Beyond physical automation: the information challenge

Traditional approaches to factory automation have concentrated primarily on physical processes — robotic assembly lines, automated material handling and sensor-driven quality control. However, Peak’s experience working with manufacturers reveals that achieving true operational autonomy requires addressing a parallel challenge: optimizing the thousands of complex decisions that orchestrate manufacturing operations using artificial intelligence (AI).

While physical robots can execute tasks with precision, the real complexity lies in the information layer. Every production run, inventory adjustment and supply chain decision requires processing vast amounts of interconnected data. This is something that has traditionally relied heavily on human expertise, intuition and gut feel.

The decision bottleneck in smart manufacturing

Consider a typical manufacturing scenario: demand forecasts shift, supply chain disruptions occur, and production capacity varies — all at the same time. In conventional operations, these situations trigger cascading manual decisions across departments: procurement teams adjusting orders, production planners reshuffling schedules, and inventory managers reallocating stock.

This decision-making process represents a fundamental bottleneck in achieving true manufacturing autonomy. Even the most sophisticated physical automation systems remain dependent on human-driven decisions about what to produce, when to produce it, and how to optimize resource allocation.

The evolution of management theory: from scientific management to AI-optimized operations

Manufacturing’s current transformation echoes previous paradigm shifts in management theory. Frederick Winslow Taylor’s Scientific Management in the early 1900s systematically optimized individual tasks through time-and-motion studies. 

Later, Henry Ford’s assembly line principles revolutionized production flow optimization. The Lean Manufacturing movement, emerging from Toyota’s Production System, extended optimization thinking to entire value streams, emphasizing waste elimination and continuous improvement.

Each of these management revolutions shared a common thread: applying systematic, data-driven approaches to optimize decisions that were previously made through intuition and experience. Today’s AI-powered supply chain optimization represents the next evolution in this continuum, only with unprecedented scale and sophistication.

Where Lean Manufacturing required human practitioners to identify waste and optimize processes through observation and analysis, AI systems can now continuously optimize thousands of interconnected decisions simultaneously, processing variables and relationships beyond human cognitive capacity.

Traditional machine learning and optimization: the foundation layer

The current state of AI in manufacturing planning typically involves two complementary approaches working in tandem:

Machine learning for pattern recognition and forecasting

Traditional machine learning (ML) models excel at identifying patterns in historical data to generate demand forecasts, predict equipment failures, and classify product quality. These systems can process vast datasets to predict future conditions, automating the analytical work that human planners previously performed manually.

Mathematical optimization for resource allocation

Optimization algorithms solve complex constrained problems: determining optimal production schedules given capacity limitations, minimizing transportation costs across distribution networks, or balancing inventory levels against service requirements. These systems codify the decision logic that experienced planners have developed over years of practice.

When combined, these approaches create powerful planning capabilities. ML models generate forecasts and predictions, while optimization engines determine the best allocation of resources given those predictions and operational constraints.

The agentic revolution: autonomous decision making at scale

The emergence of agentic AI represents a fundamental advancement beyond traditional ML and optimization approaches.

While conventional systems require human operators to interpret results and implement recommendations, agentic systems can autonomously execute decisions and adapt their strategies based on real-time feedback.

This cognitive automation layer addresses several critical manufacturing challenges:

Autonomous inventory optimization

Rather than simply recommending optimal stock levels, AI agents can automatically adjust safety stock parameters, trigger purchase orders, and redistribute inventory across locations based on real-time demand signals and supply chain conditions. The system combines ML-driven demand forecasting with optimization algorithms for inventory positioning, then autonomously executes the resulting decisions.

Dynamic production planning

Agentic systems integrate ML-based demand predictions with mathematical optimization for resource allocation, then automatically implement schedule adjustments when disruptions occur. Instead of generating reports for human planners to review, these systems directly update production schedules, reallocate resources, and adjust capacity utilization while maintaining service level commitments.

Predictive service management

AI agents can forecast potential service level issues using ML models, determine optimal corrective actions through optimization algorithms, and automatically trigger those actions without human intervention. This could be adjusting production schedules, expediting suppliers, or even reallocating inventory.

The synergy of traditional and agentic approaches

The most significant value emerges when traditional ML and optimization capabilities are integrated with agentic automation. This combination creates planning systems that are simultaneously:

• Predictive: ML models continuously refine forecasts based on emerging patterns
• Optimal: Mathematical optimization ensures resource allocation maximizes defined objectives
• Autonomous: Agentic capabilities enable real-time execution and adaptation
• Learning: Feedback loops allow the entire system to improve decision quality over time

For instance, an integrated system might use ML to predict demand volatility, optimization algorithms to determine safety stock levels that balance service and cost objectives, and agentic capabilities to automatically implement inventory adjustments across multiple locations as conditions change.

Integrating AI-optimized decisions with physical automation

The most significant opportunity lies in orchestrating these AI-powered decision capabilities with existing automation infrastructure. Manufacturing execution systems (MES), enterprise resource planning (ERP) platforms, and industrial IoT networks generate enormous amounts of operational data.

AI planning systems can consume this information to make autonomous decisions that then trigger actions across both information systems and physical automation.

For instance, when demand patterns shift unexpectedly, an integrated AI system might simultaneously:

• Generate updated demand forecasts using ML models
• Optimize production schedules through mathematical algorithms
• Automatically adjust MES parameters via agentic execution
• Trigger automated material handling systems to reposition inventory
• Autonomously generate and approve purchase orders for critical components

This integration creates a feedback loop where physical automation informs intelligent decisions, which in turn optimize physical operations. This can lead to a level of operational autonomy that was previously simply unattainable.

Learning from management history

Historical management transformations offer important lessons for implementing AI-optimized supply chain decisions. Scientific Management initially faced resistance from workers who feared job displacement. Lean Manufacturing required fundamental changes in organizational culture and thinking patterns. Now, AI-powered planning systems demand new organizational capabilities and management approaches.

The most successful implementations recognize that technology adoption requires parallel evolution in management practices. Just as Lean Manufacturing required training workers in continuous improvement methodologies, AI-optimized planning requires developing organizational capabilities in data management, algorithm governance, and human-AI collaboration.

The path to true lights-out manufacturing

While fully-automated factories represent the ultimate vision, the practical path forward involves progressively automating decision-making processes alongside physical operations. Manufacturers implementing AI supply chain solutions are discovering that cognitive automation often delivers more immediate and measurable value than additional physical automation.

Peak’s work with manufacturing customers demonstrates this principle. Companies implementing AI-driven inventory optimization have achieved 4-8% reductions in total supply chain costs while improving service levels. These kinds of outcomes directly support a business’ broader automation strategy by reducing variability and improving predictability in operations.

Economic imperatives driving adoption

The business case for AI-optimized planning extends beyond operational efficiency. Labor shortages, supply chain volatility, and increasing customer expectations for responsiveness are creating economic pressure for manufacturers to reduce dependence on manual decision-making processes.

Manufacturing leaders recognize that achieving competitive advantage requires operating with greater agility and consistency. AI provides a scalable solution that improves with experience and operates continuously without the limitations of human availability or cognitive load.

Integration challenges and opportunities

Successfully implementing AI-optimized planning requires addressing several integration challenges:

Data infrastructure

AI systems require access to real-time data across multiple operational systems. Manufacturers must invest in data integration capabilities that connect ERP, MES, supplier systems and IoT platforms.

Algorithm governance

Organizations need frameworks for monitoring and managing AI decision-making processes, ensuring alignment with business objectives and regulatory requirements.

System orchestration

The most complex challenge involves orchestrating decisions across multiple systems and processes while maintaining operational stability and safety.

The future of manufacturing autonomy

The trajectory toward fully autonomous manufacturing operations depends not just on advances in robotics and sensor technology, but equally on the development of sophisticated AI systems capable of optimizing operational decisions at scale.

As these technologies mature, the distinction between physical and cognitive automation will blur, creating integrated systems where intelligent decision making seamlessly orchestrates physical operations. The manufacturers leading this transformation are those investing in both dimensions of automation. They’re the ones recognizing that true operational autonomy requires both intelligent machines and data-backed, AI-optimized decisions.

Following the pattern of previous management revolutions, the competitive advantage will ultimately belong to organizations that can most effectively integrate these new capabilities into their operational DNA. 

Just as Lean Manufacturing became table stakes for competitive manufacturing, AI-optimized supply chain planning is rapidly becoming essential for operational excellence.

The question for manufacturing leaders is not whether this transformation will occur, but how quickly they can develop the AI-powered planning capabilities necessary to compete in an increasingly autonomous manufacturing landscape. Those who master the integration of traditional ML, mathematical optimization and agentic automation will define the next generation of manufacturing excellence.

The role of artificial intelligence in manufacturing

AI is helping manufacturers navigate through uncertainty, empowering them to be more reactive, more agile, and more consistent with their decision making. However, there’s much more to AI than shiny robots on your factory floor and some of the other applications that first spring to mind when people talk about 4.0. 

In fact, it’s some of these common connotations that lead many manufacturers to view AI as something not entirely tangible. Many of those we speak to claim that AI and 4.0 sound positive and are on their future roadmap, but their data, their current technology landscape, and their thinking are not in a place yet to even consider AI.

While this attitude is legitimate, and AI can be indeed seen as a part of Industry 4.0, I think a more fruitful way of conceptualizing AI in the manufacturing sector is more in the lineage of how manufacturers now see Lean; something that is the core fabric of their organization, and is a must-have skill set to compete in the modern world.

Consequently, by viewing AI under the lens of a process improvement technique or philosophy akin to Lean, the pitfalls of having AI as an ‘expensive toy’ or delaying an AI implementation can be avoided, and can propel manufacturing into the next competitive paradigm – just as Lean has done.

Artificial intelligence in manufacturing: some similarities with Lean

Through conversations I’ve had with many of Peak’s manufacturing customers, I’ve noticed a number of both surface and deeper similarities between AI and Lean. Here’s some information on just two of those similarities…

Changing the way organizations compete

The introduction of Lean demonstrated that you could follow multiple operational objectives simultaneously, which was contrary to the proceeding logic within industry. I would argue that AI does exactly the same, potentially much faster and much more dramatically than Lean. At a high level, you can basically say that with AI, you can switch from one operating objective to another in more or less a flip of switch! This allows organizations to operate the most efficiently, whilst still achieving end user value.

Enabling humans to work smarter, not harder

Another key feature of the nature of Lean is to cognitively reduce the burden on organization members. The classic example of this is the use of visual management techniques such as Kanban systems to signal pull within the shop floor.

Lean designs systems around shop floor operators and does not give them hard rules to follow and then berate them for not doing it. We have seen the use of AI to follow Lean’s lead of reducing cognitive strain, but in a slightly different way. It takes complex data and combines it in such a way that it gives a simple output to the business user for them to make the right decision.

Further reading on this topic

In my view, AI is analogous and complementary to Lean thinking in manufacturing. Although technological advances have been key throughout the sector’s history, they’ve all led to the enablement of new ways of thinking; new ways of looking at production systems and the way manufacturers are organized by paradigms like Fordism, Taylorism, and Lean. Decision Intelligence is the next chapter of that journey.

My new whitepaper, AI in the lens of Lean, not Industry 4.0, is now available to download if you’re interested in exploring this topic in more depth. In the guide, I delve deeper into similarities between AI and Lean, and introduce you to the benefits of Decision Intelligence – the commercial application of AI to optimize decision making in businesses. I hope you find it interesting and look forward to hearing your thoughts!

As the COVID-19 pandemic continues to disrupt the majority of industries, its impact on supply chains has been nothing short of seismic. As teams face increasing pressure to make the right decisions at the right time, squeezing every last drop of insight and information out of your vast amounts of data has never been more important.

These tricky times call for a new approach to data-driven decision making, and there’s now a real need for supply chains to focus on, what we call, the greater data ecosystem.

Yes, you can make effective decisions based on data from your current systems, or by joining up a few previously-siloed sources across your organization – but there’s potential to go even further than this. The more data you have to play with, the more informed your supply chain decisions will be. Let’s take a closer look at a few different data sources that Peak is exploring with our customers to drive intelligent supply chain decisions.

Customer systems and D2C data

This is all about linking data from your own supply chain systems with your customers, as well as consumer behavior data points (for those with direct-to-consumer channels.) For instance, this could be their ERP system or even the logistics systems between your business and your customer.

For example, suppose you’re a consumer packaged goods business or a manufacturer, with a better handle on Electronic Point of Sale (EPOS) and any other sell-out data from your customers’ systems, you can better predict what demand is going to be like, and better understand their stock levels in order to help you anticipate yours. You could even factor into account things like receipts data; what baskets are shoppers generally buying together, and how can this help you better anticipate how groups of products are going to sell together.

This closer relationship with your customers’ systems allows you to better serve them and increase efficiency and anticipate demand fluctuations. In short, it’s all about creating more competitive supply chains which are more cost-effective, with better service levels and a more accurate view of demand.

Supplier systems

By leveraging data points from your suppliers’ systems, you can plan ahead in the most efficient way and execute an effective just-in-time (JIT) inventory management strategy, holding minimal assets to save cash and space whilst still fulfilling customer demand. Our customers who are employing this methodology are able to understand when a supplier is going to deliver, to what location, and anticipate the arrival of goods and raw materials whilst also better understanding working capital implications.

Environmental and global data

Don’t underestimate the power hidden away in external, third-party data sources and the impact it can have on your supply chain decision making. Think about the ways your business can utilize, let’s say, macroeconomic data to understand what could be driving issues connected to supply and demand. Yes, we immediately think of things like GDP, or maybe even exchange rates, but there is now a plethora of data out there, that may be more industry and company-specific, that helps predict demand or implications for business performance. For instance, a sad but apposite data feed could be the level of COVID-19 near a supplier, which may hamper their ability to supply. Potentially, AI and machine learning could help understand the impact of these incidents with supply performance.

Network Data Sharing

This one may seem a little more blue sky for many businesses at first, but the benefits can be enormous if you can imagine not just working closely with your retailers, but also with competitors and those providing similar products – allowing you to gain a unique view of exactly what is happening across the rest of the market. This leads to a better understanding of wider trends and the ability to make better smarter decisions. With a mutually beneficial relationship with the wider network, you can understand supply issues, and work with competitors or neutral parties to deliver better products and services to your customers creating a form of ‘coopetition.’

Introducing a new type of business system

Tapping into the greater data ecosystem and utilizing it in your decision making offers an untold number of benefits for supply chain teams. However, to truly unlock this potential, a new approach – and a modern architecture – is needed.

In the same way that business functions have their own systems of record, the ability to power decision making based on a wide range of data sources hinges on the introduction of a new, centralised enterprise business system. For Peak, this is CODI, our Decision Intelligence system, which gives teams the ability to leverage unlimited data points at scale and speed.

With CODI, we’re able to connect the dots between data points with AI, to prescribe recommendations and actions to help you optimize your decision making across the entire supply chain.

For instance, by feeding external data into both demand and supply planning systems and leveraging it with AI, you can optimize that connection between these two core areas of your business. Not only does it allow you to better sense demand with a higher degree of accuracy, but also enables a better understanding of how supplier and operations constraints are affecting supply – automatically making micro-adjustments to optimise the way demand is being fulfilled.

There’s no denying that trade-offs are part and parcel of any supply chain strategy at some level, despite the aspirations of concepts such as ‘ambidexterity.’ However, there is another way of looking at them, and a different approach you can take. It involves placing an added focus on switching between these focuses, essentially decreasing the time between these states to give agility, enabled by artificial intelligence (AI).

The continued uncertainty that clouds the supply chain sector is resulting in teams being forced to make extremely complex, and potentially costly, decisions. Given current levels of demand, what products should you use? How often should you run certain machines? Which suppliers should you use? How much of certain materials are you going to order? These are all hugely complicated decisions, and it takes vast amounts of management resource and time to make the correct ones.

Of course, there are tools and systems out there that can help. However, in my experience, I think it’d be fair to say that the vast majority of information systems are built for more ‘steady-state environments.’ Given the huge amounts of volatility we’re currently seeing, these systems are under more pressure and aren’t built for this kind of scenario. This results in more stressed supply chain teams, spending a lot of time running around and trying to make the right trade-off decisions with little more than gut-feel to rely on.

Investing in agility

AI can be a real game-changer for supply chains in the current climate. It can power smarter, data-driven decision making, allowing you to ensure your supply chain is as agile as possible. Being agile and dynamic enough to react to volatility, for me, offers benefits that could be potentially more fruitful than trying to understand that volatility in the first place.

Yes, even with AI, there will still be some trade-offs that you need to make. However, rather than having everything set in stone from the beginning, you can be more flexible and reactive. For instance, you may initially optimize for, say, resilience, revenue or profit. With AI, to change those optimizations is essentially just a flick of a button or a case of rerunning a particular model. As the environment continues to change, you can adjust the way your operations are running and adapting. This aids decision making around the configuration of supply, and takes away some of the hours you’d otherwise be losing stuck in spreadsheets, combing through lots of data manually.

Take demand forecasting as an example. This is a hot topic at the moment, and many may find themselves reaching towards a new demand forecasting solution in a bid to better understand current volatility and make the correct decisions around inventory and how much stock to hold.

For me, this is potentially the wrong approach, since a wider perspective is required. The brutal truth is that there are some aspects of demand that are simply just random, that no system in the world can help forecast. You can have the most accurate forecast possible, but levels of uncertainty will always be there. Rather than focusing on just one piece of the puzzle, the focus should be on agility as a wider concept.

What does this mean? It means looking at things like safety stock levels and making them more dynamic while understanding profitability and trade-offs in the stock you’re holding to try and make your organisation more optimized for the external and internal business conditions it finds itself in. So, rather than putting in place fixed rules based on a forecast or supply assumptions that are updated infrequently, for a particular context, you’re instead constantly reassessing what you should be holding based on multiple factors; potential demand, volatility, profit, and how your supply base is behaving. With AI, you can also predict supplier behaviours and feed this data back into your safety stock to add even more intelligence.

To summarize, instead of teams having to deal with multiple systems and spreadsheets which are sucking up all of your management time and resources, it’s possible to model different scenarios and ask AI some crucial questions; ‘Given this level or certainty of demand, what’s the optimal level of stock we’ve got to hold? What’s the optimal shift patterns or production runs we’ve got to make? What’s the minimum order quantity?’ With this approach, and by investing in agility, you’ll alleviate all of those trade-off concerns whilst powering reactive, dynamic decision making across your entire supply chain.

The good news is that retail businesses have done the hard work for you, and there are certainly some valuable lessons to be learned from those businesses that do D2C well. As we all know, e-commerce adoption has grown significantly in retail in the past two decades, with some rather high profile winners and losers. 

Those brands who are winning – look at fast fashion giants such as ASOS and boohoo as examples – have spent years perfecting their e-commerce offering to ensure they deliver the ultimate customer experience. The key ingredient to this is data, and being able to squeeze every last drop of value from data from a wide range of sources. When this is done well, you can power smarter decision making across the entire CPG value chain, from customer acquisition through to order fulfillment.

However, making sense of a vast amount of data sources, particularly when you’re dealing with new D2C channels and added complexity throughout your business, isn’t easy. That’s why the leading retail businesses are utilizing artificial intelligence (AI) in order to generate valuable insights from their data to help them compete and win in the new era.

 CPGs looking to launch or enhance a D2C offering can apply the same techniques to their data, too. As one example, AI can help you improve the accuracy of your forecasting in order to optimize decision making, allowing you to know when and where to streamline inventories whilst still ensuring you meet demand across all of your channels. 

A common stumbling block for CPGs looking to start D2C is often initial customer acquisition. According to Ali Holmes, Senior E-commerce Director at PepsiCo UK, the “most significant challenge building a D2C model from scratch is economics. The cost of customer acquisition and bringing consumers online is initially expensive.”

However, you can follow in the footsteps of the e-commerce giants who are optimizing their advertising and marketing strategies with AI, to drive higher ROAS by targeting the right customers with the right products at the right time, via the right channel (take a look at our work with Footasylum for more on that!) Let’s not forget about AI’s unique ability to blend data from sources that don’t typically mix, too – this holistic view also allows you to make smarter decisions around pricing, margin protection and order fulfillment, to name just a few.

CPGs entering the D2C market are in a unique position to build an effective, efficient model from the ground up, rather than having to rip and replace or optimize their existing systems. By investing in the right technology now and following in the footsteps of the most successful retailers, you can skip the mistake-making period that they had to deal with in the early years, and capitalize on the best practices they’ve already laid out for you – all while using AI to accelerate your competitive advantage.

Why D2C, and why now?

There are several benefits to introducing a D2C offering into your CPG business. The COVID-19 pandemic has seen e-commerce usage skyrocket, with consumers heading online to order everything from pet food to toilet paper. Fuelled by the images of empty supermarket shelves that dominated the media at the height of lockdown, many established CPG businesses – think the likes of PepsiCo, Nestle and Heinz – all launched D2C offerings as part of a seismic shake-up of their existing channels and a bid to satisfy the demands of consumers. 

These brands capitalized on a big opportunity to develop closer, one-to-one relationships with their product’s end consumers. This is great for building brand loyalty, while shorter marketing channels with less intermediaries (i.e. retailers or marketplaces) between you and the consumer means you can potentially keep a bigger chunk of the profits, too. However, the biggest benefit of going D2C is being able to gain access to – and keep! – a wealth of customer data, which has long since been a desire of more visionary CPG companies.

Traditional CPGs have had to work with  supermarkets and retailers when it comes to data access, whereas a D2C offering can act as a key source of insight . The PepsiCo’s of this world are using these new platforms to A/B test offers and promotions, conduct important customer and user experience research, and using their websites as a testbed for new products, ranges and bundles. The majority of CPGs who have embarked on a D2C journey see it not as a new primary sales channel, but more a place to collect valuable data to inform their future decision making.

However, starting a D2C offering from scratch isn’t easy, and making sense of this wealth of new data at your disposal is easier said than done. Thankfully, there are lessons to be learnt from retail businesses – particularly those rapidly-growing, data-first e-commerce brands – who have spent years perfecting D2C and offering an exemplary customer experience.

We’re currently working with a number of leading CPGs to help them make a success of their D2C journey, introducing Decision Intelligence into their operations to maximize marketing and acquisition efficiency and make data-driven decisions to help streamline the supply chain. 

Although D2C for CPG is still nascent, we’re already exploring some really exciting use cases for Decision Intelligence in a D2C setting for some of our consumer goods customers. Let’s take a look at a few examples of how an AI-driven approach to your D2C data can positively impact your entire consumer goods business…

AI for demand sensing

By getting closer to consumers and encouraging people to buy directly, CPGs are in a brilliant position to collect vast amounts of incredibly powerful data. AI can help you paint a clearer picture of what people are buying, how they behave on your website, who’s likely to make a purchase and, crucially, what content will stimulate them to buy. The power of personalization – ensuring consumers are served the right message, with the right product, at the right time – is driven by data that is worth its weight in gold. 

But it doesn’t stop there. AI can also automatically feed this data back into your demand forecasts, augmenting it with your traditional sources meaning that you’re no longer forecasting based on just historical demand. With a more accurate forecast, and with a clearer idea of who is likely to convert and purchase a particular product, you can introduce this information into both your campaign and base forecasts.

AI for demand shaping

Traditionally, a CPG that is overstocked with inventory will rely on discounters to help them shift any unwanted stock. However, D2C offers businesses an exciting opportunity to also target specific segments of consumers directly to help clear inventory at a better price point. In short, AI for demand shaping enables you to activate, or deactivate, your new customer segments depending on what inventory you have to play with, through targeting consumers with relevant products in a highly sophisticated, highly personalized way.

This AI-driven approach has the potential to revolutionize your inventory strategy, which we feel is one of the most interesting aspects of D2C. It’s particularly prevalent for those CPGs who are dealing with perishable goods, who serve retailers that have relatively erratic levels of demand. In these circumstances  you may have the problem of often being over-stock, or you may under-stock and have supply issues when the unpredictable retailer comes back asking for more product.

However, with an AI-driven data-led approach to your inventory strategy, you could ensure you have enough quantity to always meet the retailers’ requirements, with your D2C channel effectively acting as a safety net; because you know that you’ll be able to quickly activate your highly targeted micro segments who always buy that particular product. Essentially, what we’re saying is that you can employ different inventory strategies to ensure you’re overservicing your primary channels, at no obsolescence cost of inventory. This is very much blue sky thinking at this stage, but the potential is certainly there. Similarly, a better, data-driven understanding of costs on the supply side can be taken into account against the cost of acquisition, retention and servicing customers.

Our vision at Peak is to enable CPGs to transform from a reactive to a proactive entity, built upon predictive capabilities. Peak’s Decision Intelligence platform acts as an intelligence layer that sits across your systems, at the center of your business, using AI to power optimization and decision making across the entire organization. Get in touch with one of our expert team to find out more.