Neuralift AI builds trust in AI-powered marketing segmentation using W&B Weave
The explosion of consumer data – including transactions, online and social media interactions, and demographic information – has provided marketers with unprecedented insights. Combined with the power of AI, this wealth of data enables marketers to paint full, accurate pictures of their target audience, allowing for greater personalization and potentially improved efficacy in their campaigns.
Neuralift AI, a marketing segmentation and strategy platform, aims to leverage this volume of consumer data to provide marketing teams with “superpowers in actionable marketing segmentation,” according to Co-founder and CDO Mike Maloney. The platform incorporates a generative AI element that not only summarizes insights into micro-details on segments and cohorts, but also generates copy and campaigns optimized for those segments and their targeted KPIs, transforming data-driven insights into specific, actionable steps.
Ensuring the accuracy and reliability of both segmentation analysis and AI-generated responses is paramount for Maloney and his team. “Trust is a big word for me,” said Maloney. “We want to continually give an experience to our customers that they can trust. As we’re testing and updating the explainability and GenAI portions of our product, we need to be constantly running evaluations to ensure we’re always delivering trustworthy, non-hallucinated responses.”
This intense focus on trust and explainability led Maloney and the Neuralift AI team to try out W&B Weave, a lightweight toolkit to help software developers track, evaluate, monitor, and iterate on their LLM applications.
The segmentation difference with Neuralift AI
Neuralift AI harnesses millions of consumer data points—spanning transactions, social media interactions, web page activity, demographic info and much more—to deliver precise customer segmentation analysis. The platform then goes a step further—the “lift” part of Neuralift AI—by giving CMOs and their marketing teams applicable, timely insights optimized for each micro-segment, with specific KPIs such as driving loyalty program sign ups or increasing cross-sales on related items.
“Marketers want to know who their customers are, but there’s so much data that no one person can sit there using SQL-based techniques and segment all that data,” said Maloney. “Understanding who your customers are and what they value is key. That’s what Neuralift does.”
Neuralift AI excels at sifting through vast amounts of data to identify hyper-affinity segments, such as “Selective September Joiners,” “Tuesday Meat Lovers,” and “Winter Wine Aficionados.” This precise segmentation allows marketers to optimize customer targeting and personalization and lift their marketing ROI. Along with segment discovery and creation the platform’s AI analyzes each segment, then generates actionable recommendations, including suggested next steps (e.g. email or text outreach), marketing campaigns (e.g. sales promotions or loyalty programs), and tailored email and ad copy to effectively reach these target buyers.
“The idea is, whatever your marketing team wants to know about specific customer segments, or whatever your data is suggesting about a segment, we’ll make those insights more valuable with suggested actions for how to best market to that segment,” said Maloney.
To help humanize the segmentation data and better communicate insights and opportunities to marketers, the Neuralift AI platform includes an explainability portion which explains why exactly this segment was derived, and why marketing suggestions make sense for this target buyer at the right time – for example, like a promotional offer to a customer at risk of unsubscribing or churning.
Developing the AI-driven aspect of the Neuralift AI platform presented significant challenges for Maloney and his team. They grappled with painstaking evaluation processes and the need to eliminate hallucinatory responses. To make the platform valuable for CMOs, it required an extraordinarily high level of trust, reliability, and explainability; that’s why the team picked W&B Weave.
Continued innovation in deep learning
The emergence of LLMs gave Neuralift AI a chance to enhance and humanize the platform with GenAI insights. To do that, they needed new tools to help with logging and evaluating LLM applications to ensure that the summaries, insights, and suggested actions were accurate and trustworthy. Because LLMs produce varied outputs from the same input, their non-deterministic nature demands specialized monitoring and analysis. W&B Weave provides developers with rich visualizations to evaluate models—from aggregate performance across datasets to side-by-side comparisons of individual generations—helping distinguish creativity from hallucinations.
“I could literally just add a library to our code and a few decorators, and all of a sudden I’ve got a whole bunch of information about the GenAI portion of our product, right in Weights & Biases, which I was already using and very familiar with,” said Maloney. “That was extremely powerful and useful right away.”
W&B Weave now automatically logs Neuralift AI’s inputs, outputs and traces, giving Maloney and his team quick, easy answers to lots of questions like:
- How many tokens did this call to OpenAI use? How long did it take?
- What were the input tokens?
- What was my input? What was my input JSON?
- What was the data I passed up? What’s all the context around it?
“Whether it’s Llama, OpenAI, Turbo, whatever, we can make sure that we’re at least getting the performance we’re getting today, if not improving it,” said Maloney. “And then from that, we can use W&B Weave to make a decision on do we move to this model, do we move to a whole different provider? It’s really been kind of fundamental to how we test and how we move to new models.”
W&B Weave Evaluations streamlines reporting, enabling Maloney to compare datasets and operators side-by-side and generate custom field reports. This functionality allows him to quickly identify contradictions and calculate average contradiction rates.
Rapidly identifying and analyzing contradictions was crucial for Maloney in validating the Neuralift AI segmentation platform’s accuracy in suggested actions. W&B Weave streamlines this process, allowing Maloney to pinpoint contradictions, investigate their root causes, and to prevent misleading customer information.
“I love W&B Weave for a number of reasons, and it all goes back to trust,” said Maloney. “We’re doing something called ISO 4200, which is all about meeting international standards to ensure your AI is being ethical and unbiased. W&B Weave lets me demonstrate that really easily.”
Maloney and the team use Weights & Biases not only to evaluate LLMs but also to train deep learning models. They rely on W&B Models for experiment tracking across the hundreds of segmentation models they built, and can better analyze results across multiple runs and iterations. W&B Tables is instrumental in helping the team visualize printouts and steps around new releases. They were familiar with the W&B UI and rely on it to deliver robustness and flexibility at scale.
“With W&B Models, we could see how things evolved over time, whether we’re doing an experiment with different code or in the same dataset or different datasets, we could look at things in different dimensions to make sure that our models were continuously improving,” said Maloney. “The ability to report and monitor in near real-time as we’re doing development, that was my favorite part of W&B Models.”
“We want to exceed our customers’ expectations at every turn and using Weights & Biases, we can continuously iterate and improve on what we’re doing, so our users can improve their marketing efforts and make them more relevant, actionable and effective. W&B Weave is featured heavily in how we aim to continuously improve and build a high-quality applied AI product.”