AI Integration with Existing ERP System of a Small Manufacturer

PräzisionsTech AG is a Lucerne-based company with 28 employees that manufactures precision parts for the medical technology and watch industries. The company processes over 200 different materials and produces 45 different product lines. The ERP system was implemented 12 years ago and is deeply integrated into all business processes. A replacement would have taken 18+ months and cost CHF 150,000+. At the same time, the system lacked modern features like automatic reordering, demand forecasting, and quality pattern recognition.

Production manager Peter Hofmann knew the system's weaknesses exactly: raw material reordering was based on fixed minimum stock levels defined years ago that were not adjusted to current demand. Result: on one hand, materials piling up in the warehouse for months (tied-up capital), on the other hand, sudden shortages leading to production stops, three times in the last year for 1-3 days each. Each production stop cost the company CHF 5,000-8,000 in downtime costs. Simultaneously, there was a growing realization that quality problems showed patterns: certain material batches led to higher scrap rates, but without data analysis, these patterns remained invisible.

Three core problems: First, static reorder points led to simultaneous overstock and shortages. Capital tied up in unnecessary inventory was estimated at CHF 45,000. Second, production planning was manual and reactive: orders were processed in the order received without optimizing machine utilization or setup times. Third, quality data (scrap rates, dimensional deviations, rework) was collected but only viewed as monthly totals: connections between material batches, machine parameters, and quality issues remained undetected.

Instead of replacing the proven ERP system, we developed three AI modules connected via the existing database interface. The reorder module analyzes historical consumption data, current order status, and supplier lead times to calculate dynamic reorder points and optimal order quantities. The production planning module optimizes order sequencing considering setup times, machine utilization, and delivery deadlines. The quality module detects correlations between material batches, machine parameters, and quality results, providing early warnings when a batch shows irregularities. All three modules read data from the existing ERP and write recommendations back. The ERP itself remains unchanged.

The database connection to the 12-year-old ERP system was the biggest technical challenge. The system used a proprietary database format for which we had to develop a special connector. The second challenge was data hygiene: 12 years of data contained inconsistencies that had to be cleaned first. We developed an automatic data cleansing script that also continuously ensures data quality. The production manager was involved in development from the beginning and could contribute his domain knowledge, for example, which materials cannot be stored together or which machine setup times are realistic.

After 6 months in production, the results are convincing: not a single production stop from material shortage since go-live. The previous year had three. Overstock decreased by 30%, freeing up CHF 13,500 in tied-up capital. Monthly savings total CHF 4,100 through optimized order quantities, reduced setup times, and early quality detection. The quality module already identified two material batches early as irregular. In the past, these problems would only have been noticed at the customer. Production planning reduced average setup times by 20% through more intelligent order sequencing.

The central insight: you don't have to buy a new system to benefit from AI. The existing data from the old ERP system was gold: 12 years of historical data provided excellent training material for the AI models. The investment of CHF 22,000 for the AI integration was a fraction of the CHF 150,000+ for an ERP change, with faster ROI and no migration risk.