthetics is always more difficult than empiricism.
· Logic and Rule Complexity: As AI works on rules-based architecture – defining the rules can be challenging. For example, is a 5mm widget classed in the ‘fixings’ or ‘decorative’ category when creating a new business quote for a customer? Defining complex rules can be challenging. In the instance of producing quotes, a DGS package might be better replaced with CPQ software (configure, price, quote) but the definitions and criteria of rules is challenging from the outset. Managers should assign specific personnel with enforcing the rules and ensuring consistency across an organization.
· User Training and Adoption: Ensuring that users are properly trained to use software is essential. A Digital Adoption Platform (DAP) in conjunction with the DGS would be invaluable in this context.
In summary, to use DGS effectively, an empirical and disciplined approach to document output is essential across any business, from the sales team to creative marketing procedures. Branding and presentation is as important as content.
But once procedures are established then maintained, and appropriate software integrations are made, document generation platforms are definitely a major weapon in any business’s armory for success.
