Building a Formula 1 team from scratch places enormous demands on manufacturing systems. Cadillac Formula 1’s experience reveals how correcting an apparently simple data error can involve reconstructing a complicated history of production transactions.
At Cadillac Formula 1, correcting the records for a single type of component has sometimes taken two weeks. As individual components move through production and are incorporated into larger assemblies, each stage adds further records to their manufacturing history. By then, a change to the original record may require the team to unravel a complicated history of transactions before it can safely put everything back together.
Chris Green, chief information officer at Cadillac Formula 1, described the challenge at IFS Unleashed in Orlando, where the team was discussing its early experience with IFS Intelligence. The newly introduced software is intended to let employees interrogate operational records in natural language and, within defined controls, initiate actions in enterprise systems. Cadillac is using it to identify anomalies in manufacturing information and investigate whether corrections that currently demand considerable manual work can be automated.
The need to correct these manufacturing records reflects the speed at which Cadillac established its operations and how its requirements evolved as the team developed. Cadillac began implementing IFS before receiving permission to compete in Formula 1, establishing the systems needed to manage production, finance and other business activities. As the team expanded and gained experience, it became clear that some earlier decisions about how components were identified and tracked needed to be revisited.
“We had 366 days to get ready for Melbourne,” Green explains. “By the time we received the green light, we had already built our core manufacturing and financial processes in the system, and we needed to move very quickly.”
How manufacturing data becomes difficult to change
Manufacturing records contain information that connects individual components to production orders, assembly histories and the wider business processes that depend on accurate data. At Cadillac, component serialisation is particularly important because the information is used in financial reporting, engineering processes and safety-related activities. Errors in these records can therefore have consequences across several areas of the business, making them increasingly difficult to correct as production progresses.
As the team developed its procedures through its first racing season, its requirements for identifying and tracking components evolved. Serialisation was already relevant to several important business processes, but the rules initially applied to particular parts did not always reflect what the team later needed. Resolving those differences required going back to the records created during production.
“We are using Intelligence to identify anomalies and issues in our data,” Green continues. “As we have built the team, we have learned more about how we want to structure that information. In some cases, we intended to serialise parts and did not; in others, we did not initially realise that we needed to.”
The difficulty for Cadillac is that manufacturing records become increasingly interconnected as components move through production. An error in the original serialisation information can affect subsequent transactions, making it difficult to correct without compromising the accuracy of other records. The team must therefore ensure that any changes preserve the integrity of the manufacturing history.
“Some of that data was put into the system 18 months ago, and those parts have since been incorporated into assemblies and subassemblies,” Green says. “To correct the original data manually, you effectively have to go through the history of the part, reverse it all out, correct the information at source and then play the transactions forward again.”
Reconstructing a history of shop orders
At the racetrack, Cadillac handles preassembled sections of the car rather than every individual component separately. The enterprise records behind those assemblies, however, retain the manufacturing history that created them. An error concerning an individual part can therefore sit several levels below the assembly that employees are accustomed to seeing and handling.
Revisiting those records can become a substantial exercise because each manufacturing transaction depends on earlier activity. Correcting the initial entry is only one part of the work; the system also must reflect what happened to that part afterwards. Green gave an indication of the scale of the task.
“We have examples of one part type taking two weeks to correct,” he adds. “Some of those parts are 20 to 50 shop orders deep in the structure of our data system. We were looking at enormous data correction tasks to get that information right.”
Cadillac has been exploring whether IFS Intelligence can trace those dependencies and carry out the required sequence of transactions. Green describes the intended process as reversing the manufacturing history, correcting the records at their source and replaying the transactions in the correct order. Such an operation would require the software to understand the relationships between historical shop orders, rather than merely identify that an entry is wrong.
The team was still testing the functionality when Green spoke in Orlando. New code had been made available for testing that morning, and he expected the automated approach to offer a substantial reduction in the work required. During the presentation, the possibility of reducing a two-week task to minutes was discussed, but Green did not present it as an established production result. “We believe we will be able to use Intelligence to reverse those transactions for us, correct the data at source and then play all the transactions back in,” he says. “It understands the transactional nature of our data and the history of those parts.”
The work is also revealing additional discrepancies and questions about part classification. Green says the application was helping identify records that needed further attention, as well as cases where existing rules could have been applied more consistently. This provides a way to investigate the scope of the problem while the more demanding process of rebuilding transaction histories is being developed.
Preventing errors before they spread
The possibility of automating historical corrections has prompted a broader question inside Cadillac about preventing errors at the outset. Green linked the problem to differences in employees’ familiarity with enterprise software and the difficulty of maintaining consistent records as departments expand. His concern is how to make accurate data entry routine, regardless of how frequently someone uses the system.
“We have a lot of people who use IFS every day, but we also have people who use it occasionally,” he says. “One of our challenges has been making sure that those people can interact with the system in a way that allows us to maintain high data quality.”
For occasional users, navigating an enterprise application can involve learning processes and terminology they do not encounter regularly. Cadillac is consequently interested in giving employees a more conversational way to perform certain activities without having to navigate each step through conventional screens. Green says Cadillac plans to introduce a conversational interface that allows employees to request parts, materials and services using everyday language, rather than navigating the conventional requisitioning process within IFS.
The proposed interface has not yet been described as a completed deployment. It will extend the use of AI beyond repairing historical manufacturing data towards improving how new requests enter the enterprise system. That could reduce some of the opportunities for incorrect or incomplete information to be introduced, although Cadillac will still need to determine how the approach performs in routine use.
“If we get the data correct at the beginning, everything downstream flows,” Green explains. “If we get it wrong at the beginning, the cost continues to build up until we correct the information at source. We want people who do not use the system every day to be able to interact with requisitioning in a natural-language way.”
Green has also begun using the intelligence platform himself for planning and budgeting, asking questions about information that previously required his team to build reports. Those early experiences have been useful, but the manufacturing application poses a more demanding test because it involves correcting a sequence of consequential transactions. The central question for Cadillac is whether the software can execute that work accurately and repeatably, rather than simply make it easier to find the records involved.
For other manufacturers, Cadillac’s experience illustrates how data-quality problems can grow as production proceeds and the same information is reused across departments. Repairing the consequences may be far more labour-intensive than the original entry, especially where records are linked to complex assemblies. If Cadillac can establish a reliable automated correction process while improving the quality of new information entering the system, the benefit will come from reducing the need to reconstruct those histories in the first place.
