Reduce Scrap and Setup Costs in Manufacturing — Through People, Not Machines

Scrap and setup costs in a typical Werkzeugbau or Maschinenbau Mittelständler eat 5 to 12 percent of revenue every year — and most of that loss is invisible because nobody adds it up across shifts. The standard answer from the Industrie-4.0 catalogue is more sensors, more SMED workshops, more visual quality systems. Those work, eventually. But they take six-figure capex, an MES integration project, and 12 to 18 months to land — by which point the senior operator who knew the trick has retired and the apprentice has quit in week six.

This guide takes a different angle. Most scrap and most setup overhead in a 30-to-300-employee shop is not a machine problem. It is a people problem dressed up as a process problem: a Sonderbauteil trick that never made it from the senior operator to the next shift, a Werkzeugkorrekturwert that lives in one head, a setup that gets re-rüstet from scratch because the previous shift\u2019s notes evaporated. Fix that with workforce-AI first — and you cut 30 to 50 percent of the scrap and 10 to 15 percent of the setup time without buying a single new machine. BASS Tools did exactly this and reported −45 % scrap, −11 % setup and standstill time in six months.

Ask an Inhaber-CEO of a 100-person Werkzeugbau what scrap costs him per year, and the typical answer is somewhere between 200 and 500 thousand Euro. Then ask the same question to the controller — and the number triples. Then ask the foreman — and the number doubles again, because his number includes the re-setups that nobody else counts. Scrap and setup loss are mis-counted everywhere because the cost lives in five different places: material, machine hours, operator hours, delivery delays, customer credit notes. Nobody owns the total.

A realistic decomposition for a typical manufacturing or Werkzeugbau Mittelständler with 100 people and 30 million Euro revenue. Material loss: 1 to 3 percent of revenue, which lands directly in the materials line. Machine hours lost to setup: 8 to 15 percent of available capacity, valued at full burdened cost (machine + operator + overhead). Quality re-runs: 1 to 2 percent of throughput at full margin loss. Customer credits and rush-delivery surcharges: 0.5 to 1.5 percent of revenue. Add these up — and 5 to 12 percent of revenue is a conservative estimate. For a 30-million Euro shop, that is 1.5 to 3.5 million Euro per year of avoidable loss, mostly invisible.

The Industrie-4.0 narrative says: install vision systems, instrument every machine, run an SMED workshop. All three work. But for a 30-to-300-employee Werkzeugbauer, all three solve the wrong problem first.

Vision systems detect bad parts after they are made. They reduce ship-to-customer scrap, but they do not reduce the scrap itself. The reason a part is bad is almost always upstream: the wrong Werkzeugkorrekturwert, the material batch quirk that the previous operator knew about and the new one did not, the Sonderbauteil setup that was rüstet from scratch because the trick from last week never made it across the shift change.

SMED workshops cut setup time on the lines that get the workshop. Real benefit: 50 to 70 percent setup-time reduction on the targeted line, like the classic 90 → 31 minute case study. Problem: SMED is a one-time event. The gains erode over 6 to 12 months as the senior operators who internalised the workshop retire or move, and the next cohort never gets the workshop. SMED needs an ongoing knowledge-transfer mechanism to compound. That mechanism is workforce-AI.

The fastest, cheapest scrap fix in a Werkzeugbauer below 300 people is the inverse: capture what the senior operators already know, get it into the next shift, build a continuous-improvement (KVP) loop that compounds. That is what BASS Tools did. The −45 % scrap and −11 % setup-time numbers from BASS came not from new vision systems or new SMED workshops — they came from getting the existing knowledge into the next shift before it was needed, and from a +350 % increase in captured KVP measures month over month.

Three concrete loops do the work. Each runs on the channel that fits each role (WhatsApp, Signal, Teams, SMS, push, email or shopfloor tablet), each respects the existing role of the Senior-Vorarbeiter and Senior-Operator, each ships in 14 to 30 days. Together they account for the BASS-Tools delta.

Half of the scrap in a typical Werkzeugbauer is upstream of the next shift. A part comes out bad in the early shift not because of the early-shift operator, but because the night-shift operator left a tool in a marginal condition and the handover did not flag it. A Sonderbauteil gets re-rüstet from scratch because the trick from yesterday never made it across.

The fix is the 5-field digital shift handover: production status, quality and Sonderbauteil notes, tool/equipment status, people notes, open follow-ups for the next shift. AI summarises voice memos into the structured template. Senior-Vorarbeiter signs off before delivery. Next shift starts with a complete picture instead of half-remembered fragments. That alone took BASS Tools the lion\u2019s share of the −11 % setup-time reduction inside three months.

Behind every recurring scrap pattern in a Werkzeugbau sits one senior operator who knows the workaround and four newer ones who do not. Material-batch quirks, the specific Werkzeugkorrekturwert that works on the difficult Sonderbauteil, the CNC programme variant that only one machinist runs because nobody else figured it out — this is where the long-tail scrap lives. And this is exactly the knowledge that walks out the gate when the senior retires.

Four to six structured 90-minute capture sessions per Senior-Operator, AI-transcribed, organised by part family / material / Werkzeugkorrekturwert. The output is a queryable internal knowledge base — the next-shift Vorarbeiter searches by part number and gets the trick that the senior wrote up on a Tuesday afternoon. The Senior gets explicit recognition in the works newsletter — knowledge capture is a contribution to the legacy, not a replacement signal. The full method is in Wissenstransfer Renteneintritt.

The third loop is the slow burn. Every operator on every shift has 20 to 50 small process frictions per month — a tool that always sits in the wrong holder, a part that always needs an extra check, a CNC parameter that always needs the same correction. Most of these never get reported because the formal KVP process is too heavy: write a card, file it, wait two weeks, get told it cannot be done. The result: the +20 percent of process improvement that an engaged shopfloor would generate sits unused.

The AI-driven KVP loop fixes this with a 30-second daily pulse — one open-text question, one rating, one optional voice memo. The AI clusters thousands of small notes into a small number of actionable improvement candidates per week. Senior-Vorarbeiter and HR review the cluster, pick the top three, assign them, close them. The KVP rate at BASS Tools went from baseline to +350 % per month after this loop went live. Not because operators suddenly started caring — they always cared. Because the friction to report something dropped from 20 minutes to 30 seconds.

BASS GmbH in Niederstetten, 150 employees, Gewindetechnik since 1947 — the published six-month results: −45 % scrap, −11 % setup and standstill, +350 % KVP measures captured per month. The decomposition is instructive because it tells you what to expect on each loop.

Approximately 60 percent of the scrap reduction came from Loop 1 (shift handover summary). Reason: the largest scrap category in any three-shift Werkzeugbau is parts that come out bad on the next shift because of an unflagged tool condition or material quirk from the previous shift. Closing the handover gap closes the largest loss category first.

Approximately 25 percent of the scrap reduction came from Loop 2 (Senior-Operator knowledge capture). Once the rüst-knowledge for the difficult Sonderbauteilen lived in a queryable knowledge base instead of one head, the apprentices and Quereinsteiger stopped re-discovering the tricks the hard way.

Approximately 15 percent came from Loop 3 (KVP), but the curve was different — small monthly gains compounding rather than a step change. By month six the +350 % KVP capture rate was producing 2 to 3 confirmed process improvements per week, each individually small, cumulatively significant.