Artificial intelligence
A process engineer in everyone's pocket.
Everyone on the floor gets expert help on demand, and your experienced people get their time back for what adds value: the line, the floor, the product and your customers.
AI that works inside your rules. Decision models make the calls, with written rules you can read and check by hand. AI helps around them. You choose the sites and groups each AI feature runs for.
AI governance
Fine-grained control over every AI feature.
Six AI features, and you decide where each one runs. Switch it on or off per site, keep it to the groups you choose, and read every change and every model call in the audit. A person still creates every action and picks every owner. Which models we use, and where they run, is shared on request.
-
Vera briefing
Home
Puts the next best step in front of a person, one card at a time, from Score Card KPIs, goals, open actions and the nightly Top 3 losses. The loss maths has no AI in it; the model only explains why and words the card.
Per-site switchVera Users
-
Refocus Problem Statement
Root Cause Analysis
Drafts one statement from the team's 6W2H answers, marked "Draft · edit before saving". The team edits it and saves.
Per-site switchLost elimination team
-
Voice dictation
Throughout the app
Speech to text wherever you type; the speaker edits before saving. A speech model turns the audio into text.
Per-site switchDictation Users
-
AI credit metering
Settings
Every model call is metered against your site's own credit balance, with usage per site. Record-only and shadow modes run before anything counts.
Site credit balanceUsage per site
-
Access control
Every Vera request
Vera uses Plant Toolbox the way that person would, never beyond their own access: checked when it offers, when it calls and in the normal action chain. Any write is a confirm card a person approves.
Vera UsersA person approves writes
-
The Plant Toolbox harness
Settings › Vera
Switch each AI feature on or off per site, 23 switches in all. The harness picks the model, tests the connection, and audits every switch change and every model call.
Settings accessEvery change audited
Switches · this site 6 of 23
- Vera briefingHome · Vera Users
- Refocus Problem StatementRoot Cause Analysis
- Voice dictationThroughout the app · Dictation Users
- AI credit meteringEvery model call Record onlyShadowCounting
- Vera stays inside each person's accessOffer, call and action chain Locked on
- The harness picks the modelNo model picker Locked on
Groups
- Vera Users
- 14 people
- Dictation Users
- 9 people
- Changes switches
- Settings access
Credits · this month
Used 351Balance 1,862
Model lane
Plant ToolboxThe model
The harness picks the model; there is no model picker. Connection test ok
Audit · latest
- 08:14model_callVera briefingJ. Lee · ok · 3 credits
- 08:12switch_changeRefocus Problem StatementM. Chen · Off → On
- 08:10connection_testThe modelM. Chen · ok
Decision models
Every model by name: what it reads, how it decides, what you see.
Each model is a written rule with stated inputs and thresholds, so any result can be explained in a sentence and checked by hand.
-
01
KPI status
six states
On target · Off target · No goal set · No data · Not available · Future period
How it decidesChecked in order: Future period, Not available, No data, No goal set, then On target or Off target. Compared on the displayed decimals. No yellow band.
ReadsThe KPI value, its goal from Settings › Goals, the direction (higher or lower is better) and the operating calendar
-
02
Decision Tree
Line Action Plan
Shift A, Line 4: Green 85
How it decidesScore = the weights answered Yes. A No on a Force Red on No question sets Red at 0. End on False stops the walk. At or above Upper is Green, at or above Lower is Yellow, else Red. Each site tests its questions in Simulation before they go live.
ReadsYes or No answers to ordered Trigger questions, each with a weight; a Lower and an Upper Bound; the line's shift
-
03
Barometer roll-up
Plant and Site boards
Every line's color on one plant board
How it decidesOne tile per line. Dark when the line has no shift or is scheduled down, grey until the questions are answered, split at shift handover to show both teams.
ReadsEvery line's barometer, its shift schedule and any planned outage
-
04
Vera briefing
four computed facts
On Home, most severe first, each fact with a drill link
How it decidesFour facts, found the same way every time: the biggest miss, the biggest mover, overdue actions, off-target results with no action. The facts are computed by rule, so the same input gives the same facts; where AI is switched on, a model only words the card.
ReadsThe Score Card KPIs, goals, coverage and open actions of your view
-
05
Chase Logic
line engine
Each downtime event opens with the machine that caused it
How it decidesWhen the line stops, attributes the stop to the machine whose fault caused it, not to the machines starved or blocked by it.
ReadsEach machine's state (running, down, starved, blocked), read about every 10 s from a collector at each machine
-
06
Downtime → RCA thresholds
off until a threshold is set
Fri Oct 9: Filler jam, 47 min → RCA #219
How it decidesUnplanned and longer than the RCA threshold: a Start RCA prompt. Longer than the smaller comment threshold: a comment prompt.
ReadsA classified stop (planned or unplanned, and its duration) and two thresholds you set for the company account
-
07
Coverage
a Vera fact and a Needs an action card
Needs an action, week 41: 2 of 5 have an action
How it decidesA result is covered when an action is open, linked to that KPI, and saved on that area's own board for that period.
ReadsOff-target results for the last complete period, and the open actions linked to a KPI
-
08
Effectiveness verdict
Did the actions work?
#4031: 86% of the gap closed
How it decidesBefore: 28 days before the action was raised. After: up to 28 days after it closed. A move needs at least 2%. Effective when at least 25% of the gap is closed.
ReadsDone actions linked to a KPI; the KPI before and after; the goal and the direction
-
09
Owner ranking
Most relevant
#4182: Maya Chen, Luis Ortega, Jordan Blake
How it decidesLists you first, then past owners of actions on the same KPI and area, then owners of open actions on that board, newest first. A person still chooses.
ReadsThe KPI, the area and the board the action will live on
-
10
Product Rate What-If
try a rate before you change it
The result at the rate you try, beside today's
How it decidesRecalculates the result with the rate you try, beside the result at today's rate. Nothing goes on the plan.
ReadsA product's rate on a line, and a rate you want to try
Escalation is not a model: a person escalates an action one tier in one click, and de-escalates it no lower than the board it started on.
See the tier ladderA process engineer in everyone's pocket.
In a demo we walk one shift: the Decision Tree before it starts, a KPI miss judged against its goal, and the action a person creates from it. We'll show you the harness for your sites, too.