Flow designer — swimlanes
Most of Consulting OS is invisible to the customer — because that's how it should be. The flow designer is different. It's visual, interactive, and it's the first thing every new tenant runs into. This piece is about the v4.2 release that just shipped, and why we spent 6 weeks on an editor.
What is a flow?
A flow is an engagement template: steps, roles, decision points, skippable sections, scoring rules. A new tenant doesn't build a flow from scratch — they pick one of three that cover 80% of their use cases, then customise.
What ships in v4.2
Visual graph editor
The previous version made you build flows with form fields. Building a 12-step flow that way took 40-50 minutes, and the connections ("if step 3 succeeds, jump to step 7") stayed hidden. v4.2 now gives a full graph editor: nodes, edges, drag-and-drop, visual display. That same 12-step flow now takes 8-10 minutes.
Role-based lanes
Every node has a role (consultant / customer-stakeholder / customer-tech / third-party). The canvas is sliced horizontally into lanes, and nodes drop into the lane matching their role. At a glance you can see who's blocking, who holds the ball.
Decision branches
Between nodes, alongside the normal sequential edges, you can place conditional branches. Example: "More than 5 critical items in audit findings? → Yes → go to Deep audit branch. No → go to Standard implementation branch." The condition is a JSON rule or a simple expression.
Conditional skips
Some steps can be skipped under certain conditions. Example: if the customer already signed an NDA on a previous engagement, the NDA-sign step auto-skips (but the state transition gets recorded).
Scoring
Steps can carry a scoring rule. For example at the end of the Discovery step we calculate a 0-100 "engagement readiness" score from the questionnaire. Below 40, the flow halts and the consultant gets a halt notification.
Undo/redo with 50-state history
The previous version kept 5 states, which after a single accidental node delete wasn't enough to recover. v4.2 stores 50 states per session, persisting through the workflow design session.
The 3 starter templates 80% use
1. ERP integration (8 steps)
Kick-off → Read-only access → Mapping audit → ADRs → Connector implementation → Sandbox testing → UAT → Production cutover. The consultant lane has 6 nodes, the customer-tech lane 2, the third-party (ERP vendor) lane 0 (because usually they only need to issue a read-only credential).
2. WMS redesign (12 steps)
Kick-off → Walk-through (physical site visit) → Current process audit → Process redesign workshop → ADRs → New flow prototype → Customer feedback → Implementation → Pilot (1 warehouse) → Measurement → Roll-out → Hand-over. Here 4 lanes are in use because third-party (WMS vendor, integrator partner) also enters.
3. AI assistant build (6 steps)
Use case definition → Data audit → Prototype (within 1 week) → User testing → Refinement → Production. Short flow, strict scoring: after prototyping, user testing requires minimum 70% accuracy and minimum 4.0/5 user satisfaction.
What v4.2 does NOT give
- No version control for flows. A flow edit overwrites the previous version. If you need git-style flow versioning, that's planned for v4.4.
- No multi-user real-time editing. One person edits at a time. If there's parallel editing, last-write wins (with a warning).
- No AI flow suggestion. "Generate me a flow for introducing a new CRM" — that doesn't exist. We consider flow design a human-led task.
Why did we spend 6 weeks on this?
Two reasons:
- New tenant onboarding stalled on flow building. The new tenant launched setup, hit flow building, and either struggled with it for 30 minutes or gave up. v4.1 had 23% 14-day activation. v4.2 has 41% (2 weeks of data). Too early to declare clear victory, but the numbers look good.
- Existing tenants asked for it. 18 of 47 tenants flagged the form-based flow editor as "frustrating" in feedback. It topped the feature request list for 3 months.
The takeaway
The visual editor isn't icing on the cake. By the new-tenant activation numbers, it's on the critical path. Sometimes you need to spend 6 weeks on an editor because without it, the product won't spread. v4.2 promises a lot — now the data will tell us if it delivered.