Improvement, concepts

Improvement holds Opportunities, Learning, and Verified Savings: find a loss, investigate it, take action, and check the result. Ask, Reports, Data, and Alerts are shared sidebar tools. Energy belongs to Performance and job records belong to Jobs.

Improvement and related tools

Peer comparison is part of Performance, alongside production, downtime, changeover, energy, and shift comparison.

Looking for live, drillable OEE-family KPIs on one node, with a data-completeness read? Use the KPIs & targets tab on that node’s own page, not a screen in Improve, reached by drilling into any node from Floor’s Facility Map (or from the “Want live drill-down?” link on the Reports page below). See How to check plant status and drill into a problem.

A saved investigation is a kind=incident Work Item, managed from Now. See Now, Work Items and How to manage Work Items.

See how-tos for step-by-step tasks on each of these.

Domain objects

Gateways, sources, and tags (as seen from Improve)

Telemetry in Spall flows: gateway (an edge device on the floor) → source (a connection the gateway polls, a PLC, a sensor bus) → tag (one named signal on that source, e.g. hydraulic_pressure). By the time telemetry reaches the Historian, it’s addressable as a device_id + metric_name pair. The Historian’s device and tag pickers are populated <select> dropdowns sourced live from GET /telemetry/devices and GET /telemetry/tags?device_id=..., you pick from what actually exists instead of typing an ID from memory. Full configuration of gateways/sources/tags lives in Setup. Improve only reads the resulting telemetry.

Incidents and the Historian

An incident is a kind=incident Work Item with one title, time window, asset, and free-text notes, captured from a flagged slice of the Historian’s telemetry trend, living in the same unified queue as andon calls and repairs (see Now, Work Items). The point is still that an investigation shouldn’t start from scratch each time, reopening an incident-kind Work Item jumps straight back to the Historian with the same device preselected and the same window in view, and an AI similar-incidents lookup surfaces past incidents that look like the current one, as a starting point rather than a verdict. See How to manage Work Items.

Jobs

A job is a planned/running/closed unit of production work against a product and a work center. Jobs (/work-orders) is the canonical, plant-wide record list, a record you review here on Improve, not a floor action. Sequencing and reassigning queued jobs before they run happens on Floor’s Dispatch Board. See How to manage jobs.

Energy and Verified Savings

Energy attributes metered kWh/$ down to the machine (run-state weighted) for branches with an energy sensor, and surfaces idle power waste. Verified Savings is the closing loop on Opportunities: link an acted-on insight or Work Item to the action taken (a job, a work request, a dated note), and the ledger tracks the dollar-quantified before/after, proof an action actually helped, beyond a claim. See How to monitor energy use and How to track verified savings.

ML insights and value opportunities

Two related but distinct automated-analysis surfaces live on the Opportunities screen:

• Value Discovery, a board of opportunities ranked by dollar impact, answering “where is capacity/money leaking right now.”
• ML insights, zero-config, per-asset anomaly/prediction insights generated automatically. Each open insight shows expected-vs-observed evidence and a confidence value alongside the verdict, a Models panel lists which signals each model actually uses (so the automation isn’t a black box), an insight can be dismissed, and “Explain further” calls into the AI layer for a plain-language elaboration.

One insight kind, “running slower than its best run,” compares a machine’s recent individual cycles against the fastest pace it has actually sustained (its best run of consecutive cycles, not one lucky cycle), and shows a small chart of the golden pace next to the recent cycles. It needs a real cycle boundary to work from, a part counter, a machine that reports each cycle as it finishes, or, failing those, a run/stop signal, and it waits for enough cycle history before comparing anything, so a newly connected machine shows “learning” instead of a guess.

Another insight kind, “recipe drifted from its best run,” compares the setpoints a machine holds right now against the setpoint values it held during its own best-performing recent stretch. It only compares against a setpoint recipe it actually knows, either one you captured yourself, or one it captured automatically once it knows what the machine is currently making. A recipe you capture yourself always takes priority over an automatic one. The card names which setpoints differ in plain language, shows the best-run value next to today’s value, and prices the gap the same way the cycle-pace insight above prices its own, only when the machine’s current performance is worse than it was during that best run, not just because a setpoint changed. When there’s no configured rate to turn the gap into a dollar figure, it still shows which setpoints differ, without inventing a number.

This is the same automated-analysis pipeline that powers Maintenance’s RUL panel and the proactive nudge banner on Facility Map, Value → Optimize → Explain. The banner names the machine the finding is about and links straight to it. An insight that has stayed open for a while shows how long it has been flagged, so the banner never reads like something that just happened. Once a finding is old enough that “just observed” language no longer means much, the banner stops showing it at all. It stays visible, with its age shown, on the Opportunities screen’s full ML insights list.

Context density

Above the Value Discovery board, Opportunities shows how much of the floor’s downtime carries human context: the share of stops with a reason code (a quick pick from the list, a stop auto-labeled Minor stop (auto) doesn’t count toward this share, nobody actually looked at it) and, separately, the share with a written note or comment (someone explained what happened), each shown against the prior comparable window so you can see the trend, beyond a single snapshot, e.g. “Reason-coded stops: 62% (up 21pp vs prior period).” A line and shift breakdown is available to open below it, so a supervisor can see where annotation is thin and follow up, without singling out an individual. This panel only appears once there’s downtime data in the window to measure, it never shows a made-up percentage on an empty window. The comparison is always against your own plant’s history, never against another customer’s numbers. Answers from Ask your factory take this into account too: if a window’s stops are mostly uncoded, the answer says so instead of presenting a reason-based ranking as if it covered everything.

Ask your factory

A plain-language Q&A box that answers questions using real, current data, maintenance/work items (including the notes people wrote on them and the discussion in their comment threads), the shift logbook, jobs/dispatch, downtime, quality/scrap, shift performance (including which shift-and-machine combination is losing the most and whether it’s getting worse), fleet/source health, alerts, and the $-ranked opportunities board, and cites the data sources that fed the answer. It does not free-associate from the question text alone. A human comment or note is credited distinctly in the sources (e.g. “3 operator notes on Press #2”) and attributed by role, never by the person’s name. If AI answers aren’t enabled for the tenant, the screen says so instead of inventing an answer. This is intentional, not a bug.

How the pieces relate

Reports and Historian both surface data that already exists elsewhere in the app (Reports exports/compares shift KPIs, Historian browses raw telemetry), while Ask-your-factory and Opportunities are the “investigate” pair: Ask-your-factory answers a question you already have in words, Opportunities surfaces questions you didn’t know to ask (automated). An investigation gets preserved as an incident-kind Work Item (managed on Now, not here) instead of a dedicated Improve bookmark. Energy is a node-scoped drill-down lens over the same telemetry/KPI data (a node’s own deeper KPI drill-down lives on its own page, on Floor, see Floor concepts), and Verified Savings closes the loop by proving an acted-on opportunity actually paid off. Jobs is the record-keeping counterpart to Floor’s Dispatch Board. All of them draw on the same underlying telemetry/KPI/alert data that Now and Floor screens present in real time, Improve is where that same data gets sliced, compared, exported, and explained.