For over 175 years, iconic dairy brand HP Hood has kept grocery store shelves stocked with milk, cream, eggnog, cottage cheese, ice cream, yogurt and other delicious dairy staples. Founded in Massachusetts in 1846, Hood continues to deliver on its commitment to quality and sustainability while introducing innovations like the LightBlock Bottle that have helped modernize an industry known for its heritage traditions. In addition to its core business of Hood branded dairy products, HP Hood is also a market leader in oat milk (Planet Oat), almond milk (Almond Breeze) and lactose-free milk (Lactaid).
While the dairy industry may be a somewhat unlikely locus of fast-paced innovation, says Hood food scientist Curtis Park, the marketplace is changing rapidly in ways that will soon require traditional brands to evolve. “Our development timelines have been getting shorter every year. This cuts down our testing time, leading to increased chances of problems at the factories,” he explains.
Park, who joined Hood after completing his PhD in Food Science at North Carolina State University, recalls having a sense that statistical methods could hold the key to the future of food science, though it was several years before he saw it in action. “I knew how powerful statistics could be – or at least I believed it could be really powerful – but in a way, I hadn't been entirely convinced yet,” he recalls.
Walking in the door at Hood, the culture he saw was one of steady – if slow – progress. Traditional single factor experimentation was often effective, but the timeline required for a development cycle was long and improvements more reactive than proactive. In the face of growing market demands, Park saw the potential of using data in new ways to accelerate scientific research. So, when their new Senior Director of R&D, Joe DeStephano, took the helm, Park leaped at the opportunity to explore how statistics could help the company’s science go further.
Two years later, and in no small part thanks to Park’s efforts, Hood has undergone a cultural shift toward statistical enablement. Park, now Principal Scientist, has played a key role in advocating the adoption of robust statistical approaches to food science – moving, for example, from single to multifactor experimentation – a shift he sees as a panacea for the pace and quality demands of the marketplace.
DOE reduces costly failures, cuts experimentation time by 50% and makes research more reproducible
R&D scientists at Hood, Park explains, contend with both ingredient and processing factors in formulation development. Taste, quality, shelf life, stability and cost all factor into the optimum product formula, and without statistical methods, arriving at a final recipe can at times feel like trial and error. “It’s like throwing darts at the board, hoping that you’ll find what you're looking for,” Park explains.
As a first step toward improving that approach, Park began working with statistical consultant Lynne Hare whom Hood had recently brought on to support the company’s quality and operations departments. In early conversations, Hare was adamant that a statistical approach known as design of experiments (DOE) could quickly streamline Hood’s R&D processes.
“Lynne started talking to us about the power of DOE, and I was really interested because it was something I had been looking for but just didn't know what it was until I found it,” Park recalls. “[Seeing DOE for the first time] just flipped a light switch on. I realized this is exactly what I'd been seeking – a better, more systematic way of doing experiments.”
Fundamentally, Park says, it was a reliance on Microsoft Excel that was limiting scientists to one-factor-at-a-time experimentation because it does not enable DOE. “We knew we needed a statistical software package to get started [with DOE], and Lynne had no hesitation in saying R&D needed JMP®,” the industry standard statistical discovery software for multifactor experimentation.
Since licensing JMP for every scientist in Hood’s R&D group, the team has gone from no DOE to DOE as a standard best practice – all in the space of just over two years. “The money we pay for our JMP licenses, we make up easily in one DOE. That's all it takes,” Park says. “You can save that much money just on [reducing] costs, on having a successful trial. If you have a failure, that could be $20 to 40K down the drain.”
Cost reduction also stems from significantly shorter research timelines. DOE, he says, “cuts [experimentation time] considerably. And that accumulates over time and really reduces the time needed to find the formula we will commercialize."
DOE also builds product knowledge, Park says. Reproducible, systematically documented experiments have created a sizeable historical data set that Hood’s scientists can now return to as a source of deep product and process knowledge that will inform future product iterations. Moreover, visibility around R&D has benefited the team’s quality and operations partners, where Park says communications have been significantly improved around what’s being done by R&D and why.
A combination of initial successes and training secures widespread buy-in for DOE
Not all of Hood’s scientists were immediately persuaded that a new strategic direction was necessary – or indeed that DOE was a fitting solution. So, together with Hare and DeStephano, Park began building an initial case study that could persuasively demonstrate the power of statistics to a skeptical audience.
“I decided to take one of our biggest problems and solve it using DOE in JMP,” Park says. “I figured the best case to use would be something that other people had tried to solve in the past without success. That way, we could show them: ‘Hey, you couldn’t solve it this way, but if you use [DOE], now you can.’ So we did, and we solved a problem that had been going on for years.”
Once the case study had been socialized around the department and to leadership, Park recalls, “Everybody got on board with the solution instantly. There was no second guessing.” The next hurdle was to prove to scientists that DOE would work in their area, not just on parallel challenges, and the introduction of a new R&D statistical training program was the last piece in securing support.
Hood R&D’s statistical training program is now standard onboarding for new employees, and those looking for something more he points to Statistical Thinking for Industrial Problem Solving (STIPS), a free online training course from JMP. STIPS, Park explains, allows users to acquire new skills and complete modules at their own pace. “I'm a little fanatical, I think, about JMP. And everybody in R&D knows it,” he laughs.
Discovering JMP and DOE, Park reflects, “It really has been a career changing experience – I feel very strongly that it has changed me as a scientist. Problems are much easier to solve now. And yes, I still use my subject matter knowledge. It guides the DOE to where I can find solid answers that I can be very confident in – and get other people to be confident in as well.”
Query Builder streamlines data access processes from 1 hour to 30 seconds
Another project the team tackled in the early phase of JMP implementation was the improvement of the specification development processes. Data access barriers had long presented a challenge for the R&D group, with quality data siloed between databases that didn’t talk to one another. To increase performance metrics, Park worked with Hood’s IT department and Hood Industrial Statistician Bill Henry to connect the company’s QA database to JMP using the tool’s Query Builder platform.
Now, scientists can easily access data from various sources, run a query and look at how the data is trending over time or how the process is performing against specifications. The system also provides tools Park has used to measure process capability and augment specification development. “Query Builder has been huge,” Park says. “What used to take us probably at least an hour I can now do in 30 seconds.”
Ultimately, he explains, small-scale efficiencies in aggregate deliver outsize benefits to the business by freeing scientists to focus their time on more value-adding activities. “We want to automate the things [like data access tasks] that are not valuable for a scientist to spend their time doing,” Park explains. “Instead, we want them to be spending that time thinking critically about what the results mean and what we should do about them.”
What Park describes as “a little bit of scripting” has helped automate repeat processes that further bolster time savings. For example, he says, “If you know that there’s a way to make something that usually takes five minutes take 30 seconds instead – sure, it's only four and a half minutes difference, but when you multiply that by, for example, 350 specification data sets, it's a lot of time. And that time could be a barrier for people.”
Park’s scripts now enable scientists to hit a single button that will automatically pull in and visualize the correct data. Even on a small scale, automation has brought outsize gains. “Standardization has definitely been a big benefit for us,” Park says. “The more standardized we get [across R&D, operations and quality], the better off we are as a company in making decisions.”