Logo of AccediaContact us
Logo of AccediaOpen menu icon

AI in the Automotive Industry and What Has to Be in Place First

    Blog Post

    |

  • By

    Dimitar Dimitrov

Published

Nov 16, 2026

AI-powered robotic arms assembling an electric vehicle battery pack.
  • An automotive AI use case depends on the systems it reads and writes, so a chief information officer (CIO) should list them before funding it.
  • Integration is the most-cited AI challenge in automotive manufacturing. It was named by 36 percent of 316 senior leaders in the Automotive Manufacturing Solutions (AMS) survey (2026).
  • Security checks, plant change control, and a named owner for the integration come before the first model.


AI in the Automotive Industry Today and the Systems It Depends On


AI in the automotive industry means software that learns from plant, vehicle, or supply chain data to support a decision. Its applications include driver assistance, design and engineering, plant quality and maintenance, and supply chain planning. Each one reads from systems that already exist.


The AMS automotive AI survey (2026) found that nearly half of North American automotive manufacturers describe their digitalization as still at the awareness stage. That matters when the CEO or CFO has asked a Tier 2 supplier's IT team for an AI plan. It has to start from the systems the company already runs.


Before investing in AI, an automotive company needs a list of the systems each use case connects to and an integration owner. An original equipment manufacturer (OEM) customer asking for more traceability data can be the trigger, because the answer draws on several sources that must agree. Judging the benefits of AI in automotive comes second, since a use case with no interface to its records cannot pay off. Listing those systems per candidate shows the integration it needs.


How to List the Systems an Automotive AI Use Case Has to Connect To


To list the systems an automotive AI use case has to connect to, write down what it reads from and writes back to. Then note whether an interface exists today.


The systems are the enterprise resource planning (ERP) system and the manufacturing execution system (MES). They also include the plant network, operational technology (OT), and engineering and supplier data. Run the list plant by plant, because plants differ.


Consider a manufacturer that bought its plants one at a time and built a corporate data lake without plant-floor connectivity. Its reports never matched the plant floor.

Use caseERPMESPlant network and OTEngineering and supplier dataWhere it can break
Predictive maintenance
Reads asset records, writes maintenance orders
Reads downtime and shift records
Reads machine signals, writes nothing to equipment
Reads part and tooling specifications
Machine signals from mixed-age equipment
Visual quality inspection
Reads part and lot records
Reads and writes inspection results
Reads camera and sensor data
Reads drawings and tolerances
ERP and MES records that disagree on the lot
Supply chain and demand planning
Reads orders and stock, writes planned orders
Reads production status
Not needed
Reads supplier and part data
Supplier data with no interface to read from
OEM traceability reporting
Reads receiving and shipment records
Reads lot, operation, and shift records
Reads machine data per lot
Reads supplier lot data
The first step in the chain with no interface

ERP and MES


An AI use case reads orders, lots, and routings from the ERP system and the MES. If those two disagree, the model inherits the mismatch. Likewise, where shift patterns and asset events sit in separate systems, an automotive predictive maintenance model may not tell a planned stop from a fault.


Plant Network and OT


Reading from plant equipment is harder than reading from the ERP system, because machines differ in age, make, and protocol. A model may read machine signals, but it should never write back without a separate change-control review. Commands to older equipment meet obstacles of their own, as digital twin development for legacy plants shows.


Accedia used Azure IoT Edge to consolidate data from robotic machinery of differing generations and makers at a Swedish bearing and rotating-equipment manufacturer. The work covered global plants. Connect mixed equipment first, before any model depends on it.


Engineering and Supplier Data


Check whether part specifications and supplier data sit in tools or portals nothing else reads. For each use case, name the system that holds them and its interface, if one exists. Data quality and who owns each data environment need a separate review, because automotive data management affects how well an AI investment is protected. An OEM traceability request needs all four system groups, so it tests the list directly.


What an OEM Traceability Request Shows About Your Integration Gaps


An OEM traceability request shows which of your systems cannot yet answer a question about a part, a lot, or a shift. Answering it means tracing the item from receiving through each operation to shipment. Each step may sit in a different system, and the request fails at the first step with no interface.


Suppose a Tier 2 supplier ran its ERP system on Epicor or Plex, and its OEM customer asked for more traceability data. In that scenario, the managed service provider (MSP) ran the infrastructure but could not build integrations. A freelancer's scripts pulled the data once, then broke at the next ERP upgrade.


The AMS survey (2026) of 316 senior leaders ranked these AI implementation challenges highest.


  • Integration with existing automation and legacy IT, at 36 percent.
  • Cybersecurity, at 31 percent.
  • Data governance and data quality, at 31 percent.
  • Skilled labor shortage, tied with budget approval at 30 percent.


A script that fails at an upgrade is an integration cost that was never funded, so fund the interface and its upkeep before the model. Where the ERP system is older, Accedia's legacy software modernization work can connect new software to it in phases without taking it offline. Each funded interface then needs access and change-control rules before a model uses it.


Security and Change-Control Checks That Come Before the First Model


Four checks come before a model reads from or writes to plant and ERP systems, and they belong in the funding plan. An OEM that wants more data from production widens what the AI can touch. The MSP that runs the infrastructure then needs written rules to follow, because it cannot guess which access is allowed.


  • Access per interface. Set read access for each interface separately, limited to the tables and signals the use case needs.
  • Logging at the interface. Log every read and write at the interface, so each one leaves an audit record.
  • Which systems stay read-only. List the equipment and ERP functions that stay read-only, and keep their interfaces read-only.
  • Change control at ERP or MES upgrades. Name the plant owner who signs off on interface changes when the ERP system or MES is upgraded, and before any write goes live.


Once a model can act on its own, the same rules set when an agent may act are the core decision in agentic AI in manufacturing. Each of these checks also needs someone who keeps it running after go-live.

AI Defect Detector for Manufacturers

Upload a photo of a failed machine part, and the Accedia AI Defect Detector will identify the component, assess the defect and its severity, recommend next steps, and provide a confidence score.

Who Runs the Integration After Go-Live: Your IT Team, Your MSP, or a Partner


After go-live, the integration needs a named owner, and the choice is between your IT team and an outside provider. An MES integrator that only maintains the production system cannot own an AI project for a metric such as yield.


IMD's 2025 AI maturity research treats maturity as organization-wide change. In our view, that includes the team that runs the system after launch, so name it before go-live.


  • Your IT team. Check whether it has people who know the plant's ERP, MES, and OT interfaces. Skilled labor shortage was cited by 30 percent of leaders in the AMS survey (2026), so confirm the people exist before assuming this option.
  • Your MSP. Check whether the contract covers integration work or infrastructure only. An MSP that cannot build integrations cannot repair one either.
  • A partner. Check whether it stays on after launch and has worked with your ERP and MES.


Whichever option fits, agree in writing who fixes an interface when the ERP system or MES is upgraded. A partner suits the case where neither internal IT nor the MSP can take on that work. Accedia's manufacturing team works on software, plant systems, and enterprise data for manufacturers, including the integration between them.


Five Questions to Answer Before You Fund an Automotive AI Use Case


Five questions decide whether an automotive AI use case is ready for funding. They give a CIO a list to take to the CEO or CFO. The list fits a manufacturer that bought its plants one at a time and a supplier whose OEM customer asked for more data.


  1. What does the use case read from and write to?
  2. Does each of those systems have an interface or export today?
  3. Who owns each interface when the ERP system or MES is upgraded?
  4. What security and change-control review does the connection need?
  5. Who runs the integration and the model after go-live?


Their answers also feed any wider digital transformation in the automotive industry program. A question without an answer becomes the first funded task.


Start With the Integration You Already Need


An automotive AI use case succeeds or stalls on what it has to connect to, so fund that integration before the use case. A request from an OEM customer for more data shows which of your systems cannot answer it. For a plant where several of them fail that test, AI is not the next step yet.


Pick one use case and start from the OEM data request already on your desk. Write down what it reads from and writes to, then add each interface's owner and security review. If you want help with that list, Accedia's automotive AI consultants can assess your use case and data readiness and give first guidance on integration.


FAQ

  • How is AI used in the automotive industry today?

    AI is used in the vehicle, in design and engineering, in plant operations, and in supply chain planning. Inside the car, it supports driver assistance and in-car assistants. In plants, it is applied to quality inspection and predictive maintenance on production equipment. Each of these uses depends on data from systems that already exist.

  • What does an automotive company need in place before investing in AI?

  • What is the biggest challenge in implementing AI in automotive manufacturing?

  • Who should run an AI system in an automotive plant after launch?

  • Which are the top automotive software development companies in Europe?

  • Author

    Dimitar Dimitrov

    Dimitar is a technology executive at Accedia, specializing in software engineering and IT professional services. He combines corporate strategy, business development, and people management to lead with flexibility and focus on customer success. His leadership has driven triple-digit revenue growth, backed by close attention to detail and a real enthusiasm for technology.

    Related Insights from Accedia