Traditionally, personalization has coexisted with scalability in a delicate balance, each influencing the other’s feasibility. While discussing the limitations of our platform’s advertising capabilities, it has become evident that our team has consistently faced constraints on the amount of content we can produce due to these restrictions. Personalization was once synonymous with scaling tailored content to massive audiences, often in the hundreds of thousands or millions, due to limited resources and bandwidth among marketers. Generative artificial intelligence adjusts, driving down the cost of content creation and making bespoke content at scale a reality.
To facilitate the implementation of tailored content, we’ll outline a straightforward process that transforms a standard piece of content – for instance, a product description – into a customized version aligned with customer preferences and characteristics. While our initial approach tailors content by phase, there remains untapped potential to create ultra-targeted variations, prompting us to expect a surge in advertisers leveraging similar strategies for personalized email subject lines, SMS messaging, website experiences, and more.
This workflow requires:
- Viewers’ segments and attributes are categorized and analyzed based on the insights provided within our customer data platform (CDP).
- A comprehensive hub for disseminating valuable information.
- What’s your creative vision for our chatbot?
For this demonstration, we will utilize the power of knowing where knowledge is uncovered to construct an innovative structure using a zero-copy methodology.
The solution facilitates seamless integration between buyer information, including advertising phase definitions, and product descriptions, while providing access to generative AI capabilities through the Databricks platform. By combining these components, we’ll craft customizable product descriptions that harmonize with specific segments, showcasing the power of generative AI in blending knowledge to produce unique and engaging content.
A Step-by-Step Walkthrough
E-commerce platforms often rely on concise and engaging product descriptions to showcase their offerings online or through mobile apps. Our merchandising teams have meticulously developed each description in collaboration with advertising experts, ensuring seamless alignment with the targeted audience. Regardless of the user, a uniform product description for all products is displayed to each visitor.
The seamless integration of buyer insights from customer data platforms (CDPs) into a centralized knowledge repository. Here’s the refined text:
Integrate buyer knowledge from Customer Data Platforms (CDPs) into a unified Knowledge Lakehouse, streamlining access to vital consumer understanding.
Assuming that we have duplicates of these product descriptions stored within the Databricks platform, we aim to develop a mechanism for ingesting customer insights from our Customer Data Platform (CDP). With seamless connectivity enabled through the open-source Delta Sharing protocol, supported by Databricks, we can effortlessly integrate our Amperity Customer Data Platform with zero copying required.
By simply executing a few straightforward queries, users can seamlessly ingest Amperity CDP data from the Databricks Knowledge Intelligence Platform.
Step 2: Discover buyer segments
While advertising teams approach phase design uniquely, our framework incorporates key attributes such as predicted lifetime value, popular product categories and subcategories, value preferences, price sensitivities, and geographic location. The sheer diversity of unique values across these various fields is poised to yield tens of thousands of potential combinations, potentially even more (Figure 3).
While creating multiple versions may be necessary, we expect most marketing teams will require careful review and approval before deploying any generated content to their target audience. As organisations refine their understanding of the technology, and strategies adapt to ensure seamless content production, we expect this human-in-the-loop approach to ultimately become fully automated, streamlining processes and elevating quality and reliability.
For this initial exploration, we will focus on 14 combinations that arise at the intersection of the top product subcategories, which serve as a proxy for buyer curiosity, and predicted worth tiers, used as a surrogate for model loyalty.
Step 3: Summarize segments
Each of our 14 distinct segments aggregates clients with specific profiles, encompassing various lifetime value tiers, price sensitivities, geographic locations, and more. To help generate captivating content, we’ll analyze and summarize consumer demographics across key segments. If it’s a non-compulsory step, however we should uncover for instance which a considerable proportion of consumers in one stage is from a specific geographic area or has a robust low-cost sensitivity, the generative AI model might leverage this information to craft content more attuned to that demographic (Figure 4).
Immediate design and checking must occur to ensure that the product meets all requirements and specifications. This step involves creating a detailed design of the product’s features, functionality, and performance characteristics, and then verifying its compliance with relevant standards and regulations.
As we proceed to develop an imperative for a large language model capable of generating custom descriptions. We’ve chosen to employ the modality model that accommodates a standard directive, namely: I’ll improve the text in a different style as a professional editor and provide the revised text without any explanations or comments.
Please provide the original text you’d like me to edit, and I’ll get started!
(Note: If it’s not possible to improve the text, I’ll return “SKIP” only.) content material. As evident in Determine 5, our comprehensive immediate integrates phase information into a rich tapestry of detailed instructions, presenting the final product description as an integral component of its overall content.
It’s evident that multiple hands have contributed to this text over time. By experimenting with diverse approaches and prompts, we managed to eventually arrive at a method yielding acceptable results. Several notable classes were unearthed during this expedition.
- I’d like you to clarify the role for me, “As a sales associate at our retail store, your primary responsibility is to assist customers with their shopping needs and provide them with an exceptional buying experience.”
-
Effective steering on using details about knowledge requires strategic distribution of information throughout the phase. “To craft a bespoke product description, utilize information from {segment_description}, thereby tailoring a generic description to suit specific needs.”
- Embrace course corrections when crucial. We introduced “You might be writing copy for the whole phase” earlier to avoid anomalies like “As a bike owner and a New Yorker,” which was previously generated by the AI, even when not all clients within the phase resided in New York – although the phase may still exhibit a bias towards that demographic.
The company’s algorithm will generate multiple versions of each product description by incorporating different adjectives and adverbs to create a unique tone for each variant. This step is crucial in capturing the diverse preferences of customers, increasing brand recognition, and ultimately driving sales.
Once all the building blocks are in place, generating variants for each segment becomes a straightforward process involving iteration through every description and phase combination. Within a concise set of descriptions and manageable segments, this task can be completed at a rapid pace. If we currently have numerous mixtures, we would consider parallelizing the workload by leveraging more robust methods for generative AI models available through Databricks. The generated description variants were submitted to a desk for review (Determine 6).
To ensure high-quality customized descriptions, it is essential that someone with exceptional advertising copywriting expertise reviews the output carefully. Lacking is a definitive method to measure the precision and productivity of language models in generating high-caliber results. To ensure that the copy is yielding desired results, someone simply needs to review it. If the copy isn’t up to par, then revamping the nearby text and reauthoring the narratives becomes a necessity.
It’s crucial to acknowledge that we initially asked the model to provide evidence supporting the content it generated. One finds markers inside many descriptions that clarify the distinction between the original input and the algorithm’s generated output, thereby providing transparency on how the results were derived. The reasoning behind this approach may not always be presented to the buyer, but it can still prove beneficial to integrate these insights into the model’s outputs until consistent good results are achieved.
Classes discovered
As a pioneering example, this train showcases the transformative power of generative AI in propelling humanity toward a highly customized and targeted advertising landscape.
Artificial intelligence now combines personalized experiences with massive scalability for the first time. By tapping into granular information about individual buyers, we can craft highly personalized content that was previously inaccessible, unlocking new heights of precision and effectiveness through the power of generative AI.
Despite generating AI fashion designs with precision and speed, crafting high-quality, dependable content still necessitates concerted human effort in both prompt construction and output evaluation. Currently, this limits the scope of applying these methods, but we expect a gradual expansion as more reliable models, consistent prompting techniques, and novel research approaches further unlock the potential showcased here?
At Amperity and Databricks, we are committed to empowering manufacturers to leverage their data to deliver unparalleled customer experiences. Through our collaborative efforts, we strive to accelerate the adoption of analytics, empowering a more impactful advertising experience for our shared clients.
