Large-scale language models (LLMs) have taken over the business world, and now every company is trying to use Generative AI. While tools like ChatGPT are clearly powerful, it’s not clear how businesses can reliably use this technology. Drive value.
For most companies I’ve interacted with, “Using AI“means creating chatbots, pilots, AI agents, or AI assistants. But as the initial excitement around these solutions wanes, organizations are realizing the key challenges in building systems around LLM.
The key challenge is that LLM is inherently unpredictable (even more so than traditional machine learning systems). Therefore, It is not easy to get them to solve specific problems predictably..
For example, one solution to the hallucination problem is for a “judgment” LLM to review the system’s responses for accuracy and appropriateness. However, increasing the number of LLMs increases the cost, complexity, and uncertainty of the system.
This doesn’t mean that Generative AI (and friends) aren’t worth pursuing. AI has made a lot of companies very wealthy, and I don’t think that’s going to stop any time soon.
The key is that value is created by solving problems, not by using AI (by itself). The promise of AI is that companies Identify right Problems to be solvedExamples include Netflix’s personalized recommendations, UPS’s delivery route optimization, and Walmart’s inventory management.
“Solving the right problem” is easy to say, but not so easy to do. To help, here I share three AI use cases for sales that every business is interested in. My hope is to spark your imagination and show you how to implement them with concrete examples.
Here are three use cases:
- Functional Engineering — Extract features from text
- Structuring unstructured data — Make it suitable for text analysis
- Lead Scoring – Identify the biggest opportunities
Featuring engineering It consists of Create variables that can be used to train machine learning models. Or do some analytics. For example, given a set of LinkedIn profiles, extract things like current job title, years of experience, industry, and then express them as numbers.
Traditionally, this is done in two ways: 1) manually creating features or 2) purchasing features from a third party (e.g. credit scores from FICO, company revenue from D&B). However, LLM created a third way. To do this.
Example: Extracting features from a resume
Let’s say you’re qualifying leads for a SaaS offering. The software helps midsize businesses protect themselves from cybersecurity threats. The target audience is IT leaders who are deciding which vendor is right for their company.
We have 100,000 professional profiles and resumes collected from various sources based on tags such as “IT”, “Cybersecurity”, “Leader”, “VP”, etc. But the problem is that the quality of the leads is low. They often include non-IT leaders, entry-level IT professionals, and people who do not fit the customer profile.
To ensure your sales efforts are focused on the right customers, The goal is to filter leads to include only IT leaders.Here are some ways to solve this problem:
- Idea 1: Manually review all 100,000 leads. problem: Not practical for individuals or small sales teams
- Idea 2: Write rule-based logic to filter resumes. problem: Resumes come in a variety of formats, so logic doesn’t work well.
- Idea 3: We pay data providers for this information. problem: This significantly increases the cost of customer acquisition (around $0.10 per lead).
Given the problems with the above idea, let’s think about how we can solve this problem with a large-scale language model. A simple strategy is to create a prompt that tells LLM to extract the desired information from the resume. An example is given below.
Analyze the following text extracted from a resume and determine whether the
person works in the IT industry. Return a `0` if the person does not work in
theIT industry, and a `1` if they do. Then, provide a brief explanation for
your conclusion.Resume Text:
{resume text}
This solution perfectly blends the three ideas above: (1) it reviews each lead looking for specific information like a human would, (2) it is automated by a computer program, and (3) it costs less (~$0.001 per lead).
**bonus**: For those interested in implementing something like this, I’m sharing an example Python script. here Extract years of experience from LinkedIn profiles using the OpenAI API.
Data from emails, support tickets, customer reviews, social media profiles, and call logs are all examples. Unstructured data. This simply means: It is not composed of rows and columns. Like an Excel spreadsheet or a .csv file.
The problem with unstructured data is that it is not suitable for analysis, making it difficult to gain insights. This is in contrast to: Structured data (in other words, A number composed of rows and columns) Converting unstructured data into structured formats is another area where recent advances in natural language processing (NLP) and deep learning can help.
Example: Converting a resume into (meaningful) numbers
Consider the same business case as in the previous example. Let’s say you successfully selected 10,000 IT leaders from 100,000 leads. Your sales reps can pick up the phone and write emails, but you want to make sure they can do the following first: Summarize the list Prioritize the lead Similar to previous customers.
One way to do this is to define additional features (e.g. industry, compliance requirements, technology stack, geographic location) that provide more detailed information about the ideal customer profile. This can be extracted similarly to Use Case 1. However, Identifying such indicators can be difficult.And developing additional automated processes costs money.
Another approach is to use what is called: Text embedding. Text embedding is simply Numeric representation of a chunk of text It makes sense semantically. Think of it like translating your resume into a set of numbers.
The value of text embeddings is that they transform unstructured text into structured numeric tables, which are much more amenable to traditional analytical and computational approaches. For example, in this context, text embeddings can be used to: Mathematically ranks leads that are most similar to past customers. And what is the biggest difference?
Here’s a final use case: Lead ScoringIt consists of the following Lead quality assessment Based on key predictive factors (e.g. job title, company revenue, customer behavior, etc.) This is not new, but recent advances in AI have improved the ability to parse unstructured data that can be fed into lead scoring models.
Example: Grade leads based on quality
Let’s discuss how we can finalize the ongoing business case. Use text embeddings to prioritize your audience.. Let’s say you have a list of 1,000 past leads, 500 of which purchased and 500 of which did not. For each lead, you have a profile with key information such as job title, work experience, current company, industry, and core skills.
These leads can be used to train predictive models that estimate the likelihood that a customer will purchase a product based on their profile. There are many nuances to developing such a model, but the basic idea is this: We can use the predictions from this model to define a grade for each lead. (e.g. A, B, C, D) to classify and prioritize 10,000 new items.
**bonus**: For the more technical reader who wants to implement this approach, this article walks through all three use cases applied to real sales data from my business. video. Also, example code is available for free at: GitHub.
AI holds tremendous potential for businesses, but to realize that potential, we need to identify: right There are issues that need to be addressed through this.
The ubiquity of tools like ChatGPT can easily limit solution ideas to the AI assistant paradigm. To help expand the space of possibilities, we’ve shared three practical AI use cases that use an alternative approach.
Learn more about AI for business 👇