Recommendations for Franklin Motors and the Suisse Motors joint venture
Recommendations for Franklin Motors and the Suisse Motors joint venture
Ike Castillo · Solution Consultant · Atlassian Advisory Services · August 2026
Franklin adopts one operating model for how work is recorded, decided and found. The joint venture proves it first, because it's the only part of Franklin with no habits to unlearn. Suisse's Data Center is connected, not migrated, so the venture can start now. AI comes after the knowledge is curated — and only where a baseline and an owner exist.
one program team from each company, running the new rhythm from day 30.
VP Engineering.
residency, key ownership and the IP boundary, before anything is provisioned.
CTO.
written goal updates, and a recorded briefing where the status call used to be.
Both.
The rest of this is how I'd get there, in the order the decisions have to be made.
Confluence used heavily, no central source of truth, no standard practice. Teams decide well locally; alignment breaks across boundaries.
rapid reorganization and shifting priorities make it hard to advance shared objectives consistently.
engineering is fully on Atlassian; other departments run redundant tools.
Cloud migration complete, so no platform debt carried in. Production already modeled in Assets, so the venture extends an existing pattern. High satisfaction across engineering, manufacturing and selected business teams. Real appetite for what's new.
which sites, how many time zones.
entity versus subsidiary changes the workspace boundary.
it sets the sequence for retiring redundant tools.
Engineering isn't the problem. The gap is between engineering and everyone else.
share regular progress updates: monthly for goals, weekly for projects.
set working agreements.
lead every meeting with a page; adopt organized transparency.
define clear roles for AI teammates.
documenting clear, shared goals across teams and grounding work in trusted data accessible to people and agents.
designing clear paths for how work flows and progresses across teams of people and agents.
enabling people through continuous learning and experimentation that fosters innovative human-AI collaboration.
Atlassian State of Teams 2026. Only 14% of teams implement all three. Pages six, eight and nine are those three pillars.
Published practice from Atlassian's Teamwork Lab. I haven't invented a methodology — I've taken the frame and applied it.
The work isn't inventing the practices. It's choosing which ones, in what order, for a company with a two-hour window and two parent cultures.
Suisse works a Central European day. Where Franklin's engineering sites sit wasn't specified, so here is every offset that could apply.
raised after the window closes, answered tomorrow.
a full working week, if each exchange needs both parties awake at once.
no overlap at all. Every exchange is a day behind, by construction.
The overlap is the constraint. Everything that follows exists so a decision doesn't have to wait for it.
Plant and shift teams don't live in a wiki, and they don't have to. Assets prints QR codes and the Jira mobile app scans them — a technician scanning a rig at a station is the written record. What still needs someone at a desk is the narrative: the decision page, not the event.
| Cadence | Practice | Where |
|---|---|---|
| Daily | Work status updated in Jira by the person doing it. At the handover boundary, a two-minute Loom replaces the morning call. Loom transcribes one language per video; viewers can enable translated captions, but they aren't exportable — so the venture agrees one working language for the record. | Jira, Loom |
| Per decision | A DACI page within 24 hours naming driver, approver and contributors. Automation writes each one into a Confluence database, shaped to match the native decisions flow on the cloud roadmap. | Confluence |
| Per dependency | Every cross-company dependency is a written ask with a named owner, a committed date and a visible state. Negotiated in writing, not in the overlap window. | Jira |
| Per meeting | Page first, with a decision on it. Recurring meetings carry an expiry date and must earn renewal. Status meetings first. | Confluence |
| Weekly | Each team publishes a written update, read before leadership meets. | Atlassian Home |
| Monthly | Goal owners write their own update from the success measures. | Goals |
| Quarterly | Project Team Health Monitor, run by the team. They pick two reds, not eight. | Team Playbook |
No meeting without a page. · No decision without a named owner and a written record. · Status is read, not presented.
Every item here is a published Atlassian play. The venture adopts them as a set, in week one.
Guard Standard is included with Teamwork Collection Premium and Enterprise. Guard Premium is a separate subscription, billed per unique managed user.
Cloud organizations merge; they don't separate. If the venture is ever spun out, you want a boundary that moves in one piece.
residency is set per site; customer-managed keys are one policy per organization applied at site creation, and once enrolled the data can't leave that realm.
it delays the venture a quarter for no gain this year, and Suisse can't buy Data Center expansions after March 2028 with the platform going read-only in March 2029. A joint venture is a growth event; putting venture seats into a license pool that can't expand is buying a problem on a clock.
a one-way door.
It costs a second admin function. It buys two things that are painful to retrofit.
The venture hub in Confluence: goals, the decision register, the program calendar, how-we-work. One space per discipline underneath — vehicle program, powertrain, software, manufacturing, supply and quality — all from one template.
draft, in review, decided or superseded
the Jira it governs.
Atlassian calls this organized transparency: make early work visible, but flag what it is. Search ranks what's current and owned.
Atlassian's published Lucid Motors story runs this in production — a prototype drive unit becomes an object from its build ticket, is QR-scanned through testing and teardown, and links to the vehicle it sits in. When a car shipped to New Zealand for winter testing has a battery fail, the team in California can see which battery was in it and who built it.
A wiki nobody curates produces an assistant nobody trusts. The structure comes first.
| The move | What changes | Owner |
|---|---|---|
| Approvals move into the system | Gate reviews, engineering change requests and spend approvals become approvable only in Jira and JSM. Finance and procurement come in without a mandate. | VP Engineering |
| The renewal calendar becomes the retirement calendar | Nobody switches a tool off. You stop renewing it, one category at a time, against actual renewal dates. | Tooling owner, with the EM |
| The replacement lands 90 days early | The department has its project, templates and space before the tool it replaces comes up. | Department lead |
| Adoption runs on community learning | A champions network with a real time budget and a monthly demo day. Teams learn this from each other, not from training. | Department lead |
| The transfer is explicit | A practice that moves its measure gets written into Franklin's template library at day 90. The next department adopts it at its renewal. | VP Engineering |
40+ teams consolidated from six tools onto one platform.
30+ teams unified on shared workflows.
4,000+ meetings removed in five months; the monthly business review went from 125 people to 15.
All three are published Advisory Services engagements, self-reported. None is a forecast for Franklin.
A renewal date is a decision that's already scheduled. Declining one is easier than winning an argument.
Mercedes-Benz publish a 70% reduction in defect lead time on this pattern, baselined before go-live on a new vehicle program. The triage agent is a Rovo agent — configuration. The in-car voice intake that captures the defect at the moment it happens is a custom Forge build. The agent is the easy half.
Atlassian State of Teams 2026
12,035 knowledge workers and 173 Fortune 1000 executives, January–February 2026. Atlassian calls the gap the fragmentation tax and puts it at $161 billion a year across the Fortune 500. 89% say AI improved speed; only 48% say it improved collaboration.
| Agent | What it does | Owner · measure |
|---|---|---|
| Defect intake and duplicate triage | Reads the defect, checks duplicates, confirms logs are attached, routes it. | Venture quality lead · lead time from defect found to fixed |
| Recorded review to work item | Turns a Loom design review into Jira items with reproduction steps. | Program manager · actions lost between review and backlog |
| Change impact from Assets | Returns the parts, rigs and suppliers a proposed change touches. | Chief engineer's office · days from request to impact assessed |
No agent closes, reprioritizes or dispositions a safety-relevant defect. It proposes; a named engineer decides. Enforceable now rather than aspirational — agents are first-class participants in Jira and every agent interaction is auditable there.
Only 14% of teams build all three pillars. Pages six, eight and nine are those three pillars.
| Decision | What's true today | Who owns it |
|---|---|---|
| Where the processing happens | Content stays pinned to Germany. AI processing sits outside data residency, and by default Rovo sends context to third-party providers under zero-retention terms. Cloud Enterprise organizations can request Atlassian-hosted models only. | Suisse DPO, Franklin counsel |
| Permissions decide what an agent can reach | An agent acts as the person who invokes it and sees only what they can see. Data Center results appear only for people who have connected their own account. | Venture admin |
| Untrusted content is an instruction channel | Hidden text in a page, a ticket or an uploaded file can steer an assistant that reads it. Permissions don't contain that, because the agent already acts as someone entitled to the data. Keep connector scope small. | Security leads, both companies |
| The credit budget | Teamwork Collection Premium includes 700 Rovo credits per user per month, pooled and reset monthly; single-product Premium is 70. An agent request costs 10, Deep Research 100. Extra usage isn't billed today and needs explicit opt-in. Credits appear in the admin dashboard from this month. | CTO, with the account team |
| Guard's AI controls are recent | Prompt scanning in Rovo Chat and data security rules on connectors are new Guard Premium capabilities. Plan the pilot as if they aren't there: small connector scope, hand-classified content. | Security leads |
| European employee representation | Nothing reported per named individual. Team level only, agreed with the works councils before the first measure is taken. | HR and works councils |
| Developer AI is a separate conversation | Coding agents and code intelligence belong with the platform teams under a different set of controls. | Franklin platform teams |
None of these is a blocker. Each is a decision with a named owner, landing before the first agent switches on.
| Measure | Paired with | Why it's here |
|---|---|---|
| Cross-boundary readership | Page currency | Answers the silo finding directly. If Manufacturing never reads what Engineering writes, the silo is intact. |
| Cross-company decision latency | Decisions reversed within 30 days | Faster decisions that get unmade aren't faster. |
| Engineering change lead time | Defect escape rate at the gate | Speed at the gate can't come out of quality. |
| Time to find an answer | Correct on the first attempt | Sampled and timed by hand. The only measure that can't be gamed by the system it measures. |
| Status meeting hours per 100 people | Gate predictability | Fewer meetings only counts if the program still lands. |
Baselined in week one. Targets agreed by day 14. Team level only. Goal progress comes from the owner's monthly update, not automatically from linked Jira work.
Cross-boundary readership is a proxy, and proxies get gamed — you can drive it up by cross-posting noise. That's why page currency sits beside it, and why time-to-answer is sampled by hand. If readership moves and time-to-answer doesn't, the number is lying.
Underneath these sit the platform measures — standard configuration, duplicate spaces archived, admin hours reclaimed. Leading indicators for the five above.
A Teamwork Collection Accelerator for the venture, then a Momentum Plan for Franklin as practices transfer department by department. I'd start at Expanded and only call it Transformational once the venture has moved a measure.
Identity federation engineering, a Forge build like the one Mercedes did, or plant-scale rollout are a defined next piece of work with the right roles on it. Better scoped separately than absorbed.
Forrester, commissioned by Atlassian in January 2026, eight interviewed decision-makers aggregated into one composite: 260% ROI and $8.3M net present value over three years, with 60% of users non-technical. A commissioned study on a composite — discount accordingly. The figure that matters for a venture on a clock: customers cut project timelines by about a quarter.
Your administrators run this from day ninety. We're not building a dependency on us.
VP Engineering
One program team from each company, running the new rhythm from day 30. I'd pick the one whose gate everyone is already watching.
CTO
Residency, key ownership and the IP boundary, before anything is provisioned. None of it unwinds afterwards.
Both of you
Written goal updates, and a recorded briefing where the status call used to be.
Everything on the previous thirteen pages is mine to run. These three are yours.