Elli Peterson's TechCrunch

A vehicle launch can look healthy in a program review and still be deteriorating underneath. Tooling milestones may be green, prototype builds may be complete, and suppliers may have submitted PPAP packages, yet the plant can remain exposed to unstable cycle times, configuration errors, and defects that escape into finished vehicles. I learned this during a launch in which daily firefighting obscured a deeper problem: engineering, supplier quality, material planning, and final assembly were each seeing different versions of the same risk. The experience changed how I evaluate AI in Automotive Manufacturing. The technology matters, but its value depends on whether it connects decisions across the concept-to-start-of-production chain. The practical scope of AI in Automotive Manufacturing extends far beyond installing computer vision at an inspection station. It includes finding weak signals in engineering changes, predicting supplier readiness, stabilizing JIT and JIS material flows, de...

AI in Credit Collections looks different at every stage of the consumer lending lifecycle. A newly delinquent revolving card, an unsecured installment loan nearing charge-off, and an auto account eligible for repossession should not enter the same treatment strategy. Their payment structures, loss severity, customer options, compliance controls, and recovery paths differ. The real design challenge is to make intelligence sensitive to those differences while keeping servicing decisions explainable and operationally enforceable. For lenders modernizing this lifecycle, AI in Credit Collections can connect pre-delinquency assistance, early-stage outreach, loss mitigation, late-stage collections, and post-charge-off recovery. That connection is especially important when servicing, payment, bureau, collateral, and agency systems provide conflicting versions of an account. A model-driven treatment is only as dependable as the status, consent, dispute, and payment facts available at the momen...

My first serious encounter with Generative AI in MedTech did not begin with a dramatic diagnostic breakthrough. It began with an overworked design assurance lead, a conference room covered in traceability printouts, and a product team trying to reconcile user needs, risk controls, verification protocols, and submission commitments before a design review. The model produced a polished summary in minutes. It also quietly merged two distinct hazards and attributed a verification result to the wrong device configuration. That experience captured the promise and the danger of the technology: it can compress days of specialist effort, but fluency is not evidence. Since then, I have worked through use cases spanning research and product development, regulatory affairs, quality management systems, and post-market surveillance. The most useful way to understand Generative AI in MedTech is not as a universal automation layer but as a controlled capability embedded in defined workflows. The succ...

My first serious encounter with AI In Investment Management was not a dramatic trading-floor breakthrough. It was a portfolio-review meeting in which three teams produced three different answers to a seemingly simple question: why had a balanced mandate underperformed its benchmark by 42 basis points? Investment research blamed security selection, portfolio construction pointed to a duration mismatch, and performance attribution identified stale classifications in the underlying data. The models were not the immediate problem. Fragmented portfolio, benchmark, and reference data were. That meeting shaped how I now evaluate every investment AI initiative: start with the decision, trace its data lineage, and design controls around the practitioner who remains accountable. The most useful way to understand AI In Investment Management is as an operating capability rather than a collection of clever models. It can connect security screening, model portfolio construction, pre-trade complianc...

My clearest lessons about AI Use Cases in Construction did not come from polished demonstrations. They came from projects under pressure: a commercial tower whose concrete quantities kept moving, an infrastructure package waiting on late design information, and a closeout team reconstructing decisions from scattered RFIs, submittals, and daily reports. Those jobs taught me that artificial intelligence creates value only when it is attached to a defined construction decision, supplied with controlled project data, and placed inside the workflow of the people accountable for the outcome. A useful overview of AI Use Cases in Construction should therefore be read through a delivery lens. The important question is not whether a model can recognize drawings, summarize documents, or predict a delay in isolation. The question is whether that capability helps an estimator close a scope gap, enables a superintendent to protect the look-ahead schedule, gives project controls earlier evidence of ...

My first encounters with AI Use Cases in CPG were less glamorous than the conference presentations suggested. The work involved reconciling stubborn SKU forecasts, explaining promotion lift that disappeared after an event, and helping supply planners respond when packaging constraints invalidated an otherwise sound production schedule. Those experiences taught me that artificial intelligence creates value only when it is embedded in the decisions that category teams, demand planners, RGM leaders, brand managers, and customer supply teams make every week. A model can be statistically impressive and still fail if it arrives after the consensus forecast is locked or recommends an action that cannot be executed at the shelf. A useful way to evaluate AI Use Cases in CPG is to start with the operating rhythm rather than the algorithm. In a large branded manufacturer, decisions flow through stage-gate reviews, demand-plan reconciliation, S&OP, executive IBP, trade promotion planning, pro...

My clearest lessons about AI in Electronics Manufacturing did not come from a polished strategy workshop. They came during prototype builds, yield reviews, supplier escalations, and late-night containment calls when a line was producing failures faster than engineering could explain them. In high-tech electronics, an algorithm creates value only when it respects the physical realities of solder paste, component variability, product configuration, test coverage, and production timing. The most successful initiatives I have seen began with a stubborn manufacturing problem, not with a desire to deploy an impressive model. A useful overview of AI in Electronics Manufacturing shows how widely the technology can be applied, but the factory-floor experience is more nuanced. A model must fit into NPI gates, BOM release controls, SMT engineering routines, quality escalation paths, and serialized traceability systems. It also needs owners who understand why an AOI call differs from an ICT failu...