Nine out of 10 R&D professionals at large consumer goods companies use some form of AI in their work, but only 19% incorporate AI into their daily workflows. And new research from Turing Labs, a startup building AI/ML tools for CPG innovation teams, shows that fewer people say they’re actually making a big difference.
The fact that 91% of the 290 senior R&D leaders in the US and Europe surveyed by Turing Labs* use AI on a daily basis or are actively piloting/scaling it does not, at first glance, suggest that they are behind the game, admits Manmit Shrimali, co-founder and CEO of Turing Labs. But when you dig a little deeper, most of the activities are around concept generation, basic research, and summary tasks, where a general-purpose LLM excels, he says.
“Really, the big impact has been on the consumer goods marketing side, where marketers are now able to spend a lot less on ad agencies and have a lot more in-house ad management. That’s where we’re seeing success right now, because the data is so easy there. But it’s not product innovation.”
The gap between AI activity and AI impact
But when we get into more substantive attempts to improve the pace and quality of innovation, the problems begin. According to research from the Turing Institute, 20% of AI initiatives are never implemented, 14% are implemented but later abandoned, and 24% are still in use but have minimal or no business impact. On the other hand, 57% of respondents said that GenAI’s formulation output is too generic to be used without significant manual rework and validation.
According to Shrimali, the problem is two-fold. One is the mismatch between general-purpose tools and highly domain-specific workflows. He says formulation is not just a matter of search and content generation, but also includes ingredient chemistry, regulatory constraints, manufacturability, cost, taste, texture, shelf life, nutritional goals, consumer positioning, and the commercial context surrounding the product.
As a result, tools that look impressive in demos often fall short when tested against real problems. “Customers often dump thousands of technical white papers into ChatGPT and try to ask a question, only to then discover that they get different answers to the same question. So this kind of LLM approach can be a disaster.”
Second, Brian McCarty, director of engineering, argues that companies that see AI simply as a way to accelerate current processes, rather than rethinking how innovation is organized in the first place, will not reap the full benefits.
“If there’s one thing the FMCG industry is getting wrong when it comes to AI adaptation, it’s that they’re only trying to use AI to speed up existing workflows.”
Beyond formulation optimization
Founded in 2019, Turing Labs started out as a formulation optimization platform, but has since expanded into what Shrimali describes as an “innovation system” for FMCG companies, which he claims can typically be divided into three buckets when it comes to AI.
“The first person recognized that they could build it in-house, but realized it would take too long. The second person is constantly running pilots and trying everything without the resources to give a single platform sufficient legitimacy. The third person asked the right questions, chose one vendor, and dug very deep. They know that just because they can build something in-house doesn’t mean they want to do it; they know they have the talent.” [in AI] It’s very expensive. That’s why companies end up coming to us. ”
McCarty said some large consumer goods companies have also learned the hard way that trying to build their own systems doesn’t always work. He argues that some companies have “spent millions of dollars on the promise of data lakes and one large intelligent system.” But in most cases, he says, this didn’t happen. “What they got was isolated victories, frustrated product developers, and diminished confidence that AI can change the way products are actually created.”
“We do not provide a GPT wrapper”
Turing Labs, which has raised $19.25 million from backers including Insight Partners and Moment Ventures and works with private label teams at major consumer goods companies and retailers, is trying something more ambitious, Shrimali said.
Rather than positioning itself as a general concept generator, Turing Labs’ platform is built around domain-specific models and workflows that span the entire innovation cycle, from commercial KPIs and market positioning to formulation, regulatory constraints, ingredient evaluation, and commercialization, he said.
“We’re not offering a GPT wrapper or a general-purpose AI. There’s no shortage of concepts and ideas in this industry. To win in the market, we need new operating models. We need reasoning systems that tell us which levers to pull, how to build winning products, and how to pay for our innovation efforts.”
It also helps CPG clients understand their risks, McCarty says. “The risk profile that FMCG companies take on often doesn’t seem right. They often don’t take on things that have the potential for great success, such as multi-billion dollar brands, but at the same time when they do take risks, it’s based on gut feeling rather than measured and calculated. This is one area where they can help make better decisions going forward.”
Beyond chatbots
Shrimali, a Turing Lab customer, said it usually starts with a business problem, such as “taking back market share from private label pasta sauce.” Turing Labs’ agents “work on this problem end-to-end to identify where the gaps really are (product, price, pack, regulation, claims, etc.), what to make, the optimal formulation, what the cost is at scale, where the regulatory risk is, and what cost savings across the portfolio can fund startup.”
“Blend optimization” [an area where the firm originally focused] is still there, but now one of several agents. ”
The system combines public data, expertise, proprietary data generated by Turing Labs, and customer data. These data are isolated and not used to train other customers’ models.
According to Shrimali, users can interact with the system through a chat-based interface as well as browser-based workflows. However, chatbots alone are not sufficient for making high-stakes R&D decisions, as the quality of results “depends largely on the way the questions are asked.” So the company sought to balance LLM-style interactions with a more structured workflow that reflects how product developers actually work.
Where agents can help and where they still fall short
As for AI agents, they have improved dramatically, but they are not ready for prime time in every case. They still fail, misunderstand context, require human intervention, and require proper orchestration, context, guardrails, and human collaboration, rather than being treated as autonomous problem solvers who are handed tasks and left completely alone, McCarty says.
But one example where they can perform well is in regulatory work, he says. Currently, product developers send formulations to regulatory teams and sometimes wait days to get feedback. More proactive AI workflows can provide early warning of regulatory risks, translate rules into constraints for formulation, and suggest alternatives.
In this model, AI goes beyond simply checking compliance and helps teams think about how regulatory changes create opportunities for offensive and defensive innovation.
“Structurally Defensive” CPG Innovation
Shrimali acknowledged that the fundamental issues facing R&D executives in this study are not new. If this survey had been conducted five to 10 years ago, respondents likely would have expressed similar general complaints. They have to do more with less, waste time on unproductive tasks, deal with incompatible legacy IT systems, and spend blood, sweat, and tears on innovations that, like most new consumer products, ultimately fail.
But what’s notable in 2026 is that many large food companies are “asking their R&D teams to create the future,” even though their budgets, staffing, and incentives remain heavily tilted toward defense. existing A product, he insists.
“The deeper problem is that innovation in consumer products has become structurally defensive and risk-averse. Look at how pharmaceutical companies are winning. [the food industry] About GLP-1. While organizations rely on innovation to drive growth, much of their R&D capacity is directed toward protecting their existing portfolios. Breakthrough innovations are not only difficult to implement; We have few resources. ”
Meanwhile, he says, the gap between brands and private labels is narrowing, with smaller, more nimble companies beating Big Food in the innovation bet. In such an environment, simply using AI to reduce reformulation time or come up with packaging innovations may not be enough.
For Shrimali, the more relevant metric for AI in food R&D is whether the R&D is contributing to revenue, margins, shelf space, and driving truly differentiated products, rather than simply reducing time to market.
Research highlights
According to Turing Labs research:
- Research and development efforts over the next 12 months will be overwhelmingly focused on maintaining and optimizing existing products rather than developing new ones.
- The majority of research and development efforts are invested in concepts that will never make it to stores. And once a product is launched, more than half require reformulation within 12 months.
- 52% of respondents reported needing to make changes within the first year of launch, and 49% of those made adjustments between 6 and 12 months.
- 60% of respondents say the quality gap between branded and private label products has narrowed over the past three years, and 3% say it has completely disappeared.
*For companies that formulate and manufacture food and beverages with 1,000 or more employees.
