A US lab's AI now answers equipment breakdowns 130x faster
Genetic testing firm Natera built an AI helper that reads repair tickets and suggests the likely cause before an engineer opens them. Its biggest accuracy gain came from adding missing manuals, with no change to the AI model.

The short version
- Natera, a US genetic testing lab, built an AI helper that diagnoses broken lab equipment from repair tickets.
- The first response to a ticket now arrives in about 8 minutes. It used to take 17.4 hours.
- A senior technician's departure showed how much know-how lived in people's heads. The system writes that know-how down so it stays.

What Natera built
Natera runs DNA tests, and its labs depend on many precision instruments. When one breaks, patient samples wait. The lab gets more than 65 repair tickets a day.
Expertise was the weak point. Each instrument family had its own specialist. When a senior technician who knew one family well left, that knowledge left too. No system held it.
Natera's lab engineering team built a system it calls SWARM, with help from Amazon Web Services. It reads each new repair ticket before an engineer opens it. Then it writes a short diagnosis report. The report gives the most likely cause and ranked steps to try. It also says how confident it is.
Engineers can also talk to a second version of the helper. They can ask follow-up questions and upload photos of the faulty machine.
How it works, in plain terms
The helper learns from the lab's own documents: vendor manuals and technical guides. It also uses 33 written "skills". 24 of them cover specific families of instruments. The other 9 handle common steps, such as closing a ticket or calling in more help.
Natera kept the skills separate on purpose. When a diagnosis is wrong, the team can find the one skill that failed and fix it.
Patient data stays out. Tickets contain only details about equipment and workflow. Every use of the AI is logged, as a clinical lab requires.
The results
Natera measured the system over 90 days of real use. The average first response fell from 17.4 hours to about 8 minutes. That is roughly 130x faster.
Coverage is high as well. 96.2% of tickets get an automatic first diagnosis without an engineer starting it.
Accuracy is good but not perfect. When the helper rated itself highly confident, it was right 74% of the time across 154 checked cases. Engineers record what they actually found and fixed, and each answer is graded against that.
The most useful finding is about data. For the instrument family with the worst scores, adding the missing vendor manuals lifted accuracy from 71% to 75%. Nothing about the AI model changed. Natera's lead engineer, Ariel Jirau, put it this way: "Diagnostic accuracy is a data quality problem first."
What it means for companies in Indonesia
Many plants, hospitals and telcos have a version of this problem. A few senior technicians hold most of the know-how, and their notes live in their heads. When they retire or resign, repair times climb.
Natera's approach is a practical answer. The AI does the first reading of the ticket and points to likely causes. The engineer still makes the call. Over time, the lab's knowledge sits in documents and skills that anyone can check.
The data lesson matters most. Before buying a smarter AI model, check whether your manuals and repair records are complete. Natera's accuracy gain came from documents alone.
What to do next
- Find the repair queue where tickets wait longest for a first response, and measure that wait today.
- List the experts whose departure would hurt most, and start recording how they diagnose problems.
- Collect every vendor manual for your critical equipment in one place.
- Grade any AI helper against what engineers actually fixed, and track accuracy by equipment type.
- Keep a person responsible for every repair decision.
