If anyone’s sick of hearing about data silos, it’s marketers.
But fragmented data is still a big problem for them, making it hard to understand the impact of specific channels on campaign performance.
On Tuesday, mar tech company GrowthLoop announced the launch of its new customer data platform, called The Loop, which uses generative AI technology to analyze campaign tactics and make specific recommendations about what’s working and what else is likely to work.
This comes less than a year after GrowthLoop partnered with generative AI company Typeface to release Marve, an AI product that helps marketers quickly produce audience segments and customer journeys.
While these two products are very much connected, The Loop goes further, CEO Chris O’Neill told AdExchanger.
How the platform works
Using first-party customer data from enterprise clients, The Loop can generate insights about how a campaign performs, including sales performance, but also why it performs.
With Marve, marketers can use natural language to create audience segments – for example, people who’ve bought a ticket to a Boston Red Sox game in the past twelve months – together with a recommended customer journey to reach that audience.
The Loop builds upon those capabilities by analyzing the Red Sox segment and customer journey against other data points, such as transactions, and then generating recommendations for campaign optimization.
“Marketers can see the sales they produce for each audience across any channel or campaign and make improvements based on what made money, not just clicks,” O’Neill said.
The platform does this by sitting across several cloud data warehouses, including Google BigQuery, Snowflake and Amazon Redshift.
Traditional data transferring usually involves copying data from one location to another, which can lead to a higher possibility of security risks. However, GrowthLoop uses something known as zero-copy architecture, which means the data can be accessed across different databases without being physically moved.
The platform is powered by several multimodal language models, which are deep learning models trained on large datasets, including Google’s Gemini, Snowflake Cortex, Meta’s Llama and tools from OpenAI.
The Loop can compare, contrast and even combine these models based on which are best for a specific set of activities.
The idea for this came from an internal hackathon, where GrowthLoop engineers were inspired by a type of function known as “continuous queries,” which allows for incoming data to be analyzed in real time.
“Someone said, ‘Wouldn’t it be great to validate these models to see which ones are doing relatively better at this final piece of the closing the loop?’” O’Neill said. “That’s where it started.”
Ensuring accurate results
Of course, proprietary cross-functional systems like The Loop rely on more than just language learning models to operate. Often, retrieval-augmented generation (RAG) is used to ensure that the AI tool is reporting accurate information without hallucinating.
“It’s not the model trying to guess what things are,” said Scott Brinker, VP of platform ecosystem at Hubspot, who was briefed on The Loop.
Instead, he said, the model “is helping to translate between things like natural language requests, into things that become concrete requests.”
To put it in more practical terms, let’s say a user asks a very basic generative AI model to name five fruits that end with the letter “Y.” But the model doesn’t have a concept of what a “fruit” actually is; it’s just one word it knows out of millions. There’s a good chance of it generating an incorrect answer.
To combat this, The Loop could ingest a database of fruit names, and then, when asked, be able to retrieve information from that specific database using a RAG process.
“It’s translating the query there, but the underlying data is not being synthesized by the LLM,” Brinker said.
Being able to deliver these kinds of accurate results to marketers at a faster pace – and without having to wait for a data analyst to retrieve the requests – is one of The Loop’s biggest selling points, according to O’Neill.
One senior marketer who’s had a chance to work with the product told O’Neill that her team can experiment with their data at a rate eight times faster than what they could do previously.
Similarly, some campaign measurements which would have taken six weeks with existing tools can now be done in under ten minutes, he said.
This rate of speed will hopefully allow for even more experimentation, something Brinker believes to be the “single greatest marketing lever in a digital world.“
If, say, roughly 10% of a marketer’s experiments work out and it’s usually only possible to run around 10 of them a year, that mean just one winner.
“Now, we can actually run hundreds of these a year, and we get 10 winners out of it,” Brinker said. “That’s a pretty magical impact on what marketing’s going to be able to do.”
