Can LLMs reduce business risks?

Organizations everywhere generate millions of words of text each month building bridges to their markets. They create text for marketing and text for sales. They generate product descriptions and slideshows and emails and customer profiles. They generate comments to put in code or in spreadsheets for financial statements. They produce text for investors, for future customers, for existing customers, messages from one department to another, updates from one team to the next. They draft, revise, and circulate plans, contracts, and notices of payments due. Millions of words; a huge investment. 

Written and spoken texts are the lifeblood of every organization. And money keeps them pumping and circulating. So a company without texts is like a company without money: dead.

To save money and produce texts faster than ever before, it makes good sense to find ways to automate their creation, delivery, and even their uptake. And today we have the tools to do exactly that: hallucinations and other issues aside, LLMs do a great job of generating text. On many topics. In several languages. For different reading levels. In different styles. At the speed of business.  What could possibly go wrong? Can't we simply replace the language professionals – copywriters, information developers, translators – who generate text for us today?

The answer is buried deep inside a different question: 

Why, in the first place, do organizations need so many texts of so many kinds in so many languages to function? 

We can answer this deeper question with two words:  Risk Mitigation. 

The crucial role of text in the enterprise is to manage – and reduce – risk.  If texts don't reduce risk, then they increase it.

Marketing. For example, we produce marketing copy – text, with or without images – to avoid the risk of having an unrecognizable and indistinct brand with a negative reputation.  No brand awareness, no pool of potential buyers, no sustainable revenue. A company name, a logo, or an image aren't enough. Potential clients need to know why a company exists and build up a positive mental model of it – based on the text that reaches them about it. Better text about a company means more people know about it, and they in turn can create text about it, as with mavens, influencers, and viral marketing. Bad copy turns people away from a brand and sucks the blood from a company's veins.

• Can LLMs create unique, targeted, market-specific text to effectively mitigate this kind of risk? Can they do it reliably enough to eliminate humans with sophisticated linguistic skills and a clear sense of the risks involved? Nope.

Sales. We produce sales documents – the sales enablement function – to avoid the risk of having sluggish or no direct sales.  No sales, no company.  Sales professionals adapt existing texts – written or oral – to highlight how the product can meet the needs of specific clients. A single, standardized product spec sheet is nowhere near as effective as an interactive discussion of how your product can solve company X's problem in a cost-effective way.

• Can LLMs create adaptive, client-specific sales text to effectively mitigate this kind of risk? Can they do it reliably enough to eliminate humans with sophisticated linguistic skills and a clear sense of the risks involved? Nope.

Legal. We produce legal texts – like contracts and briefs – to avoid the risks of exposure to legal action and to support legal actions against others. Companies document compliance with business, environmental and human resources legislation using very carefully reasoned text. Summarizing the legislation about X, Y, or Z – without identifying the risks and opportunities of a particular course of action – will not do the trick.

• Can LLMs create relevant, well-reasoned legal text to effectively mitigate this kind of risk? Can they do it reliably enough to eliminate humans with sophisticated linguistic skills and a clear sense of the risks involved? Nope.

Product Management. Careful companies produce specific kinds of texts – Product Requirements Documents – to avoid the risks of poor product-market fit and delayed time to market, both of which can be fatal to a line of business or to a whole company. Summarizing existing product features without process savvy, in-country knowledge, and user feedback will not do the trick. Better PRDs mean better products launched faster to happier clients. 

• Can LLMs create relevant, well-reasoned product documents to effectively mitigate these kinds of risk? Can they do it reliably enough to eliminate humans with sophisticated linguistic skills and a clear sense of the risks involved? Nope.

Operations. Companies everywhere generate extensive documentation of their internal processes and actions – to avoid the risk of much higher development and support costs for their products and services. Documentation allows them to understand and decide once then leverage this understanding many times. Plans, reports, emails, and text messages all help move processes along effectively. Shared responses to common support questions avoids the time and cost of re-researching a problem each time.  Feedback from support to engineering and sales, for example, although rarely sought out or used, is almost always available and often very valuable. 

• Can LLMs generate useful, relevant process documentation to effectively mitigate these kinds of risk? Can they do it reliably enough to eliminate humans with sophisticated linguistic skills and a clear sense of the risks involved? Nope.

International Growth. As they try to conquer new markets, companies produce a great many of their texts in other languages – to avoid repeating all of the above risks in each new market. Localized materials enable increased awareness and sales, along with enhanced reputation, in new markets. But nearly-literal translation of marketing and sales copy, of legal documents, and process documentation not only does not mitigate all these risks, it increases them!

• Can LLMs generate accurate, culturally sensitive, legally compliant texts to effectively mitigate all these kinds of risk? Can they do it reliably enough to eliminate humans with sophisticated linguistic skills and a clear sense of the risks involved? Nope.

Using LLMs to generate a company's texts is like using leeches to improve its health. Quick, dirty, and ultimately counterproductive.

If text is the lifeblood of an organization, then much too much is at stake to rely on black-box automation. We need highly skilled language professionals to carefully assess which texts to produce for what reasons. When we understand why we need the texts in the first place and the often dire consequences of getting them wrong, then we can more reliably measure and reduce business risks instead of multiplying them at scale.