[{"data":1,"prerenderedAt":249},["ShallowReactive",2],{"/en/workflows/branch-signal-evidence-coach":3},{"id":4,"slug":5,"locale":6,"translationGroupId":7,"localeSwitchApproved":8,"title":9,"description":10,"documentationMarkdown":11,"workflowJson":12,"category":227,"tags":228,"integrations":232,"difficulty":236,"author":237,"verified":33,"featured":33,"date":238,"modified":238,"icon":7,"imageSrc":7,"path":239,"alternates":240,"seo":241},"1e459ac8-25b9-4f43-90d7-88507f2190db","branch-signal-evidence-coach","en",null,true,"Branch Signal Evidence Coach","A menu-driven coaching flow that helps teams sanity-check branch numbers, spot dirty signals early, and decide when automation is safe vs when judgment must stay in the loop.","## How it works\nThis workflow turns a “we have numbers, so we must be right” moment into a calmer, more decision-ready check. It gives leaders and operators a quick set of reality tests for branch metrics, conversations, and attribution—before a confident meeting runs off a cliff.\n\nThe flow starts by attempting a Knowledge Base answer (great when the question is familiar), then guides the user through a practical menu: trust tests for branch numbers, dirty-signal detection, automation vs judgment, working with messy evidence, common branch comparisons that go wrong, and how to build a signal culture that produces decisions—not just slides.\n\n## Key features\n- Knowledge Base-first responses, with routing to a structured coaching menu when needed\n- Button-based guidance for common decision-shaped questions (trust, hygiene, attribution, culture)\n- “Switch-like” routing so one tap reliably lands on one outcome\n- Loops back to the main menu after each answer to support multi-question sessions\n- Optional handoff to a human support queue when the user needs help beyond self-serve\n\n## Step-by-step\n1. **Trigger:** The workflow starts when a new user message enters the flow.\n2. **Knowledge Base attempt:** The flow tries to answer via the **Knowledge Base Policy** (fast wins for repeatable questions).\n3. **Menu of decision checks:** If a KB answer isn’t appropriate, the user sees a button menu to choose what they need:\n   - Trust branch numbers\n   - Spot dirty signals\n   - Automation vs judgment\n   - Use messy evidence without “sanitizing” it\n   - Avoid attribution & comparison traps\n   - Build a signal culture\n   - Talk to a human\n4. **Routing:** The user’s button tap is evaluated in sequence (like a switch statement). The first match routes to the correct guidance.\n5. **Guidance delivered:** A targeted, operator-friendly message is sent for the chosen topic.\n6. **Loop for more help:** After the guidance, the flow returns to the main menu so the user can run another check.\n7. **Handoff (optional):** If the user selects **Talk to a human** (or no button matches), the workflow hands off with a short context-setting message.\n\n## Setup requirements\n- **Calypso Knowledge Base:** Recommended (the flow is designed to attempt KB answers first). No additional credentials are required inside this workflow.\n- **Routing/handoff:** If you want the human handoff path to work, ensure your Calypso workspace has a department/queue that can receive routed conversations (the template references **Analytics Support**).\n- **Channels:** Works anywhere Calypso interactive buttons are supported (e.g., Web Chat, WhatsApp).",{"id":13,"teamId":14,"name":9,"version":15,"workflowVersion":16,"nodes":17,"connections":188,"routingEnabled":8,"active":33},"wf_branch_signal_evidence_coach_v1","calypso-public-library","1.0.0",1,[18,34,40,52,86,96,104,110,116,122,128,134,140,146,152,158,164,170,181],{"id":19,"name":20,"type":21,"typeVersion":16,"position":22,"parameters":25,"category":32,"deletable":33,"connectable":33},"node_flow_configs","Flow settings","flow-configs",[23,24],120,80,{"name":9,"description":26,"tags":27,"triggerType":31},"KB-first coaching menu to sanity-check branch signals, spot dirty data, and apply automation guardrails.",[28,29,30],"signal-design","decision-systems","branch-metrics","input","policy",false,{"id":35,"name":36,"type":31,"typeVersion":16,"position":37,"parameters":39,"category":31,"deletable":33,"connectable":8},"node_input","Inbound message",[23,38],220,{},{"id":41,"name":42,"type":43,"typeVersion":16,"position":44,"parameters":46,"category":51,"deletable":8,"connectable":8},"node_kb_policy","Knowledge Base: Decision signals","knowledge-base-policy",[45,38],360,{"enabled":8,"fallbackToRouting":8,"sticky":8,"stickyMode":47,"activationOpener":48,"personalization":50},"ai_sticky_release",{"enabled":8,"instruction":49},"Answer plainly and operationally. Prefer decision checks, thresholds, and what usually breaks first. Avoid academic language.",{"useContactName":8},"response",{"id":53,"name":54,"type":55,"typeVersion":16,"position":56,"parameters":58,"category":51,"deletable":8,"connectable":8},"node_menu","Decision checks menu","interactive-message",[57,38],620,{"messageType":59,"headerText":60,"bodyText":61,"footerText":62,"sectionTitle":63,"buttons":64,"ctaDisplayText":63,"ctaUrl":63},"button","Decision checks for branch signals","Pick what you’re deciding. I’ll give you quick tests that prevent confident wrong calls.","If it looks perfect, check the plumbing.","",[65,68,71,74,77,80,83],{"id":66,"title":67},"trust_branch_numbers","Trustworthy numbers",{"id":69,"title":70},"spot_dirty_signal","Dirty signal check",{"id":72,"title":73},"automation_vs_judgment","Auto vs human",{"id":75,"title":76},"messy_evidence","Messy evidence",{"id":78,"title":79},"attribution_misreads","Attribution traps",{"id":81,"title":82},"signal_culture","Signal culture",{"id":84,"title":85},"talk_to_human","Human help",{"id":87,"name":88,"type":89,"typeVersion":16,"position":90,"parameters":93,"category":95,"deletable":8,"connectable":8},"node_if_trust","If: trust branch numbers","if",[91,92],860,160,{"buttonId":66,"operator":94},"equals","routing",{"id":97,"name":98,"type":99,"typeVersion":16,"position":100,"parameters":102,"category":51,"deletable":8,"connectable":8},"node_text_trust","Guidance: trust tests","text-message",[101,92],1120,{"text":103},"Here’s what usually deserves trust (and what’s polished noise):\n\n1) Trust metrics that have a clear *denominator you can’t “forget”*. Example: “approved loans / eligible applicants” beats “approved loans”.\n2) Trust numbers that *reconcile* (daily totals match weekly totals; branch sums match region totals). If they don’t add up, you’re measuring the reporting pipeline, not reality.\n3) Trust signals with *stable definitions*. If the definition changes mid-quarter, the trend is a costume.\n\nPolished noise tends to be:\n- Anything that can be “improved” by changing what gets counted (reclassification, exclusions, channel shifts)\n- Single-point snapshots (end-of-month photos hide the mess)\n- Rankings without error bars (small branches bounce around; it’s math, not magic)\n\nDecision move: before you act, ask: “What would have to be true for this number to be misleading?” Then go look for that specifically.",{"id":105,"name":106,"type":89,"typeVersion":16,"position":107,"parameters":109,"category":95,"deletable":8,"connectable":8},"node_if_dirty","If: spot dirty signal",[91,108],240,{"buttonId":69,"operator":94},{"id":111,"name":112,"type":99,"typeVersion":16,"position":113,"parameters":114,"category":51,"deletable":8,"connectable":8},"node_text_dirty","Guidance: dirty signal checks",[101,108],{"text":115},"Dirty data rarely shows up as “missing.” It shows up as *too smooth, too consistent, or too convenient*.\n\nFast ways to catch it before the meeting goes off the rails:\n- **Spikes at boundaries:** big jumps on month-end, quarter-end, campaign start/stop = process artifacts until proven otherwise.\n- **Perfect conversion chains:** if every step improves at once, you probably changed tracking or counting.\n- **One-branch miracles:** ask, “Did anything operational change?” If the answer is “no,” assume instrumentation or definition drift.\n- **Conversation mismatch:** if sentiment/complaints say one thing and the KPI says the opposite, the KPI is often lagging, filtered, or gamed.\n\nDecision move: pick one metric and run a “plumbing check”:\n1) What’s the source event?\n2) Where can it drop, duplicate, or be re-labeled?\n3) Who benefits if it looks better?\n\nIf you can’t explain the pipeline in two minutes, you’re not looking at a metric—you’re looking at a rumor with a chart.",{"id":117,"name":118,"type":89,"typeVersion":16,"position":119,"parameters":121,"category":95,"deletable":8,"connectable":8},"node_if_auto","If: automation vs judgment",[91,120],320,{"buttonId":72,"operator":94},{"id":123,"name":124,"type":99,"typeVersion":16,"position":125,"parameters":126,"category":51,"deletable":8,"connectable":8},"node_text_auto","Guidance: automation guardrails",[101,120],{"text":127},"When to trust automation (and when to keep humans in the loop):\n\n✅ Automate decisions when:\n- The cost of a wrong decision is low or reversible\n- The signal is frequent, consistent, and well-instrumented\n- You can measure outcomes quickly (tight feedback loop)\n- The decision rule is simple enough to explain to a new hire\n\n🛑 Require human judgment when:\n- Edge cases matter (exceptions are where losses hide)\n- Incentives invite gaming (targets tied to bonuses, rankings, approvals)\n- The label is subjective (quality, intent, fraud risk)\n- The environment is changing (new product, new policy, new channel)\n\nGuardrails that work in practice:\n- **Confidence thresholds**: only automate when confidence is high; route the gray zone to humans.\n- **Kill switches**: a single metric tripwire pauses automation.\n- **Audit slices**: review a random sample weekly, not just the failures.\n\nRule of thumb: if you can’t name the top 3 ways the automation could be wrong, you’re not ready to let it drive.",{"id":129,"name":130,"type":89,"typeVersion":16,"position":131,"parameters":133,"category":95,"deletable":8,"connectable":8},"node_if_messy","If: messy evidence",[91,132],400,{"buttonId":75,"operator":94},{"id":135,"name":136,"type":99,"typeVersion":16,"position":137,"parameters":138,"category":51,"deletable":8,"connectable":8},"node_text_messy","Guidance: messy evidence → insight",[101,132],{"text":139},"Turning messy evidence into usable insight (without cleaning away the truth):\n\n1) **Don’t over-clean.** If you remove “outliers” because they’re inconvenient, you’re often deleting the story.\n2) **Segment before you average.** Branches behave differently by volume, tenure, and local constraints. Averages hide that.\n3) **Keep raw examples nearby.** Pair each dashboard view with 3–5 real conversations or cases. Numbers without texture breed fantasy.\n4) **Track “unknown/other” honestly.** A shrinking “unknown” is often better progress than a prettier top-line.\n\nPractical output format (decision-ready):\n- What we think is happening\n- The *strongest* evidence for it\n- The most likely ways we’re wrong\n- What we’ll check next week to confirm\n\nIf your cleaning process makes the data look more confident than your operations team feels, you’re probably cleaning the signal, not the noise.",{"id":141,"name":142,"type":89,"typeVersion":16,"position":143,"parameters":145,"category":95,"deletable":8,"connectable":8},"node_if_attr","If: attribution misreads",[91,144],480,{"buttonId":78,"operator":94},{"id":147,"name":148,"type":99,"typeVersion":16,"position":149,"parameters":150,"category":51,"deletable":8,"connectable":8},"node_text_attr","Guidance: attribution & comparisons",[101,144],{"text":151},"What teams repeatedly misread when comparing branches, conversations, and attribution:\n\n- **Volume illusion:** high-volume branches look “better” because randomness averages out. Small branches swing wildly—don’t overreact to ranks.\n- **Mix shift:** branch A didn’t improve; it got easier customers (seasonality, channel, eligibility, local events).\n- **Conversation ≠ conversion:** more chats can mean better service *or* more confusion. Tie conversation signals to outcomes cautiously.\n- **Last-touch bias:** attribution loves the final step and ignores the earlier work (education, problem discovery, objection handling).\n\nDecision move: when you compare branches, insist on:\n1) Same definitions\n2) Same time window\n3) A mix-adjusted view (at least split by customer type or product)\n4) A “what changed operationally?” note\n\nIf the story is “Branch X is just better,” you’re probably missing a constraint, a mix shift, or a tracking gap.",{"id":153,"name":154,"type":89,"typeVersion":16,"position":155,"parameters":157,"category":95,"deletable":8,"connectable":8},"node_if_culture","If: signal culture",[91,156],560,{"buttonId":81,"operator":94},{"id":159,"name":160,"type":99,"typeVersion":16,"position":161,"parameters":162,"category":51,"deletable":8,"connectable":8},"node_text_culture","Guidance: build a signal culture",[101,156],{"text":163},"Build a signal culture that helps decisions happen (not just slides):\n\n- **Name your “decision metrics.”** A small set that actually changes actions. Everything else is supporting evidence.\n- **Reward flagging weirdness.** The best analysts sound annoying at first—and save you later.\n- **Require a reversal plan.** For any big call: “What would make us undo this within 2 weeks?”\n- **Keep a metric changelog.** When definitions change, write it down. Otherwise you’ll argue with ghosts.\n- **Use pre-mortems.** Before the meeting: “How could this dashboard be misleading?” Then assign someone to try to break it.\n\nWit-sized truth: if a team can’t decide, it’s usually not because they lack data. It’s because they don’t trust it—and nobody’s allowed to say that out loud. Fix that first.",{"id":165,"name":166,"type":89,"typeVersion":16,"position":167,"parameters":169,"category":95,"deletable":8,"connectable":8},"node_if_human","If: talk to a human",[91,168],640,{"buttonId":84,"operator":94},{"id":171,"name":172,"type":173,"typeVersion":16,"position":174,"parameters":176,"category":180,"deletable":8,"connectable":8},"node_fallback_handoff","Handoff: Analytics Support","fallback",[101,175],680,{"handoffMessage":177,"departmentId":178,"departmentName":179},"Got it—routing you to Analytics Support. If you can, share: (1) which branch(es), (2) the metric(s), (3) the time window, and (4) what decision you’re trying to make.","analytics-support","Analytics Support","terminal",{"id":182,"name":183,"type":99,"typeVersion":16,"position":184,"parameters":186,"category":51,"deletable":8,"connectable":8},"node_text_no_match","If no match: safe next step",[101,185],760,{"text":187},"I may have missed your selection. Please send a new message and try the menu again—pick the closest option and we’ll narrow it down.",[189,193,195,197,200,203,205,207,209,211,213,215,217,219,221,223,225],{"id":190,"source":35,"target":41,"sourceHandle":191,"targetHandle":191,"type":192},"conn_input_to_kb","main","smoothstep",{"id":194,"source":41,"target":53,"sourceHandle":191,"targetHandle":191,"type":192},"conn_kb_to_menu",{"id":196,"source":53,"target":87,"sourceHandle":191,"targetHandle":191,"type":192},"conn_menu_to_if_trust",{"id":198,"source":87,"target":97,"sourceHandle":199,"targetHandle":191,"type":192},"conn_if_trust_true_to_text","true",{"id":201,"source":87,"target":105,"sourceHandle":202,"targetHandle":191,"type":192},"conn_if_trust_false_to_if_dirty","false",{"id":204,"source":105,"target":111,"sourceHandle":199,"targetHandle":191,"type":192},"conn_if_dirty_true_to_text",{"id":206,"source":105,"target":117,"sourceHandle":202,"targetHandle":191,"type":192},"conn_if_dirty_false_to_if_auto",{"id":208,"source":117,"target":123,"sourceHandle":199,"targetHandle":191,"type":192},"conn_if_auto_true_to_text",{"id":210,"source":117,"target":129,"sourceHandle":202,"targetHandle":191,"type":192},"conn_if_auto_false_to_if_messy",{"id":212,"source":129,"target":135,"sourceHandle":199,"targetHandle":191,"type":192},"conn_if_messy_true_to_text",{"id":214,"source":129,"target":141,"sourceHandle":202,"targetHandle":191,"type":192},"conn_if_messy_false_to_if_attr",{"id":216,"source":141,"target":147,"sourceHandle":199,"targetHandle":191,"type":192},"conn_if_attr_true_to_text",{"id":218,"source":141,"target":153,"sourceHandle":202,"targetHandle":191,"type":192},"conn_if_attr_false_to_if_culture",{"id":220,"source":153,"target":159,"sourceHandle":199,"targetHandle":191,"type":192},"conn_if_culture_true_to_text",{"id":222,"source":153,"target":165,"sourceHandle":202,"targetHandle":191,"type":192},"conn_if_culture_false_to_if_human",{"id":224,"source":165,"target":171,"sourceHandle":199,"targetHandle":191,"type":192},"conn_if_human_true_to_fallback",{"id":226,"source":165,"target":182,"sourceHandle":202,"targetHandle":191,"type":192},"conn_if_human_false_to_no_match","automation",[28,29,30,229,230,231],"data-quality","automation-guardrails","attribution",[233,234,235],"Calypso Knowledge Base","Web Chat","WhatsApp","intermediate","Calypso","2026-05-21T11:05:14.446Z","/en/workflows/branch-signal-evidence-coach",{"en":239},{"title":9,"description":242,"ogDescription":243,"twitterDescription":244,"canonicalPath":239,"robots":245,"schemaType":246,"alternates":247},"Help teams trust the right branch numbers, spot dirty signals early, and decide when automation is safe—using a guided menu and KB first answers.","A practical coaching workflow for branch metrics: detect polished noise, keep messy evidence honest, and know when automation needs human judgment.","Turn branch signals into decision ready insight: trust tests, dirty signal checks, automation guardrails, and attribution traps—guided by a simple menu.","index,follow","HowTo",[248],{"hreflang":6,"href":239},1780761214035]