50% of the Data We Need: A Food Manufacturing Plant Manager's Honest Take on Continuous Improvement and AI

Data shortage isn't a challenge faced by most food manufacturing plants. Instead they have tons of data sitting in spreadsheets, physical logs, maintenance systems and are stored with heads of people who have worked on the floor for decades. The problem is that there is a lot of information at one place and very few teams at plants have enough time on their hands to analyze all of this data.

And one man all too familiar with this problem is JR Rodriguez.

JR Rodriguez is an assistant plant manager at HB Specialty Foods in Nampa, Idaho. Here the company produces industrial-scale batters for products such as French fries and breeders for chicken and fish.

Rodriguez has spent more than 20 years in manufacturing and his experience as a welder makes him a perfect fit in the field. He started his journey as a welder before moving through maintenance, production and plant leadership.

Today, his work gives us a practical view of continuos improvement in food manufacturing and lean manufacturing in the food industry. Thanks to AI, Rodriguez and his team are able to analyze information much faster. However, the plant is still building the processes and the data foundation that is required to get more from it.

In addition to this, Rodriguez is launching an operational excellence consulting practice with his wife. This venture helps him gain two perspectives: implementing continuous improvement (CI) inside an operating food plant and building frameworks to help other manufacturers improve their own processes.

Continuous Improvement in Food Manufacturing: Safety First, Then Cost

Rodrigues is very clear about his decision on which continuous improvement problems deserve primary attention. For him, safety comes before anything else.

If data related to the plant points to an inherent safety risk or something that could lead to a dangerous incident, that problem moves to the front of the queue. There is no question about waiting for a calculation of whether fixing it will improve throughput or reduce operating costs.

What follows closely behind is food safety.

In our interaction, Rodriguez revealed that HB has a dedicated quality team that handles its quality as well as regulatory responsibilities. Rodriguez approaches food safety primarily from the mechanical side. His responsibility involves looking at equipment failures, maintenance history and incidents that could create an opportunity for foreign material to enter the product.

His background therefore gives him a particular perspective in relation to operational excellence in food manufacturing. Prior to this, Rodriguez studied welding and metal fabrication, ran a welding shop, and worked his way through maintenance before gradually transitioning into food manufacturing and plant leadership.

This experience taught Rodriguez to realise the connection between the reliability of an equipment and production. For example, if there is a mechanical failure near food that is exposed, it no longer remains a maintenance problem but becomes a food safety risk.

Once primary concerns such as personnel and food safety are addressed, the improvement-related priorities shift towards cost, throughput, and making the operation more profitable.

But Rodriguez adds another measure that can disappear from a traditional efficiency discussion: whether the improvement makes work easier for the people handling it.

This emerges as an important principle for lean manufacturing in the food industry. A process can look efficient on a spreadsheet and still lead to unnecessary obstacles on the floor.

Through our interaction, what also became evident was Rodriguez's openness about HB's current maturity. He maintained that the plant is still in the early stages of making continuous improvement a concentrated focus of the organization. Much of the work remains reactive because the team is still caught up in digging through historical information and identifying where the challenges have already occurred.

How AI Speeds Up Lean Six Sigma Analysis in Food Production

Amongst Rodriguez's most practical AI use cases is not particularly dramatic, in fact, it saves time.

There is a lot of prep work that can go into a traditional Lean Six Sigma analysis before a CI leader can even get to the actual improvement work. The data from the production process must be collected and organized. Analytical work such as Pareto charts, control charts, scatter plots or root cause diagrams can take time to prepare manually.

Rodrigues uses AI tools to shorten this part of the process.

The important point is not that AI replaces the CI practitioner but changes where the practitioner's time goes.

Instead of spending hours preparing data and building analysis, more time is spent interpreting the results, deciding what deserves attention and getting onto the production floor to make the improvement.

For lean manufacturing in the food industry, that can be more useful than adding another dashboard. Rodriguez also believes that AI can help with something that many CI discussions overlook- change management.

When a proposed change affects someone's workstation or workflow, it's not always enough to explain it verbally. AI tools can help create a document, visual, or simple rendering that shows employees what the proposed change could look like.

This gives the teams on the floor something concrete to react to. As Rodriguez explained, "Being able to give feedback on something, especially a design that's going to impact you long-term, allows for a little bit easier alignment and acceptance out on the floor."

The same context is needed for OEE food manufacturing data. Rodriguez is aware of OEE systems that monitor throughput and waste, but he says those metrics tell only part of the story at the plant.

Batch counts per hour, cycle consistency, cases produced, equipment speed and even if a bearing is maintaining consistent RPMs can help explain why a larger performance metric changed.

AI is able to speed up the analysis. The operational team still has to figure out what the pattern means.

The Tribal Knowledge Problem: Manufacturing's Hidden Data Gap

The biggest constraint on more advanced AI-assisted CI at HB is not necessarily the AI itself but it is getting the plant's information into a usable form.

Rodrigues described much of the current environment as "tribal knowledge oriented." There are spreadsheets, physical documents, computerized maintenance records, and years of operational knowledge that were never created as one connected data source.

Hence spreadsheet fragmentation is one part of the problem. Different records may have been created for different purposes at different times. When Rodriguez wants to understand a broader operational pattern, relevant data points may first need to be pulled together into a consolidated spreadsheet.

Physical documents create another gap. Some equipment and maintenance information still exists in paper records. When that information matters to an analysis, the team may scan the documents or manually extract the useful data points before combining them with other records.

Then there is manufacturing tribal knowledge.

Experienced operators and maintenance employees learn things about a process that may never appear in a formal record. They know when equipment behavior feels unusual, which adjustments have historically worked, and which small changes tend to precede larger problems.

The knowledge exists, but if it remains only in people's memory, there's no way a software can analyze it.

That is why a tribal knowledge manufacturing solution is partly a documentation problem. Useful physical records need to become accessible and spreadsheet information needs to be consolidated where appropriate. Important operating knowledge needs to become documented process knowledge instead of depending entirely on individual experience.

The other half is not necessarily non-existent, some of it may just be sitting in paper records. Some may be generated by equipment but not currently collected. Some may exist in isolated files, and some may still live with the people who understand the process because they have worked with it for years.

Until that information is captured in a form that can be analyzed, AI can't provide a complete picture of the plant.

Predictive Maintenance in Food Manufacturing: The Gap Between Software and Reality

HB already uses computerized maintenance management software to record work and generate reports. Rodriguez said the system has preventative and predictive capabilities, with newer AI functionality becoming visible over roughly the past six months. The plant has not fully tapped into those capabilities yet.

For Rodriguez, the next step in predictive maintenance food manufacturing is not simply installing more technology. The processes generating the data first need to become consistent.

If a process varies depending on the operator, shift or undocumented workaround, the data can tell an unclear story; more sensors do not automatically solve the problem.

Rodriguez's longer-term vision goes deeper than standard OEE food manufacturing measures such as waste and throughput. He is keen to see how many cycles happen each hour, how many batches are produced, how many cases come off the line, whether a bearing is maintaining a consistent RPM, and how those signals change minute by minute.

Each data point tells part of the story and together, they could create the cohesive plant view Rodriguez believes is still missing. Today, his estimate remains about 50% of the required data.

Getting closer to 80% or 90% will require process standardization and better data capture infrastructure. AI may be able to analyze the information quickly, but the plant first has to produce information worth analyzing.

CI Before AI: The Sequence That Makes Adoption Work

A principle runs through Rodriguez's entire experience that AI does not shortcut continuous improvement. Instead, it accelerates parts of CI that have traditionally been slow and manual.

AI can analyze data faster and make proposed changes that are easier to communicate. Better data collection may eventually support more predictive maintenance but processes still need to be standardized, disconnected information consolidated, and tribal knowledge documented.

Rodriguez's emerging consulting work follows the same logic; he is building frameworks for manufacturers working through process standardization and operational excellence before trying to move further with technology.

The broader AI market gives that argument additional context. BCG reported that only 26% of companies in its 2024 research had developed the capabilities to move beyond proofs of concept and begin generating AI value. The research also found that people and process issues account for most AI implementation challenges.

For a plant manager, the implication is practical- the AI tool is only one part of the work. CI comes first and the data foundation follows. Therefore AI becomes useful when there is a reliable operation underneath it.

Frequently Asked Questions

Bio Block

About the Interviewer