Most manufacturing companies started the year with AI on their agenda. Their goals range from evaluating initial use cases to demonstrating measurable impact in live operations. Now it is September, and after three quarters, progress in many plants has fallen short of expectations.
There is a reason why manufacturing often lags behind other business functions. In marketing, HR, or customer service, a chatbot or relatively simple AI agent can support a wide range of useful applications, often within days. In production, however, the knowledge that AI needs is embedded in transactional data from multiple systems and distributed across thousands of orders, operations, and machines. The underlying processes are physically interconnected, and those dependencies are not documented in a format that a chatbot can simply read. Before AI can make reliable statements about a plant, substantial groundwork is required. In conventional AI projects, this is where months are lost.
That does not mean the goal is out of reach. The remaining quarter is still enough to bring the first AI applications into production and finish the year with tangible results rather than another status report.
Custom Data Models and Use Cases Make Manufacturing AI a Two-Year Project
A conventional AI project often starts with use case workshops, develops into a data platform project, and ends 12 to 24 months later with a pilot of limited practical use. Two demanding workstreams account for most of that timeline.
The first is the data foundation. Whether production data comes directly from ERP, MES, and machine systems or is already available in a company data platform or Production Cloud, it initially provides only a fragmented picture of production. Each system records data according to its own logic and for a specific purpose. Bringing these sources together is the first part of the work, and some companies have already completed that step. But data integration alone solves only part of the problem.
What is typically still missing is operational context. The relationships between products, orders, materials, machines, process steps, and material flows need to be understood. Without that context, AI can process the available data, but it cannot determine what is actually happening in production. In conventional projects, building a consistent, contextualized production model therefore consumes a substantial share of the budget before AI is used at all. In many cases, this is also where many AI projects stall or fail altogether, as the complexity of production data proves too difficult to resolve.
The second challenge begins with selecting the right use cases. Relevant problems in production are usually closely tied to specific processes, equipment, products, and operating conditions. Defining an application that delivers genuine operational value while remaining technically feasible is therefore considerably more demanding than in many other business functions.
Once the use case has been defined, it still has to be developed into a functioning AI application. Connecting a language model to a few data sources is not enough. The application must understand the relevant process logic, combine the right production data and analyses, incorporate the necessary domain knowledge, and deliver reliable results under actual operating conditions.
The result is often a highly specialized solution. An AI application developed for one process or production area cannot simply be transferred to another use case, another area, or another plant. Process logic, data structures, and operating conditions often differ too much. Much of the work therefore starts over with the next application. Conventional AI initiatives tend to scale one use case at a time, making broader rollouts slow, expensive, and resource-intensive.
ONIQ Builds the Foundation for AI in Weeks
ONIQ's Manufacturing Excellence Software significantly shortens the path from production data to operational AI. It brings the required production data model and ready-to-use AI applications together in one product, so companies do not have to build either from scratch.
ONIQ connects to the company's existing infrastructure within a few weeks. This can include ERP, MES, and shop floor systems as well as data lakes, Production Clouds, and industrial IoT platforms. The data can come from commercial software from different vendors or from platforms developed in-house.
From the relevant production data, ONIQ automatically creates the Digital Value Stream. Rather than representing isolated data points, it places them in their operational context. Product, orders, materials, operations, machines, inventory, routings, bills of materials, quantities, and disruptions are linked so that the actual flow of production becomes visible, including how an event at one point affects the downstream value stream.
The model is continuously updated with new data and represents both the current state and the historical development of production. It provides precisely the consistent production data foundation that conventional AI projects would first have to build from the ground up. In ONIQ, that foundation is already part of the software.
A second, deterministic analytics layer operates on top of the Digital Value Stream. It calculates production KPIs, provides specialized analyses and deep dives, and makes causes and relationships across the value stream transparent. Teams can see where performance is being lost, how WIP and throughput times are developing, and where the greatest improvement potential lies. Results can be traced back to individual orders, operations, or machines.
These analyses are deliberately not generated by an AI-based language model. They are calculated deterministically by the ONIQ software. KPIs and calculations used for production management and improvement need to be reproducible, traceable, and auditable. A language model alone cannot guarantee that. Instead, AI applications must operate on top of a production foundation that is already structured, contextualized, and analytically robust.
Built-In Production AI Agents Accelerate the First Applications
This foundation is what makes reliable AI in production possible in the first place. Language models can analyze complex relationships, but they need the right context and process understanding to operate reliably. Where a robust model of reality is missing, they can fill gaps with assumptions that sound plausible.
The Digital Value Stream and the analytics built on top of it close this gap. AI has access to a complete, continuously updated representation of the production processes as well as validated analytical results. Its conclusions can therefore be based on operational facts rather than plausible-sounding assumptions.
Once the Digital Value Stream is in place, production teams can start working with their first AI applications immediately. ONIQ includes preconfigured Production AI Agents for common tasks and problems in Performance Management, Value Stream Management, Shop Floor Management, and Planning. Each agent understands its task, accesses the relevant production information and analyses in ONIQ, and comes with embedded manufacturing and Lean expertise. Initial agents can be set up within hours, tested on the plant's own data, and put into operational use.
A Production AI Agent takes ownership of a defined task and works through it independently from start to finish. It retrieves the required information, investigates causes and relationships, evaluates what it finds, and derives well-founded recommendations. The agent can be run once or on a recurring schedule, taking over investigations that teams would otherwise have to repeat manually time and again. Instead of simply being notified that a KPI has crossed a threshold, the team receives the result of a structured and thorough investigation.
The agent's result is also the starting point for further analysis by the team. Users can follow up directly in AI Chat, asking questions in natural language, or move into the relevant ONIQ deep dives with the appropriate analyses and visualizations. They have access to the same production information, analyses, and KPIs that the agents use. This makes it possible to trace every recommendation back to its underlying causes, validate it, and investigate it in as much detail as required.
With the Production AI Agents already integrated into ONIQ, the second major workstream of conventional AI projects also becomes significantly shorter. Much of the analytical work that would otherwise go into specifying, developing, and validating an individual AI application is already built into the ONIQ agents.
Production Teams Build Their Own AI Agents in Natural Language
The Production AI Agents included in ONIQ already cover many common production tasks. The possibilities expand further when teams create their own agents or extend existing ones to reflect their specific requirements. Expert knowledge that may previously have existed only in the heads of experienced engineers and managers can become part of a functioning AI application within a short time.
Take an AI Agent for bottleneck management in high-mix production as an example. The output of an entire factory depends on keeping its bottleneck supplied, while countless orders, far more than anyone can track manually, move towards it through different operation sequences with varying remaining lead times. An agent set up for this task connects the current WIP with historical lead and waiting times and with the planned throughput at the bottleneck, forecasts which orders will reach it and when, and projects how its material supply level will develop over the coming days. A looming material shortage becomes visible days before it would stop the bottleneck machine, early enough to adjust priorities and production planning.
To create an agent like this, a user can start with the existing agent for bottleneck identification and describe the required changes or extensions in natural language. ONIQ provides the production model, operational context, manufacturing knowledge, and analytical tools required to execute those instructions. The resulting agent can be created and put into use quickly.
Execution schedules and notifications can also be configured in the software, allowing the agent to proactively report back whenever it detects a change in the bottleneck situation. Once configured, it repeats the analysis reliably and reports its findings directly in ONIQ or by email.
Over time, the first results from preconfigured Production AI Agents can be followed by increasingly specific applications that capture the knowledge and problems of the individual plant. What begins as a fast entry point can develop into a growing set of AI applications tailored to the analytical work of the production team.
Get the First AI Applications Live Before Year-End
For companies aiming to have their first AI applications live before the end of the year, an ONIQ Pilot in a selected production area provides a practical starting point. Depending on the existing system landscape, ONIQ can either begin with data exports from the source systems, which can typically be provided without major effort, or connect directly to live systems if that is the faster route. In both cases, ONIQ creates the Digital Value Stream and the first agents from the plant's actual data, so the team works with its own production reality from the beginning.
The Pilot also provides the evidence that matters to management and IT. Within a few weeks, teams can see in their own production area which tasks the agents can take over and what their results are worth in day-to-day operations. Companies that start with data exports can then move to a live connection on that basis. This is typically set up within a few weeks because the Digital Value Stream already exists and only needs to switch from data exports to live data.
Hosting does not have to slow down the start either. ONIQ is available as Software-as-a-Service (SaaS) in the ONIQ Cloud and can therefore be used without setting up additional infrastructure. If production data needs to remain within the company's own environment, ONIQ can also run in the customer's cloud infrastructure.
By year-end, this creates a tangible result. Agents working on real production data, findings already feeding into shop floor meetings and performance reviews, and a demonstrated business case for the next steps, from establishing the live connection to rolling the solution out across additional production areas.
If you want to test this in your own production environment, bring us one question that repeatedly costs your team time. In a short conversation, we can determine what a Pilot would look like, which data it requires, and whether results before year-end are realistic in your case.