Across recent AI Phase projects in Munich, AI systems automated 70–80% of the targeted process steps, cut per-task time by up to roughly 97%, and can potentially halve the costs of the affected process. Every number below comes with its baseline and its hedge.
Last updated: July 2026
faster inspection: 10 minutes to around 20 seconds (B.B.W. Group)
of quality inspections automated with 3D computer vision
of IT support tickets automated for netcos GmbH
classification accuracy in the ticket automation
projected cost reduction potential, pharma process optimization
Where do these numbers come from?
Three reference projects with measured outcomes – plus the one figure people misquote most.
10 min → 20 s
Computer vision QA: from 10 minutes to 20 seconds
For the B.B.W. Group, a quality assurance partner of the BMW Group, AI Phase built a 3D computer vision system that inspects components automatically: 80% of inspections automated, inspection time down from 10 minutes to around 20 seconds – roughly 97% faster – and inspection costs that can potentially be halved (from two inspectors to one). The remaining 20% stays human: poor lighting, rare defect classes and judgment calls.
Measured: inspection time per part and share of automated inspections, against the manual pre-project baseline. The cost halving is a potential, not an audited saving.
70% automated
NLP ticket automation: 70% automated at over 90% accuracy
For netcos GmbH, AI Phase built an LLM-based ticket sorting and IT support assistant. It handles 70% of incoming tickets automatically at over 90% classification accuracy, with significantly faster response times; the remaining tickets are routed to humans by design.
Measured: share of tickets resolved without human handling and classification accuracy, against the manual pre-project baseline.
+20% quality
Pharma process optimization: +20% quality, ~30% cost potential (projected)
For a Swiss pharmaceutical company, AI Phase optimized a complex process column using Bayesian optimization and random forests on existing production data: projected +20% end-product quality, output consistency raised from 80% to 95%, and a potential cost reduction of around 30%. Honest caveat: this required months of historical process data before the optimization could start.
Measured: model-based projections on historical production data – reported as projected and potential, not as audited outcomes.
~1.4% running AI costs
Running costs at ~1.4% of manual handling – read this one carefully
The most striking and most misquotable figure on this page. In the netcos project, the running costs of the AI system amount to roughly 1.4% of the original manual processing costs – that compares the monthly operating cost of automated handling with the personnel cost of manual handling for the same volume. It is a calculated comparison, not an audited savings figure, and it excludes the one-off build cost.
Measured: monthly AI operating cost vs. manual processing cost for the automated share; stated as a calculated comparison.
Project reference · Validate it on your own data in 4–8 weeks
What do these numbers mean for your industry?
Four metric families transfer across industries: automation rate, per-task time, error and consistency rates, and running costs. The absolute values do not.
These are real, measured outcomes from individual AI Phase projects – not guarantees and not industry averages. Your baseline decides your ceiling: a process that is already 90% efficient will not halve its costs, and a document base nobody maintains will not produce an assistant with 90% accuracy.
- Metrics are defined with the client before the project and measured against the pre-project baseline.
- Figures marked with ~ are approximate; the pharma results are projections, and the B.B.W. cost halving is a potential.
- Results depend on data quality, process volume and a clear process owner.
- One-off build costs are not included in the running-cost comparisons.
For an honest estimate of your own process – and for the budget and timeline questions: AI consulting & opportunity assessment · what AI projects cost · how long AI projects take
Questions about the methodology or the projects behind these numbers: the team behind the measurements · ask us directly
Frequently asked about AI results
In AI Phase projects, 70–80% of targeted, high-volume process steps: 70% of support tickets at netcos GmbH and 80% of quality inspections at the B.B.W. Group. The remainder stays human by design – for edge cases and judgment calls.
Find out what is measurable in your process
Describe your process and get a scope sketch, an honest feasibility assessment and next steps within 24 hours – free and non-binding.