Pilots

A first pilot built for real manufacturing motion

Lychee scopes early deployments around one line, one defect family, one KPI, and outputs that plant teams can judge quickly.

Scope A Pilot See product

Pilots in progress with Na-ion and Li-ion programs in the United Kingdom — 2026. Typical first pilot: first prediction in 5 days, pilot window 2–12 weeks. Early-pilot scope and methodology on the homepage pilot cards.

Reads from

AVEVA PI System · Siemens Opcenter · Aspen InfoPlus.21 · Ignition by Inductive Automation · custom historian and CSV exports

Data security

Process data stays in your environment. On-premises and private-cloud deployments supported.

Data requirement

We don't need run-to-failure data.

Real plant data is mostly cells that haven't failed yet — run-to-failure is a benchmark luxury, not a production reality. Lychee uses Weibull accelerated-failure-time regression with right-censoring: every unfailed cell trains the model. If your plant has cycling data, you have enough data.

Validated

8.9% vs 14.2%

At 46% cells censored · survival model vs. point-estimate

Validated on Severson with simulated censoring: at 46% of training cells unfailed, Lychee's survival model holds 8.9% median APE while a standard point-estimate model degrades to 14.2%.

Pilot scope

The first scope is legible in five minutes

A credible pilot starts narrow enough to evaluate quickly and concrete enough for plant teams to trust.

Pilot scope

One line

Start with a defined line or process area rather than a factory-wide rollout.

Pilot scope

One defect family

Focus on a defect pattern the plant already knows is expensive.

Pilot scope

One KPI

Tie the pilot to scrap, yield, throughput, or diagnosis speed.

Pilot scope

Weeks, not quarters

Scope the first operating window tightly enough to evaluate quickly.

Pilot scoping process

Four steps to first signal

Lychee compresses pilot entry to the shortest path — from historian credentials to actionable ranked-risk output, without a long data-engineering prerequisite.

Week 0

Data spec review

We scope data format, signal quality, and label availability before committing scope.

→

Week 0–1

Success criteria

Defect family, KPI, and win condition agreed with your QA and engineering lead.

→

Week 1 · 5 days

First signal

Historian credentials in, first ranked-risk predictions out.

→

Weeks 2–12

Pilot window

Real lots, real signals, engineering action.

Pilot structure

What a first engagement clarifies

Dimension	Plant reality	Pilot answer
Existing systems	MES, SPC, inspection, QMS, historian	Used as context and event history, not replaced
Minimum pilot inputs	Equipment or historian tags, lot context, QC results	Enough to test one defect family and one KPI
Primary outputs	Ranked risk, likely upstream drivers, investigation priorities	Designed for process and quality teams to act on
Governance model	Scoped around existing plant constraints	Aims to reduce deployment burden, not expand it
Data security	Process data stays in your environment	On-premises and private-cloud deployments supported

From benchmark to KPI

What benchmark metrics mean for your manufacturing KPIs.

Lychee translates research-grade benchmark performance into the metrics your CFO, operations director, and quality lead actually evaluate spend against — precision, lead time, and cost-benefit at your unit economics.

Research metric

Median absolute percent error

single-digit on 4 of 5 benchmarks

Manufacturing KPI

Precision · recall on early defect signal

Translates benchmark accuracy on cycle-life prediction into the precision and recall metrics your QA lead acts on for ranked-risk lot decisions.

Research metric

Cycle-life prediction

from cycles 1–100, validated across 4 labs

Manufacturing KPI

Lead time before downstream confirmation

Translates early-cycle prediction into weeks-earlier visibility on defect risk versus waiting for formation testing or end-of-line QC to confirm.

Research metric

Hybrid generalization

same model · 3 chemistries · no retuning

Manufacturing KPI

Cost-benefit at your unit economics

Translates cross-chemistry, cross-lab generalization into the unit cost of catching one bad lot — measured against your scrap, labor, and material write-off.

Deployment model