What's CEXy?
Posts
The Rise Of The "Synthetic Customer" 🦾

The Rise Of The "Synthetic Customer" 🦾

Everything You Need To Know 🤖

Ben Carmichael
January 18, 2024

Hey Team,

I hope everyone enjoyed reading my 2024 CEXy predictions last week! I can’t wait to see how ambitious (or conservative) my predictions were …

As I trawled through many predictions last week, I stumbled upon an intriguing term that immediately piqued my interest - THE SYNTHETIC CUSTOMER.

A Forbes article predicted that 2024 would bring the “RISE OF THE SYNTHETIC CUSTOMER”:

The trend for creating synthetic customers touches on several key technology trends, including generative AI, data privacy and the concept of the digital twin. This enables companies to model their behavior, predict where pain points will arise on the customer journey, and predict the best path towards an optimal outcome. Customer personas are nothing new, but in 2024 they are informed by real-time data including live transaction data and social media sentiment, meaning they are more useful than ever when it comes to generating insights into real customers.

Bernard Marr

So this week, I’m going to dive into what a synthetic customer actually is. I have trawled the internet far and wide for answers, and have summarised my learning below.

I hope you all enjoy 🙂

So, What Are “Synthetic Customers”? 🤖

Synthetic customers are:

digital entities / bots designed to simulate real customer behaviors, preferences & decision making processes.
AI-driven avatars or bots that mimic human customer interactions. They are programmed to understand and replicate human emotions, decisions, and interactions, providing businesses with a wealth of data and insights.

They can be trained on an array of different datasets, including customer demographics, purchase history, online behaviour and social media activity. This training data allows them to behave like real customers, ensuring organisations can test concepts on the synthetic customer before it gets launched to an actual market.

Synthetic customers could theoretically predict what real customers are going to do, before they even do it.

Practical Applications For Synthetic Customers 🔎

Let’s start with a story … 📘

I write this newsletter every Thursday night, spending considerable time to determine how useful the topic will be to my readers whilst also trying to think of an eye-catching subject line 👀 …

On Friday morning, I wake up to and check the stats:

Email Open Rate (%)
Link Click Through Rate (%)
Customer Replies (%)

On some occasions, I’m extremely happy with the performance, whereas on other occasions, I’m not.

Some topics that I feel will be the most popular don’t resonate and some articles that I think are useless have actually become my most popular.

The point? I wish I could launch multiple versions of my newsletter to a fake audience before I sent it out to my actual audience.

The fake audience would be an exact replica of my existing audience (demographics, job titles, LinkedIn content preferences etc) and it would give me guidance on how I can improve my content before I hit the actual schedule button.

This “fake audience” is the exact same concept as a synthetic customer:

They analyse vast amounts of data to understand customer patterns and trends.
They are capable of engaging with humans in real-time, offering realistic responses, advice and solutions.
They adapt to individual customer needs, enhancing the customer experience.

5 Use Cases For Synthetic Customers 🦾

#1 - Predicting Responses To Marketing Campaigns - Having a synthetic audience to test marketing campaigns on could help marketers refine messaging & creatives, leading to a significantly higher ROI on ad spend. Most Ad Agencies now realise that relying on ‘gut feel’ isn’t a credible method for achieving the highest ROAS. Whilst most agencies have become savvier in the way they measure marketing ROI, this is currently done after the fact. Imagine a scenario where they could test creatives with a full mock audience before they launch - it could be a game changer!

#2 - Replace Customer Personas - Customer personas are often established by service designers / researchers using statistically insignificant research methods. It is quite often a laborious task involving a number of customer interviews and journey mapping exercises. Whilst there still may be an important place for customer personas (due to the human-centric and empathetic approach of customer research), the fact is that customer personas are not created off of statistically representative sample sizes. Synthetic customers, on the other hand, are data-driven and predictive. These customers will be created using large customer datasets, and would be much more representative of the entire customer population you are trying to serve.

#3 - Predicting Customer Churn - Imagine if your organisation built a simulation of every customer and their associated journey throughout your organisation - from sign up to cancellation. For any new customer that joins, the business could use the experience of previous customers to predict the future behaviour of the new customer. For example, if they have a poor customer service experience, you could ask the synthetic customers what the likely outcome of that poor experience would be. If it’s going to lead to a negative outcome i.e. churn, you can implement some guardrails to course-correct the situation.

#4 - Improved Product Testing - By creating synthetic customers who mimick the behaviours of your real customers, you could theoretically go to market with a new product before it actually launches. You could use your synthetic customers to test different combinations of pricing tiers & product features, allowing your team to figure out the propensity to purchase at different levels. If Synthetic customers were around in 2012, maybe they could of told Kodak about digital cameras … ?

#5 - “Mock” Data Creation - Generative AI allows us to experiment with “fake” data much easier than we have been able to historically. We can simply ask GPT to produce 1000 comments from customers between the age of 18-25 who regularly shop online. Equally, we can ask gen-AI to answer survey questions for us, or even fill in the blanks. For example, surveys typically yield a response rate of 5-15%. What if synthetic customers could produce data for the other 85% of customers who didn’t respond?

Real-Life Examples Of Synthetic Agents / Customers 🤷‍♂️

Example One: GPT mirrors humans with ethics scenarios - In 2023, the Allen Institute for Artificial Intelligence wanted to see whether they could develop an AI system that made moral judgments like humans. As a first step, they figured they’d see OpenAI could already do the job. The team asked GPT-3.5 to judge the ethics of 464 scenarios, previously appraised by human subjects, on a scale from –4 (unethical) to 4 (ethical)—scenarios such as selling your house to fund a program for the needy or having an affair with your best friend’s spouse. The system’s answers, it turned out, were nearly identical to human responses, with a correlation coefficient of 0.95.

Example Two: CBA Aims To Simulate Customer Behaviours As An Early Experimentation Tool - Commonwealth Bank of Australia (CBA) is studying the use of generative AI to create 'synthetic customers' capable of emulating human behaviour, to test the response to new products and services:

“Generative AI enables machines to process, interpret and use patterns to create new outputs. We’re using this advanced technology to explore creating customer personas or ‘synthetic agents’, where GenAI chatbots act as an early experimentation tool.

“By drawing on simulated experiences of daily life to emulate behaviours, we’re testing whether these GenAI chatbots could provide qualitative and quantitative understanding of how customers might respond to changing contexts, everyday financial challenges, and new products."

Dan Jermyn - CBA Chief Decision Scientist

Key Risks To Be Aware Of When Working With Synthetic Data / Customers ⛔️

As optimistic as this article has been so far, it is worth mentioning that Synthetic customers are an extremely new concept which should not be totally trusted (yet …)

A few key risks / callouts include:

Lack of real-world experience: Synthetic customers may not be able to fully replicate the complexities and nuances of human behavior.
Bias: The data used to create synthetic customers – as well as the AI algorithms – can be biased, which can lead to inaccurate results.
Data quality: Building synthetic customers means you are relying on the quality and completeness of your data to be spot on 100%. That’s not always the case.
Limited emotional understanding: Synthetic customers may struggle to understand and simulate human emotions (an important component of the customer experience), thereby impacting the accuracy of their predictions and insights.
False confidence: We’ve all seen examples of ChatGPT confidently providing the right answer. As cool as AI has become, it is extremely important to continuously train, validate & test any models you are using to produce synthetic customer data, or any AI-related material.
Being misinterpreted as human data: As more synthetic data gets produced, it is going to be critical to figure out a way to differentiate between “real” and “synthetic” situations. When this data is mixed together (or misinterpreted), it could have dire impacts. This is something we all should be mindful of when completing analysis. For example, check out this Barack Obama deep fake video 😂

References For This Article