What type of AI solution is right for your business?
Does your business have needs related to ‘words’ or ‘numbers’?
Introduction
There's been a lot of excitement around the progress made with large language models (LLMs) and Generative AI (GenAI). These models have enabled businesses to automate a lot of tasks that were previously difficult to automate--especially those related to text and language.
As language-oriented social animals we are hardwired to experience language as intent, intelligence, perceptions, and facts. So, we are naturally excited about LLMs and how we may use LLMs in our day-to-day lives and business.
Generative AI models are a class of models that are trained to produce plausible examples of non-numeric content, such as text, images, or music.
LLMs are the subset of GenAI models that specifically produce text. GenAI is useful for tasks that require the production or processing of content and for business applications, text content is the most common type.
However, not all business-critical tasks are suitable for LLMs or GenAI. Here, we will try to draw a distinction between when you should or should not consider using LLMs or generative AI in your business.
The simplest way is to ask yourself - Does your business have Needs of Words or Needs of Numbers?
Needs of Words
i.e. language, text, content
If your business has language-oriented tasks, goals, opportunities, or problems, then LLMs are an appropriate tool. LLMs (such as ChatGPT, Perplexity, Claude and others) routinely perform well in question answering, knowledge extraction, document summarization, and other text and speech tasks.
In fact, they have already become a part of our daily lives - check out some of these trends.
OpenAI has 400 million users of its automated question answering
Google has replaced its top search result with an LLM generated answer
Scientists are re-indexing older research papers for advanced search and synthesis
Call centers are increasingly automated
If your business has any of these needs (research, searches, summarization, synthesis, other content tasks), then by all means you should be considering how you can take advantage of this newest AI technology.
For LLMs to work for you, your basic business model or the set of tasks that enhance profitability should be driven by content or language processing. In this case AI, when implemented properly, will make your business more efficient and effective.
Needs of Numbers
i.e. quantifiable indicators such as costs, prices, supply chain metrics, sales, financials
In the case of numbers, jumping all in on language processing may not be the right fit for your current business. It may still be the case that a given business may see its greatest profit and revenue opportunities in using AI tools that can bring efficiency and effectiveness in how your business manages its ‘numbers’.
LLMs will not be able to address these needs. You might be better off improving your existing toolbox with other types of AI, automation, and algorithmic tools.
For forecasting tasks, that is, predicting quantities like sales or revenue over future time periods based on past performance: consider classical or machine-learning based time-series analysis tools
For ranking tasks, such as identifying your most promising sales prospects, consider machine learning scoring models
For logistics tasks, such as planning efficient delivery routes consider operations research-based methods, and optimization algorithms
For estimation tasks, such as estimating the profit from a marketing campaign based on key features and knowledge of past marketing campaigns: consider machine learning regression models
For maintenance tasks, such as budgeting and scheduling machine maintenance: consider survival models
The statistical machine learning algorithms that we mention above can also be considered a class of AI that works over structured data (the type of data stored in a spreadsheet or database).
Each of the above examples depends on a "physics" of the world - how quantifiable quantities such as resources, products, customers, prices interact. These systems have long been captured with numbers, formulas, and equations. It is only the descriptions of these things that are linguistic.
Numeric methods, like classic time-series analysis or optimization algorithms, come from Engineering fields. The good news is that many of these methods are mature and field-tested for applications that GenAI is not well-suited for.
How to decide which AI tool is right for you?
Step 1 - Identify whether you have a business ‘need of words’ or a ‘need of numbers’
For ‘words’ tasks (search, summarization, and so on): LLMs are the way to go!
For ‘numbers’ tasks: mature standard methods like the ones listed above exist which have been utilized for a very long time.
Step 2 - Select the right tool or methodology
For ‘words’ problems: Identify the right tool for your use case, experiment with prompts until you narrow in on the answers you are looking for and be mindful of the potential pitfalls
For ‘numbers’ opportunities, finding the appropriate methodologies is now (thanks to LLMs) a variant of search. Most classic numeric methods require some experience to deploy optimally and safely. Consider using internal data science resources or engaging consultants to help your team build out these ‘numbers’ solutions.
For some use cases, you can use a combination of LLMs and other AI tools.
Step 3 - Assign the task to the right individual or team
For ‘words’ tasks, designate individuals who have experience with using the right prompts, checking sources, evaluating and comparing different features offered by the various LLM tools on the market and are well-versed about the potential pitfalls. For more on using LLM-based AI tools, read our article on this topic.
For business needs of ‘numbers’, it used to require large teams to expect expertise in all of the various techniques. Nowadays a smaller team can in fact solve these problems. They can even use LLMs to find descriptions of possible solutions, including example implementations.
In conclusion - Let the business need and not the tool be the guide
In the spirit of Marshall McLuhan's "The medium is the message" we need to be clear that the task drives the profit opportunity, not the tool.
It takes time and resources to implement AI solutions. In spite of the rush to adopt these tools, to save time on reversing costly moves, ensure that the focus of the rush is on using the ‘right’ tools for your needs.
For help with using LLMs or other AI tools and methods mentioned here, contact us for scoping, consulting, training, coaching, and evaluation by writing to conscisolutions@gmail.com
This article was written in collaboration with data scientists and algorithmic consultants John Mount, PhD and Nina Zumel, PhD
 | A guest post by
|