What an AI agent does with your X-Matrix: four scenarios
Connected to Hoshin Kanri over MCP, an AI agent reads your X-Matrix, its indicators, its scores and how it is deployed to your sites, then works from them: it prepares a committee, checks that a group objective is taken up below, says which part of the strategy an event touches, explains a score. Hoshin computes; the agent reasons.
The four scenarios below are illustrative: the company, Mécanor, and the people are fictional. The traces are not. Every tool call ran on the public demo on 28 October 2026, every figure quoted comes from its output, and every request can be replayed on the demo.
1. Preparing the monthly leadership committee
Illustrative scenario. A lean consultant supports Mécanor, a fictional precision-machining SME. The day before the monthly leadership committee, she asks her agent to prepare the agenda from the matrix instead of chasing each owner for an update.
What matters here: the agent does not summarize a table. It crosses three sources nobody crosses by hand: the meeting cadence (who reviews what), the indicators (actual against target) and the chains of the matrix (what reaches nothing).
Request
Prepare the agenda of Mécanor’s monthly leadership committee: where do our annual objectives stand, which indicators are slipping, what do we need to decide?
MCP tool call
get_management_architecture()"Revue stratégique trimestrielle" quarterly, 180 min, tracks north + east └ "Comité de direction mensuel" monthly, 90 min, tracks east + south └ "Point hebdo production" weekly, 45 min, tracks south + westMCP tool call
get_matrix_data(matrix_id: "3716b0dc…")elements.west (currentValue / targetValue) "Customer non-conformity rate (%)" 1.4 / 0.8 result "Internal defect PPM" 720 / 450 performance "Average lead time (days)" 23 / 18 performance "On-time delivery OTD (%)" 91 / 96 result "Changeover time (min)" 41 / 25 performance "Monthly invoiced revenue (€M)" 1.29 / 1.38 result "Absenteeism (%)" 5.2 / 4 result "Operators certified on ≥3 workstations" 15 / 24 performance elements.south (externalRef) "In-line quality control with machine vision — milling shop" linear.app/mecanor/project/vision-fraisage "EN 9100 certification (aerospace market)" linear.app/mecanor/project/en9100 "Mock audit in September." correlations (excerpt) Customer non-conformity ↔ machine vision: strong Customer non-conformity ↔ EN 9100: medium SMED program ↔ Changeover time: strong Scheduling overhaul ↔ Average lead time: strong Scheduling overhaul ↔ On-time delivery: strong completeness.details.chains (excerpt) "An attractive plant: retain key skills…" path.east: ["Absenteeism below 4%"] path.south: [] complete: false completeness.details.orphans.elements "Employer brand: technical school partnership + onboarding track" quadrant: south
Step not replayed
In a real deployment the agent would now read the progress of the two linked Linear projects through Linear’s own MCP server. The demo is connected to no Linear workspace, so this step did not run.
Agent answer
Monthly leadership committee, 90 minutes. It reviews the annual objectives and the initiatives; here is what needs the time. 1. Customer non-conformity (20 min). 1.4% against a 0.8% target, internal PPM at 720 against 450. The two most linked initiatives each have a Linear project: machine vision in the milling shop (strong link) and EN 9100 (medium link, mock audit in September). Review progress on those two projects. 2. Lead time and delivery (20 min). Lead time 23 days against 18, on-time delivery 91% against 96%, changeover 41 minutes against 25. The SMED program and the scheduling overhaul carry these three indicators: those are the owners to question. 3. Revenue (10 min). €1.29M invoiced per month against €1.38M. 4. Decision needed (20 min). Absenteeism at 5.2% against 4%, and no initiative carries this objective: the vision “An attractive plant” stops halfway down. One initiative exists, the employer brand, linked to nothing. Attach it, or launch another one. 5. Cross-training (10 min). 15 operators certified on three or more workstations, against 24. That leaves 10 minutes of slack.
2. Checking that the sites took up the group objectives
Illustrative scenario. Mécanor runs two plants, a machining site in Cholet and an assembly site in Vierzon, each with its own X-Matrix attached under the group one. The group operations director wants to know, before the quarterly review, whether the sites actually deployed what the group decided.
Hoshin computes the cascade: which group objective or initiative each site element deploys, what nobody took up, which site objective deploys nothing. The agent then does what the computation cannot: it says which gaps are acceptable.
Request
Have our two sites taken up the group’s objectives? What is carried by nobody?
MCP tool call
get_cascade(matrix_id: "3716b0dc…")matrices "Mécanor — X-Matrix 2026 (English)" depth 0 "Mécanor — Cholet plant, machining (English)" depth 1 "Mécanor — Vierzon plant, assembly (English)" depth 1 coverage: 60 edges[0].deployed (← takenUpBy) "Customer non-conformity: from 1.8% to 0.8%" ← Cholet "Milling shop at aerospace-grade quality" "Cut average production lead time by 20%" ← Vierzon "Ship every assembled order on time" "In-line quality control with machine vision…" ← Cholet "Milling PPM below 400" "SMED program on the 3 bottleneck machining centers" ← Cholet "Changeover under 25 min…" "Operator cross-training plan…" ← Vierzon "10 operators certified on 3 or more workstations" "Shop-floor scheduling overhaul…" ← Vierzon "Assembly OTD at 96%" edges[0].notDeployed "€16.5M revenue (+12%) with gross margin held" "Absenteeism below 4%" "EN 9100 certification (aerospace market)" "Employer brand: technical school partnership + onboarding track" unanchored Cholet: "Overtime hours cut by 30%"
Agent answer
Six of the group’s ten annual objectives and initiatives are taken up by a site. Cholet takes up customer non-conformity, machine vision and the SMED program. Vierzon takes up production lead time, scheduling and cross-training. Carried by nobody: revenue, the employer brand, EN 9100 certification and absenteeism. The first two can stay at group level. The last two cannot: a certification is passed on a site, and absenteeism is played out on the shop floor. Someone has to decide which site carries them. In Cholet, “Overtime hours cut by 30%” deploys nothing from the group. It may be a sound local objective. Worth asking whether it draws on the same teams as the SMED program; then either attach it to lead time, or own it as local.
3. An event hits the strategy
Illustrative scenario, fictional event. Mécanor’s main aerospace customer announces that from July 2027 it will only buy from EN 9100-certified suppliers. The quality manager pastes the announcement into her assistant and asks what it changes.
This is the matrix used as a reading grid: the agent does not judge the news in general, it reads it against the elements and links the company already committed to, and says what is exposed.
Request
Our main aerospace customer just announced it will only buy from EN 9100-certified suppliers from July 2027. What does this touch in our strategy, and what should change?
MCP tool call
get_matrix_data(matrix_id: "3716b0dc…")elements.south (excerpt) "EN 9100 certification (aerospace market)" status: active externalRef: linear.app/mecanor/project/en9100 description: "Mock audit in September." correlations of this initiative ↔ "€16.5M revenue (+12%) with gross margin held": strong ↔ "Customer non-conformity: from 1.8% to 0.8%": medium ↔ "Customer non-conformity rate (%)": medium ↔ "Internal defect PPM": medium ↔ "On-time delivery OTD (%)": weakMCP tool call
get_management_architecture()(excerpt) "Comité de direction mensuel" monthly, 90 min, tracks east + south "Point hebdo production" weekly, 45 min, tracks south + west
Agent answer
What it touches. EN 9100 was a lever to win the aerospace market; it becomes a condition for keeping this customer. In the matrix it strongly supports the 2026 revenue objective (€16.5M): if certification slips past July 2027, part of the existing revenue is at stake, not just growth. What is missing. Three indicators are linked to EN 9100: customer non-conformity, internal PPM, on-time delivery. They measure quality, not progress toward the audit. I suggest a dedicated performance indicator, for instance the findings still open after the September mock audit. What changes in governance. Today the topic lives in the Linear project. The monthly leadership committee reviews annual objectives and initiatives: EN 9100 should come up there every month, with July 2027 as the milestone. I changed nothing: the demo token is read-only. With a write token, I would prepare these changes in the matrix and apply them once you agree.
4. Explaining a score
Illustrative scenario. Mécanor’s CEO opens the matrix and sees a completeness score of 74. He asks the question everyone asks: why 74 and not more, and where to start.
The score is computed by Hoshin, the same way every time, from a public specification. The agent turns its details into an order of work, and notices what the sub-scores do not say: two of the gaps are the same gap.
Request
Why is our matrix at 74 and not higher? Where do we start?
MCP tool call
get_matrix_completeness(matrix_id: "3716b0dc…")score: 74 levelLabel: "Good" details (sub-scores): coverage 95 · density 44 · orphans 80 · chains 67 · indicatorBalance 90 density.NW: possible 24, actual 0 chains.details (excerpt) "An attractive plant: …" path.east: ["Absenteeism below 4%"] path.south: [] complete: false suggestions density.sparse Create more correlations between Vision (North) and KPIs (West) orphans.one Link "Employer brand: technical school partnership + onboarding track" (Initiatives (South)) to an adjacent element chains.incomplete Vision "An attractive plant: retain key skills and attract the next generation" has no complete chain (missing: Initiatives (South)) balance.missingBoth "Employer brand: …" is linked to no KPI — add both result and performance
Agent answer
74 out of 100, “Good”. Coverage (95) and indicator balance (90) are fine; three things pull the score down, and two of them are the same hole. 1. The vision “An attractive plant” reaches no initiative (chains: 67). The objective “Absenteeism below 4%” has nothing delivering it. 2. The employer brand initiative is linked to nothing, neither objective nor indicator (orphans: 80). These two answer each other: the initiative missing from “An attractive plant” already exists, it is just not wired. Linking it to absenteeism and giving it an indicator closes both at once. 3. No correlation at all between the vision and the indicators (density: 44, the north-west corner is empty). This is the longer job: choosing which indicators tell you, each month, whether the vision is moving. Where to start: points 1 and 2, a ten-minute decision in committee.
Frequently asked questions
Are the people and the company real?
No. Mécanor is a fictional SME built for the public demo, and the roles described are scenarios. What is real: the tool calls, their outputs, and the fact that you can replay them.
Why doesn’t my agent answer word for word the same?
The figures come from Hoshin and will be identical. The wording comes from the agent and varies from one model to another. Hoshin computes, the agent reasons.
Which agent do I need?
Any MCP client: Claude (claude.ai, Desktop, Code), Cursor, or another assistant that supports MCP servers. On claude.ai and Cursor the connection is a URL and an OAuth sign-in.
Can the agent change the matrix?
Yes, with a write token or an OAuth connection with write access: create elements, wire them, update a status. Per-matrix edit rights apply to agents too, and a token never grants more than its user’s role. The demo is read-only.