Many retailers and manufacturers can’t get past the AI “pilot plateau.” They’ve shoveled money into projects but can’t scale them into tangible, long-lasting results. This plateau makes it impossible to realize the financial potential of current AI initiatives or pave the way for emerging technologies like assistants and agents.
In fact, MIT reported that 95% of gen AI pilots fail, and according to Gartner, 40% of agentic AI projects will be consigned to the dust heap of failed implementations by the end of 2027.
At RELEX, we’re developing a multi-agent AI system, a scalable, adaptable solution that helps companies achieve fast ROI and drive long-lasting value.
Our AI-native platform increases the longevity of tech investments by incorporating technologies already proven to scale and a steady stream of innovations. It provides AI governance by design to boost confidence in AI-driven decisions, and it unifies data and teams to overcome costly planning misalignments.
Now with emerging agentic capabilities that increase accuracy, efficiency, and reaction time, RELEX helps you move past the desolate pilot plateau, realize AI-driven benefits, and amplify them across your business.
Why do so many AI implementations fail?
AI raises eyebrows and concerns because there are so many ways to mishandle its implementation. Companies across industries consistently find themselves up against the same obstacles and delays until finally being forced to cut their losses and abandon projects altogether.
Before we can solve the pilot scalability problem, we have to identify the causes. Why do these projects stall?
Ambiguity results in poor AI strategies
C-level leaders are under immense pressure to adopt AI. They’re handed a budget and an ultimatum: “Use AI. Make it successful.”
Many business leaders don’t even know where to start. Without clear definitions and use cases for the different types of AI, it’s impossible to know how to apply them.
Plus, the speed at which this technology is evolving makes it difficult to pinpoint what AI implementations are immediately feasible and how best to prepare for the next innovation. The murky line between technology that’s a real differentiator and what’s just a distraction contributes to the unfocused strategies plaguing so many businesses and making it impossible to determine an AI strategy that mitigates risk while ensuring long-term value.
Generic solutions prove that AI isn’t one-size-fits-all
In the expanding field of AI, many solution providers themselves just don’t have the expertise to handle industry-specific planning needs. They struggle to build tailored technologies and processes simply because they haven’t been doing it long enough and don’t understand the nuances of retail and manufacturing.
They also don’t have a solid foundation of specialized AI tools and best practices underpinning their solutions, resulting in faulty decisions, poor performance, and wasted time, money, and potential.
Siloed, outdated planning systems fracture organizations
Too many companies subsist on the scraps of data and “insights” gasped out by spreadsheets and legacy systems – the technological vestiges of an era when digital transformation was still a new concept.
These outdated solutions don’t have machine learning for complex calculations or nuanced pattern recognition, and they’re certainly not built to provide gen AI assistance or agentic decision-making. Because they haven’t been incorporating innovations over time, AI is just a sticky note tacked onto a system that can’t truly integrate and accommodate it.
Unfortunately, companies often have many such solutions strewn across their organizations. These solutions aren’t connected to each other, so planners end up making siloed decisions with siloed data. In a world where what you don’t know can hurt your bottom line, this tunnel-vision planning turns these legacy systems into financial sinkholes.
Outside AI-native solutions, AI is just a sticky note tacked onto a system that can’t truly integrate and accommodate it.
Learn more: Why retailers need to unify pricing and promotions
Implementation challenges lead to hesitance and poor adoption
Sometimes, the devil you know is better than the devil you don’t.
Many IT leaders prefer outdated systems with inconvenient workarounds to the daunting task of implementing costly and seemingly unproven innovations. And who can blame them, when 90% of AI implementations fail?
Plus, even a sound solution can be rendered useless by poor onboarding and go-live procedures that leave users frustrated and in the dark. Users need to be able to onboard smoothly and get fast, comprehensive answers to their questions. Poor onboarding and nonexistent user empowerment lead to poor adoption, undercutting and delaying ROI.
Data quality and risks remain a concern
Companies are (rightly) concerned about data and AI.
There’s the typical “garbage in, garbage out” concern around data that’s outdated, incomplete, or just plain wrong. The poorer your data, the poorer your AI calculations.
What’s more, siloed planning systems can leave your organization awash with stagnant pools of data that are all inaccurate in unique ways.
And yet, data is a company’s most precious and protected asset. It’s every CIO’s nightmare to wake up to a string of emails and notifications about a massive data breach. Breaches like this are becoming all the more common, and the consumer response and financial fallout can severely damage a brand’s solvency, reputation, and credibility.
Understandably, IT leaders are wary and scrutinous of new partnerships and third-party solution providers. To them, newer AI systems like LLMs are new and improved ways for employees to accidentally leak vital and sensitive information to the public.
AI governance and transparency are limited
With many solution providers, AI is a “black box”; no one knows exactly what it does, how it works, or why it makes the decisions it makes.
How can companies be sure these systems generate accurate, trustworthy output and not hallucinations? What if a single error is introduced, creating a ripple effect that wreaks havoc on planning decisions?
The lack of transparency and explainability naturally elicits mistrust among would-be adopters of AI-driven solutions, especially when it comes to agentic AI.
Scalability starts with strategy
Before even looking at specific AI solutions, business leaders need to determine a scalable strategy, one that gives them enough direction to make reasonable decisions but allows them to course-correct as the market, business needs, and technologies develop.
That strategy is AI diversification.
AI diversification focuses on:
- Identifying the right types of AI for the right use cases.
- Prioritizing implementations according to business needs.
- Integrating applications over time into one, unified system.
A diverse AI portfolio allows businesses to build their implementations on top of each other in a self-funded cycle. Each implementation drives value on its own, and by investing incrementally, businesses can use the ROI from one project to fund the next project, magnifying and prolonging the benefits of those initial investments.
The RELEX multi-agent system is designed to support this approach, combining AI capabilities, data management innovations, and best practices into a sustainable, cohesive planning system that allows you to build and scale your tech stack. Whether you’re just getting started with AI or already integrating generative and agentic functions, the RELEX platform adapts to your business needs, mitigating risks and incorporating advancements as we develop them.
Plus, we invest approximately 25% of our revenue back into R&D, accelerating our development cycles and passing that competitive advantage along to our customers.
How does the RELEX multi-agent system support AI scalability?
The multi-agent system is an ecosystem of interconnected, collaborative AI agents and solutions that analyze data, diagnose issues, anticipate changes, and make decisions, adapting quickly and automatically to market shifts. It scales because each of the components within this architecture is scalable:
- Agents that understand when, how, and why to act.
- A specialized AI toolkit.
- A shared data pool.
- A unified platform that houses all these capabilities and syncs planning and execution decisions.
Now let’s look at the part each capability plays in defeating the pilot plateau.
RELEX AI agents
AI agents are autonomous entities that act and adapt to reach a goal, either individually or as a network of decision-makers. They run in loops, calling on different tools and even other agents to perform tasks, constantly planning and executing until goals are met. As self-learning systems, they use real-world data to automatically adjust parameters to current market conditions.
There are two types of agents in the RELEX multi-agent system.
The first is Rebot, which was initially launched as a gen AI assistant and the first of its kind in the retail and supply chain space. Using Rebot as an on-demand advisor, companies have seen leaps in productivity, from more efficient daily operations to faster, easier onboarding that accelerates time-to-value.
Now, we’re developing Rebot into an omnipresent agent coordinating all system interactions.
For instance, a planner could ask, “Hey Rebot, what areas are most likely to see lost sales this week, what are the causes, and how can I preemptively address those issues?” Rebot will be able to recommend and take action according to user prompts – and even autonomously, depending on system settings and guardrails.
Rebot can do all of this because it doesn’t do it alone. In the background, it interacts with a network of specialized agents purpose-built for diagnostics, allocation, implementation, and supplier management.
For example, the RELEX Diagnostics agents can identify and explain supply chain issues, then recommend or take corrective action. These are complemented by scenario planning agents that can automate what-if analyses so that agents and users can understand and determine the best responses to market changes.
So, it might look like you’re chatting only with Rebot, but behind the scenes:
- Agent A is routing your request.
- Agent B is gathering information.
- Agent C is analyzing the data.
- Agent D is formatting the response.
This will be a new kind of user experience. Rebot can interact in users’ native languages, removing technical barriers and allowing them to focus on effective prompting rather than system navigation. Instead of clicking buttons and navigating a user interface, users will be able to interact directly with Rebot to coordinate planning and execution, without having to identify and coordinate the appropriate agents themselves.
With Rebot acting as the coordinator, users can focus on asking the best questions instead of figuring out which agent to talk to.
Rebot is also LLM-agnostic, meaning RELEX can transfer it from one foundational LLM to another, depending on developments in the field.
How agents support scalability: Agents increase the speed, accuracy, and sheer number of planning decisions, giving companies a competitive advantage by responding more quickly to disruptions and more readily adapting to customer needs.
How do RELEX agents know what to do?
RELEX agents make better, more scalable decisions because they have two unique advantages when it comes to handling retail and supply chain planning challenges.
The first of these is a machine-readable knowledgebase of documented best practices and industry expertise. This repository, codified over 20 years and continually updated, provides the rich context agents need to make automatic decisions that align with nuanced, industry-specific conditions.
The second advantage is access to RELEX Diagnostics, which measures product performance and identifies the underlying causes of supply chain issues like lost sales. By integrating agents with Diagnostics, RELEX gives agents the real-world context they need to determine and prioritize planning problems and recommend or initiate optimal responses.
Learn more: How RELEX turns AI agents into experts
How Diagnostics and AI-readable documentation support scalability: These agent-enhancing capabilities contribute to more competitive planning decisions and build trust in agentic actions for smoother adoption, increased efficiency, and sustained ROI.
Human-in-the-loop controls: Monitoring and scaling agent activity
For years, RELEX customers have used our configuration kit and business rules engine (BRE) to adapt planning processes to new planning requirements. These intuitive configurability settings give users the control and flexibility to make process changes without expensive coding projects. Agents will use these same systems to modify the RELEX system based on company goals and needs.
At the same time, the configuration kit and BRE provide a level of AI governance and transparency unique to RELEX, allowing users to monitor and control agent actions through a graphical interface.
This signals a shift toward increased human-machine collaboration in which agents like Rebot gradually transform from on-demand consultants into more active planning participants, becoming as autonomous as you want them to be.
User roles change as well. Instead of manual task execution, users will become responsible for setting agentic goals and strategies, determining governance parameters, supervising and directing autonomous systems, and upskilling with hands-on learning in sandbox environments.
How human-in-the-loop control supports scalability: Better user-machine collaboration drives adoption, enhances user control of agents, and gradually increases agent autonomy for faster, better, and increased decision-making that stays ahead of trends and disruptions.
The RELEX specialized AI toolkit
Agents are only as good as the tools they use, and those tools include specialized AI. Specialized AI uses machine learning, mathematical optimization, and heuristics to perform highly specific tasks with superhuman speed and accuracy. It’s the behind-the-scenes powerhouse performing the tasks agents need done.
Got a lot of data? Specialized AI analyzes it to spot patterns and make predictions. Need to understand how a supply disruption will affect fulfillment? Specialized AI is running scenarios. Plus, it runs on a feedback loop, constantly assessing how close outcomes were to its predictions, so it learns and improves over time.
Unlike agents, which require other tools to be successful, specialized AI generates value on its own. This lends to the RELEX platform’s scalability. Specialized AI is built into every one of our planning modules, and they are all connected on our unified platform. So, you can start small, applying those machine learning capabilities to the planning areas that need the most help or promise the largest return. Once you reap those rewards, you can move on to the next project, gradually build out those AI branches until you have a large, interconnected system.
How specialized AI supports scalability: Specialized AI drives rapid, nuanced responses to retail and supply chain dilemmas, baking industry expertise into the solution itself, driving business value, and vastly improving agent performance.
Scalable data management and protection
The RELEX platform provides both data quality and data protection.
The shared data pool of quality, near real-time data feeds every RELEX planning decision and expands to accommodate data volumes at retail scale. RELEX applies data cleansing capabilities to ensure data is cleaned up and stays clean for optimal AI calculations.
In terms of data security and privacy, your company’s data remains yours. RELEX uses pre-trained models and retrieval-augmented generation (RAG) techniques. This RAG approach is a way of optimizing LLM output by pulling from an authoritative knowledgebase outside the LLM’s training resources and using that information to generate a better response, without ever storing or training on that proprietary information. No information is sent to outside AI providers.
Learn more: Straight answers to AI and data security questions
How RELEX data management supports scalability: RELEX ensures data stays updated and secure while making it available across planning teams and scaling it as your business grows. Efficient, data-driven decisions drive customer satisfaction, profitability, and other business KPIs.
An AI-native unified planning platform
The RELEX unified platform is AI-native and allows companies to take an evolutionary approach. It augments specialized AI investments with generative and agentic innovations to improve outcomes, rather than trying to replace those foundational specialized tools – without which agents are rendered useless.
Unified cross-functional planning breaks down organizational silos through agent orchestration. Agents are collaborative by nature, and the RELEX unified platform integrates them with each other and with those scalable technologies like specialized AI and data management across planning functions and teams.
How the unified platform supports scalability: The platform is the realization of an AI-diverse strategy, integrating AI features and incorporating innovations continually to magnify and sustain value across organization.
Beyond the plateau: The benefits of scalable, AI-native RELEX solutions
What does this level of scalability mean for business leaders trying to make the most of their investments? Let’s trace the journey from problem to solution to outcome.
| Problem | The RELEX solution | Outcome |
| Ambiguity and industry “noise” stifles strategic thinking. | AI diversification enables a feasible, incremental approach to AI projects. | AI implementations become both ROI-driving and ROI-driven, scaling with business needs and technical developments. |
| Generic, one-size-fits-all AI solutions can’t support industry needs. | Expertise-infused agents use the specialized AI toolkit and Diagnostics insights to make nuanced, customer-centric decisions. | Companies increase competitiveness, brand loyalty, and profitability. |
| Siloed, outdated planning systems fracture organizations. | The RELEX AI-native platform unifies planning across functions. | Each AI capability complements the other capabilities, generating greater value than if it were operating alone. |
| Implementation challenges lead to hesitance and poor adoption. | IT leaders and users gradually incorporate and automate agentic decisions, increasing trust over time. Rebot helps users better understand their supply chains and strategize more effectively. | Seamless onboarding, implementations, and user-empowerment improve adoption, productivity, and efficiency. |
| Data quality and risks are a concern. | RELEX provides data quality, scalability, and protection, all in one platform. Companies retain control of their data with enterprise-grade AI capabilities and strict data management guardrails. | Companies build more resilient networks with data pools that scale with their business and fuel smarter planning choices. |
| AI governance and transparency are limited. | The configuration kit and the business rules engine give users visibility into and control of automated settings and agentic decisions. | Built-in AI governance transforms user roles and increases agentic autonomy, allowing companies to stay ahead of trends and disruptions. |
The first step: Decide to diversify
Successful AI implementations start with a feasible, flexible strategy. You don’t have to implement AI everywhere all at once. Diversifying your AI portfolio allows you to break implementations into manageable steps, building momentum and generating ROI as you go.
The better you understand your options and the types of AI at your disposal, the more confident you can be in the strategy you build. As we work toward bringing the multi-agent system to our customers, neither we nor our customers are in stasis. All the technology is already in our toolkit and already reaping benefits.
Get the details and specifics so you can see your options clearly, make decisions confidently, and begin building toward your future supply chain success.
Written by
