What is a Large Language Model —
and why should your board care?
Large Language Models (LLMs) are reshaping how companies operate, compete, and get valued. Here is what every non-technical executive needs to understand before the next board meeting — and before the next deal.
Start here: the plain-language definition
A large language model (LLM) is a software system trained on vast quantities of text — books, articles, code, websites — to predict and generate human-like language. Think of it less as a search engine and more as an extraordinarily well-read generalist: one that can draft a contract summary, answer technical questions, translate a pitch deck, or write working code, all in seconds.
GPT-4 (powering ChatGPT), Claude, and Gemini are the most visible examples. Behind the consumer products, the same underlying technology is being embedded into enterprise software at a pace that few organisations — and fewer boards — have fully registered.
“LLMs are not a feature update. They are an infrastructure shift — one that will be visible in margins, headcount ratios, and competitive moats within the next 18 months.”
How it works — without the engineering lecture
You do not need to understand the mathematics to make good decisions about LLMs. What matters is the business architecture:
The distinction matters in due diligence. A company that says "we use AI" could mean anything from a GPT wrapper bolted onto a product overnight, to a deeply integrated, proprietary system that genuinely differentiates the business. These are not equivalent risk profiles.
The four questions your board should be asking
Whether you are evaluating an acquisition target, governing a portfolio company, or steering your own transformation, these are the questions that separate informed oversight from misplaced confidence.
Build, buy, or rent?
Is the company building on foundation models (rental), fine-tuning them (partially owned), or developing proprietary models (built)? Rental is fast but creates dependency. Built is defensible but capital-intensive. Most companies that claim the latter are doing the former.
Where does the company's data go?
Every prompt sent to an external LLM potentially exposes the data within it. What data governance policies exist? Are customer records, IP, or regulated data being processed through third-party APIs without adequate contractual protection?
What happens when the model is wrong?
LLMs hallucinate — they generate confident-sounding, plausible-looking output that is simply incorrect. In a customer service chatbot, this is an embarrassment. In a legal, financial, or medical workflow, it is a liability. What human review processes exist?
Is the "AI advantage" real or repositioned?
Since 2023, many companies have relabelled existing software features as "AI-powered." In a transaction, it is critical to separate genuine LLM integration — with measurable efficiency gains — from marketing repositioning with no underlying change to the product architecture.
What this means for valuation
LLM adoption is beginning to affect how investors price technology businesses. The dynamics are still early, but two patterns are already visible:
Upside: Companies that have genuinely embedded LLMs into their core workflows are seeing measurable productivity gains — particularly in software development, contract review, customer operations, and knowledge management. Where these gains are auditable and recurring, they should be visible in margin trajectories within 12–24 months of deployment.
Downside risk: Companies whose products are heavily dependent on a single foundation model provider (particularly via API) carry an underappreciated concentration risk. Pricing changes, model deprecations, or policy shifts at OpenAI or Google can materially affect unit economics overnight. This is the LLM equivalent of single-cloud dependency — and it deserves the same scrutiny in a technology due diligence.
The one signal that separates leaders from laggards
In our due diligence work, the single most reliable indicator of genuine LLM maturity is not the technology itself — it is measurement. Companies that have moved past experimentation can tell you, with data, what their AI systems have changed: tickets resolved without human escalation, hours saved per analyst, error rates before and after deployment.
Companies still in the “AI strategy” phase typically cannot. That distinction — between AI as a measurable operational capability and AI as a narrative — is increasingly the difference between a premium multiple and a discount.
