Shepherd Labs Health Intelligence
A customer health monitoring agent built on Claude
- → A custom, interactive and interconnected set of health intelligence dashboards that aggregates customer signals from across your tool stack into a single portfolio view
- → An onboard AI agent that reasons over your customer data and surfaces insights about renewal risk, expansion opportunities, and team performance
- → A natural language chat interface that lets you query your portfolio the way you would ask a question of a senior analyst
Real World Application
Built to connect to any tool that exposes an API, Shepherd Labs replaces the fragmented dashboards of individual platforms like Gainsight, Salesforce, and Monday.com with a unified view customized to how your team actually works. CSMs can drill into individual accounts, surface flags, and track trends in signal history. CS leaders get a portfolio-level summary with agent-generated insights ready each morning. When a CSM or leader has a question about dashboard data, the chat interface is one click away to go deeper.
Synthetic Data
Shepherd Labs is built on a synthetic dataset: a fictional B2B SaaS company with 75 accounts, a CS team of 7, six health signals calibrated to balance predictive value against data cost, five customer flags, and six super user classifications that drive proactivity. The portfolio is designed to replicate the retention and expansion dynamics a real CS team navigates every week.
Summary: Portfolio KPIs and agent insights
Summary (continued): CSM risk profile, super users, and renewals
Agent Chat: Multi-turn reasoning example
All Accounts: Full portfolio table with filtering
Account Preview: Individual account signal history
CSM View: Individual CSM deep dive
Shepherd Labs includes the ability to copy or download each agent chat session. The three exports below give a feel for the agent's reasoning and the kinds of questions the system is built to answer.
CSM Workload and Coaching
**You:** Are any of my CSMs showing signs of being over-extended? **Agent:** Looking at workload distribution, two CSMs stand out as concentration risks right now. --- ## Jacob R. — 13 Accounts · $6.62M ARR Jacob carries the highest-ARR book on the team, and the risk concentration is significant: - **4 at-risk or critical accounts** requiring active intervention: Quantum Ventures (At Risk, silent churner), Westbrook Media (At Risk, silent churner), and two further At Risk accounts — all in Jacob's book - **3 rescued accounts** (IronBridge Data, Keystone Infrastructure, Zenith DevOps-adjacent) in active recovery — these require sustained attention to prevent re-regression - **2 expansion candidates** (Ember Cloud, Flux Technologies) that need proactive engagement to capture upside - At $6.62M ARR across 13 accounts, a single churn in his at-risk cluster represents $420K–$640K in ARR exposure The concern isn't account count — it's that Jacob is simultaneously managing rescue recovery, silent churner escalations, and expansion motions. That's three distinct engagement modes running in parallel with no slack. --- ## Tom C. — 9 Accounts · $3.21M ARR Tom's book is smaller but his risk density is the highest on the team: - **2 Critical accounts**: NovaStar Fintech ($630K, silent churner + support spiral) and Vantage AI ($570K, support spiral) — both declining and requiring intensive intervention - **3 At Risk accounts** alongside those Criticals - **2 recent unexpected churns** already on his record — a third would be a serious pattern signal - NovaStar Fintech alone is $630K and already past renewal (renewal_days = –48), meaning that account may already be lost and still consuming CSM bandwidth Tom's situation is different from Jacob's — it's not a capacity problem, it's a performance and support problem. The question isn't just whether he's overextended, but whether he has the toolkit to manage accounts in crisis. The two unexpected churns plus two Criticals suggest he may need coaching support or temporary account redistribution, not just a lighter load. --- ## Everyone Else — Within Normal Range - **Priya N.** carries Xenon Infrastructure (Critical) and PeakEdge Security (At Risk), but her 13-account book is otherwise healthy and she has the tenure and process discipline to manage it - **Marcus T.** has GalaxyCorp (At Risk) and two honeymoon accounts to watch, but his 14-account book is manageable at his experience level - **Miranda L., Samara K., Diana F.** are all within healthy ranges — Miranda and Samara's primary watch items are single accounts each --- **Bottom line:** Jacob is overextended by risk complexity — too many simultaneous engagement modes across a high-ARR book. Tom is a different problem: his book is smaller but his Critical accounts are severe, his track record includes two recent churns, and NovaStar may already be gone. Both warrant leadership attention, but for different reasons. --- **You:** Which of Jacob's accounts would be the best transition candidates, which CSMs would be the best transition candidates and why? Keep in mind, we don't want to negatively impact the customer experience. **Agent:** This requires balancing Jacob's overextension against two constraints: don't disrupt accounts that are stable and relationship-dependent, and don't overload receiving CSMs. The goal is to relieve pressure on Jacob's crisis accounts without moving accounts that would suffer from a transition. --- ## Accounts to Keep with Jacob These should not move. Disrupting them creates more risk than it resolves. - **Quantum Ventures & Westbrook Media** — Both are At Risk silent churners in active decline. Transition at this moment would almost certainly accelerate churn. Jacob owns the relationship and needs to see these through. - **IronBridge Data & Keystone Infrastructure** — Both rescued accounts in V-shape recovery. Stability of the CSM relationship is part of what makes rescue stick. Don't touch these. - **Ember Cloud & Flux Technologies** — Healthy expansion candidates. These are Jacob's upside. A new CSM inherits a relationship they didn't build right before an expansion conversation. --- ## Best Transition Candidates These are stable, low-complexity accounts where a clean handoff carries minimal disruption risk. | Account | Segment | ARR | Tier | Super User | Renewal | |---|---|---|---|---|---| | Axiom Financial Group | Enterprise | $580K | Healthy | Seasoned Excellent | 626 days | | Cortex Engineering | Enterprise | $480K | Healthy | Seasoned Excellent | 36 days | | Summit Commerce | Enterprise | $380K | Neutral | Seasoned Solid | 50 days | | Zenith DevOps | Enterprise | $530K | Positive | Seasoned Solid | 386 days | | Margin IO | Mid-Market | $185K | Healthy | Seasoned Excellent | 330 days | | Zencore Platform | Mid-Market | $300K | Healthy | Seasoned Solid | 753 days | **Rationale across these accounts:** All are stable or improving with no active flags, no rescue dependency, and Seasoned super users who won't need heavy onboarding support from a new CSM. Cortex Engineering and Summit Commerce have near-term renewals (36 and 50 days respectively) — those need immediate handoff execution if they move, but both are healthy enough that a clean transition shouldn't threaten the renewal. Orbit Global ($550K, Healthy, expansion-flagged) is borderline — it's stable and improving, but it carries an expansion flag, so whoever receives it needs to be ready to run that motion soon. --- ## Best Receiving CSMs **Priya N. — Primary candidate for enterprise accounts** Priya is the strongest enterprise operator on the team with 4 years tenure and a track record as the top renewals performer. Her current book has two difficult accounts (Xenon Infrastructure, PeakEdge Security), but the rest of her portfolio is clean. She has the process discipline to absorb 1–2 enterprise accounts without disruption. Axiom Financial Group or Zenith DevOps would be natural fits given her enterprise depth — and if Orbit Global moves, she could handle the expansion motion. **Diana F. — Strong candidate for mid-market and select enterprise** Diana is the rising star with a predominantly healthy/expansion book and demonstrated ability to manage that motion well. At 10 mid-market and 2 enterprise accounts, she has room. Zencore Platform or Margin IO would land cleanly in her book. She could also absorb Cortex Engineering — it's a straightforward healthy enterprise account with a 36-day renewal, which plays to her process strengths. **Miranda L. — Limited capacity for one clean mid-market account** Miranda is nearly ready for enterprise promotion and her book is predominantly healthy. She could absorb Zencore Platform or Margin IO without strain. Keep her at one addition — she still has Parcel Works as a watch item and adding too much before her promotion could create its own risk. --- ## Who to Avoid as Receiving CSMs - **Tom C.** — Already managing two Criticals and three At Risk accounts. Adding anything to his book is counterproductive. - **Marcus T.** — Has GalaxyCorp in active At Risk decline plus two honeymoon accounts requiring attention. Not the right moment to add Jacob's accounts. - **Samara K.** — Only 6 months tenure. Dragonfly Logistics is already a watch item she may not have fully recognized. Not ready for Jacob's enterprise accounts. --- ## Bottom Line Move 2–3 stable, Seasoned super user accounts — prioritizing Axiom Financial Group, Zenith DevOps, and Zencore Platform — to Priya and Diana. Leave everything with active flags, rescue trajectories, or expansion timing in Jacob's hands. A targeted transfer of clean accounts is the right pressure release valve; a broad redistribution creates more disruption than it solves. --- **You:** What type of coaching intervention do you suggest for Tom? **Agent:** Tom's situation has two distinct problems that need different interventions — a performance pattern problem and an active account crisis. Conflating them with a single coaching approach won't work. --- ## The Pattern Problem First Two unexpected churns plus two Critical accounts plus three At Risk accounts is not bad luck — it's a signal that something systematic is breaking down in how Tom manages accounts under stress. Before prescribing specific tactics, leadership needs to understand *where* the breakdown is occurring. A structured case review of the two churned accounts should happen immediately, focused on: - **Where were the early warning signals?** Were the silent churner or support spiral patterns present weeks before churn — and did Tom see them? - **What interventions were attempted, and when?** Late intervention on declining accounts suggests either signal recognition failure or escalation hesitancy - **What was the customer communication cadence?** Support spiral accounts (NovaStar Fintech, Vantage AI, Sandstorm Labs) require proactive outreach ahead of ticket volume — was that happening? The answers will tell you whether Tom has a *detection* problem, an *intervention* problem, or an *escalation* problem. The coaching looks different for each. --- ## Signal Recognition — If That's the Gap Tom's Critical and At Risk accounts share a consistent pattern: Value PerformanceScore, DAU/MAU, Time-to-Value, and super user competency all declining in parallel over a sustained period before reaching crisis. These aren't sudden drops — they're slow bleeds that were visible in the data weeks ago. If Tom isn't catching these early, the intervention is structured health score review cadence — weekly account reviews with explicit thresholds that trigger action, not just observation. Pairing him with Priya N. for joint reviews on his at-risk accounts would give him a live model of what "early pattern recognition" looks like from a CSM with strong process discipline. --- ## Intervention Execution — If He's Seeing It But Not Acting If the case review shows Tom was aware of declining signals but didn't escalate or intervene decisively, that's a different problem — likely either lack of confidence in intervention tactics or hesitancy to surface bad news to leadership. Interventions here: - **Escalation framework**: Tom needs explicit permission structures and clear criteria for when to pull in leadership, solutions engineering, or executive sponsors. If he's trying to manage Critical accounts solo, he'll lose them. - **Playbook access**: NovaStar Fintech and Vantage AI both fit the support spiral + declining super user competency pattern. There should be a defined playbook for this — executive outreach, platform re-onboarding offer, super user competency assessment. If Tom is improvising, that's a process gap, not just a Tom gap. --- ## The Immediate Crisis — NovaStar and Vantage AI Coaching is a medium-term intervention. NovaStar Fintech is already 48 days past renewal and in Critical territory — that account likely needs direct leadership involvement now, not a coached Tom trying to recover it alone. Vantage AI at $570K and Critical is close behind. For both accounts, the recommendation is: - Leadership joins the next customer call — not to undercut Tom, but to signal executive commitment to the customer and give Tom cover to reset the relationship - Tom owns the follow-through and communication cadence *after* that call — this keeps him in the relationship while adding the support he clearly needs --- ## Structural Support Regardless of what the case review reveals, Tom should not be managing Critical and At Risk accounts in isolation right now. Two concrete structural changes: - **Bi-weekly 1:1 with CS leadership** focused exclusively on his at-risk and critical accounts — not a status update, a working session where interventions are built and reviewed together - **Account redistribution** (as discussed) to remove the stable, healthy accounts from his plate so his bandwidth is concentrated where it matters --- ## Bottom Line Tom needs a root cause diagnosis before he needs a coaching prescription — the churn post-mortem is the first step, not an optional one. Run that, then layer in signal recognition coaching or intervention execution support based on what you find. In the meantime, NovaStar Fintech and Vantage AI need leadership escalation now — Tom shouldn't be managing those alone regardless of what the coaching plan looks like. --- **You:** Consolidate a summary of the recommendations from this chat. **Agent:** ## Recommendations Summary --- ### 1. Jacob R. — Workload Relief via Targeted Account Transitions **Accounts to transition:** | Account | Segment | ARR | Receiving CSM | |---|---|---|---| | Axiom Financial Group | Enterprise | $580K | Priya N. | | Zenith DevOps | Enterprise | $530K | Priya N. | | Zencore Platform | Mid-Market | $300K | Diana F. or Miranda L. | | Cortex Engineering | Enterprise | $480K | Diana F. | | Margin IO | Mid-Market | $185K | Diana F. or Miranda L. | | Summit Commerce | Enterprise | $380K | Priya N. or Diana F. | - Cortex Engineering and Summit Commerce have renewals in 36 and 50 days respectively — execute those handoffs immediately if moving - Orbit Global ($550K, expansion-flagged) is a borderline candidate — only move if the receiving CSM can immediately run the expansion motion - Keep all flagged accounts with Jacob: Quantum Ventures, Westbrook Media (At Risk, silent churners), IronBridge Data, Keystone Infrastructure (rescued, recovering), Ember Cloud, Flux Technologies (expansion candidates) --- ### 2. Tom C. — Coaching Intervention **Immediate actions:** - Run a structured post-mortem on the two unexpected churns — identify whether the breakdown was signal detection, intervention execution, or escalation hesitancy. The coaching plan follows from this answer - Leadership joins the next customer call for both NovaStar Fintech and Vantage AI — executive presence to reset the relationship, Tom owns follow-through - NovaStar Fintech is 48 days past renewal and Critical — treat as a potential loss and escalate all available levers now **Structural changes:** - Bi-weekly working sessions with CS leadership focused exclusively on Tom's at-risk and critical accounts - Redistribute Tom's stable healthy accounts (Xero Labs, Delta Dynamics, Redwood Platforms) to relieve bandwidth — his focus needs to be entirely on crisis accounts - Do not add any accounts to Tom's book until his Critical accounts are resolved **Coaching tracks (pending post-mortem findings):** - *If detection gap:* Structured weekly health score review cadence with explicit intervention thresholds; pair with Priya N. for joint account reviews as a live model - *If execution gap:* Formal escalation framework with clear criteria for pulling in leadership or executive sponsors; defined support spiral playbook so Tom isn't improvising on Critical accounts --- ### 3. CSM Capacity — Guardrails for Redistribution **Recommended receiving CSMs:** - **Priya N.** — Primary candidate for enterprise transitions; tenure and process discipline support absorption of 1–2 accounts - **Diana F.** — Strong candidate for mid-market and select enterprise; expansion-oriented book is a natural fit - **Miranda L.** — Can absorb one clean mid-market account; limit to one addition given Parcel Works watch item and pending enterprise promotion **Do not assign additional accounts to:** - **Tom C.** — Active crisis management; no capacity - **Marcus T.** — GalaxyCorp in active At Risk decline; two honeymoon accounts requiring attention - **Samara K.** — Six months tenure; Dragonfly Logistics not yet fully recognized as a risk; not ready for additional enterprise complexity --- ### 4. Open Decision Items These require leadership judgment before execution: - **Orbit Global** — Transition or retain with Jacob? Decision hinges on whether Priya or Diana can immediately activate the expansion motion - **Tom's account redistribution** — Which specific healthy accounts move and to whom should be confirmed after the post-mortem, once Tom's coaching track is clearer - **Miranda L.'s enterprise promotion** — She's flagged as ready; formalizing this would expand the receiving CSM pool for future redistribution and reduce long-term concentration risk in the team
Portfolio Health Trends
**You:** What health score trends from the last 8 weeks are most concerning and why? **Agent:** Here are the account with the most concerning 8-week trend organized by tier. --- ### Tier 1: Already Past Renewal — Immediate Exposure **NovaStar Fintech** and **Vantage AI** both sit in Critical territory with renewal dates already elapsed (–48 and –104 days respectively). Both carry the silent churner + support spiral dual flag. Their late-period scores are 37 and 26 — and critically, neither shows any stabilization. Score trajectories: NovaStar 50.7 → 42.3 → 37.3, Vantage AI 42.0 → 34.5 → 28.1. Value Performance Score is collapsing in both (NovaStar: 0.46 → 0.37 → 0.33; Vantage AI: 0.32 → 0.26 → 0.23). Tickets are rising into the late period (NovaStar: 4.9 → 6.1 → 7.4; Vantage AI: 6.4 → 9.5 → 12.7). These are not trends — they are active collapses with no floor visible. Both sit with Tom C., who has already had two unexpected churns. --- ### Tier 2: Steep, Unbroken Declines With Near-Term Renewals **Quantum Ventures** — Score: 50.7 → 41.3 → 35.3. Every signal declining in a straight line across all three periods. Value Performance Score: 0.46 → 0.36 → 0.32. DAU/MAU: 0.44 → 0.34 → 0.28. Super user competency: 0.56 → 0.45 → 0.37. Renewal in 373 days — far enough out that intervention is still possible, but the slope has not changed in 22 weeks. At Risk, silent churner. Jacob R.'s book. **Westbrook Media** — Nearly identical pattern. Score: 52.0 → 42.1 → 36.3. Value Performance Score: 0.46 → 0.37 → 0.31. DAU/MAU: 0.45 → 0.37 → 0.31. Silent churner + Seasoned Overextended super user. The super user competency decline is pronounced (0.58 → 0.47 → 0.47 — flat in late period but at a low level). Renewal is 668 days out, which means the trajectory has a long runway to worsen before it becomes a renewal conversation. Also Jacob R. **Xenon Infrastructure** — Score: 50.6 → 41.4 → 35.1. Silent churner + support spiral. Super user competency falling steadily: 0.55 → 0.45 → 0.38. Tickets rising: 5.0 → 6.1 → 7.6. Value Performance Score: 0.47 → 0.33 → 0.32. Priya N. carries this alongside PeakEdge Security and the Critical account Cascade Health Networks. Renewal is 711 days out, but the dual-flag combination is dangerous at this score level. **GalaxyCorp** — Score: 51.3 → 42.5 → 37.0. At Risk, silent churner. Seasoned Overextended super user. Value Performance Score: 0.47 → 0.39 → 0.34. DAU/MAU: 0.43 → 0.34 → 0.28. Renewal 597 days out. Marcus T.'s only At Risk account — and he's simultaneously managing two honeymoon+expansion candidates and Zinc DevOps. --- ### Tier 3: Declining But With Some Complexity **Cascade Health Networks** — Already Critical (score 41.4 → 33.9 → 27.9). The support spiral flag is confirmed by rising tickets (6.3 → 9.4 → 12.3) alongside collapsing Value Performance Score (0.32 → 0.27 → 0.22) and Time-to-Value (0.36 → 0.32 → 0.25). What makes this distinct from a pure silent churner is the visible customer distress — they're still engaging via tickets, but seeing no value. Priya N. is managing this, which is the best-case CSM assignment given her process discipline, but the score is still falling in the late period. **Wren Platform** — Score: 50.7 → 42.4 → 35.7. At Risk with no archetype flag beyond the base classification, but every signal is declining: Value Performance Score 0.45 → 0.35 → 0.30, DAU/MAU 0.45 → 0.38 → 0.33, super user competency 0.54 → 0.46 → 0.37. Seasoned Overextended super user. Renewal in 204 days — this is Diana F.'s primary watch item, and the renewal window is closing. **Sandstorm Labs** — Score: 50.6 → 41.3 → 35.6. At Risk, support spiral. Tickets rising: 4.4 → 5.9 → 7.0. Value Performance Score: 0.47 → 0.34 → 0.33. Super user competency steadily eroding (0.54 → 0.45 → 0.37). Seasoned Overextended. Tom C.'s account — adding to his already stretched book. --- ### The Pattern Worth Flagging at Portfolio Level Seven of the nine accounts above sit with either **Jacob R.** (Quantum Ventures, Westbrook Media) or **Tom C.** (NovaStar Fintech, Vantage AI, Sandstorm Labs — plus Delta Dynamics and Redwood Platforms in Positive). The Seasoned Overextended super user scenario appears in five of these nine accounts — super user competency decline is not incidental, it's a structural driver of the score deterioration across this cohort. The accounts without a visible floor in their late-period score — NovaStar Fintech, Vantage AI, Cascade Health Networks — are the most urgent. Every other account on this list has time, but no trajectory change yet. --- **You:** For NovaStar, Vantage, and Cascade Health, summarize the top 1 - 3 actions that would make the most immediate, positive impact. **Agent:** These three accounts share score collapse and rising tickets, but the signal mix driving each is different enough that the intervention priorities diverge. Actions are ordered by immediacy of impact. --- **NovaStar Fintech — At Risk · Silent Churner + Support Spiral** *Tom C. · $630K ARR · Renewal elapsed (–48 days)* **1. Escalate ownership immediately.** Tom C. has two unexpected churns behind him and is carrying the highest-risk book on the team. At –48 days past renewal with no floor in the score, this account needs CS leadership directly in the room — not as backup, but as primary. A senior sponsor-to-sponsor call is the right first move, not another CSM touchpoint. **2. Diagnose and triage the ticket spiral before the next customer conversation.** Tickets rose from 4.9 → 6.1 → 7.4 across periods. Before any value conversation lands, the team needs to know what's driving volume — unresolved issues compound distrust. Clear or escalate the open queue first, then reframe the relationship. **3. Anchor the conversation to a single, demonstrable value outcome.** Value Performance Score has collapsed to 0.33 and Time-to-Value is at 0.28. The customer has stopped seeing value and stopped progressing toward it. The only credible recovery path is a narrow, concrete win — not a roadmap pitch. Identify one high-value feature they're underusing and build a 30-day plan around it. --- **Vantage AI — Critical · Support Spiral** *Tom C. · $570K ARR · Renewal elapsed (–104 days)* **1. Get to the root cause of the ticket volume — this is the most urgent signal.** Tickets are at 12.7 in the late period (6.4 → 9.5 → 12.7), the highest of the three accounts. Value Performance Score is 0.22 and still falling. This customer is frustrated and not finding their way to value through normal channels. The ticket queue is the loudest signal of what's broken — it needs a structured review, not just response management. **2. Assess whether the super user situation is recoverable.** Vantage AI carries a Seasoned Overextended super user classification. Competency has declined from 0.57 → 0.49 → 0.47. If the admin is stretched or has fallen behind platform changes, no amount of CSM engagement will stick — end users won't get the experience they need. A direct super user enablement conversation needs to happen in parallel with the executive escalation. **3. Escalate to leadership — and assess churn probability explicitly.** At –104 days past renewal and a score of 26 with no stabilization, this account requires an honest internal assessment of whether recovery is achievable or whether the focus should shift to minimizing churn impact. Leadership needs to make that call, not the CSM alone. --- **Cascade Health Networks — Critical · Support Spiral** *Priya N. · $490K ARR · Renewal in 345 days* **1. Resolve the support spiral before any value conversation.** Tickets have risen from 6.3 → 9.4 → 12.3 while Value Performance Score has fallen from 0.32 → 0.27 → 0.22 and Time-to-Value from 0.36 → 0.32 → 0.25. The customer is actively seeking help and not getting traction. Unlike the silent churners, this account is still engaged — that's an asset, but only if the ticket issues get resolved and converted into forward progress. **2. Rebuild super user capability as the foundation for everything else.** Super user competency has declined steadily (0.30 → 0.27 → 0.23) and sits at the lowest level of the three accounts. Priya is the right CSM to execute a structured enablement program — this should be her first concrete deliverable with the customer, not a secondary initiative. Without a competent admin, platform adoption cannot recover. **3. Use the 345-day renewal window deliberately — build a milestone plan now.** This is the only one of the three with meaningful runway. Priya's process discipline is an advantage here. The right move is a co-created success plan with 60/90/180-day checkpoints — giving the customer visible progress markers and giving Shepherd Labs early warning if the trajectory isn't changing.
Quick Prioritization
**You:** What are the top 2 customers I should prioritize for scheduling a quick check-in call this week? Don't include our customers with At Risk or Critical health scores. I'm already communicating with those. **Agent:** Looking across the portfolio, excluding At Risk and Critical accounts, the two highest-priority check-in calls this week are: --- **Vellum Tech — Neutral · Positive (Silent Churner watch)** Vellum Tech carries a silent churner flag despite sitting in the Neutral/Positive border zone — which is exactly what makes it easy to miss. The super user competency score is Seasoned Overextended and declining (0.57 → 0.48 → 0.47), and while the overall health score looks deceptively stable (63–64 range), the underlying admin strain is a leading indicator. A check-in now, before the score moves, is the right call. **ARR:** $270K · **Renewal:** Dec 8, 2026 (597 days) · **CSM:** Tom C. --- **Dragonfly Logistics — Neutral · Silent Churner watch** Dragonfly is flagged as a silent churner watch item and sits at a score of 56 — stable on the surface, but the super user competency is Seasoned Overextended and declining (0.58 → 0.48 → 0.47), mirroring the same admin strain pattern as Vellum Tech. Notably, this is on Samara K.'s book, and per her profile, she may not yet have recognized the signal. A check-in this week creates an opportunity to surface it. **ARR:** $220K · **Renewal:** Jul 21, 2026 (92 days) · **CSM:** Samara K. --- **Why these two over others?** Both are Neutral-tier accounts with a silent churner flag or watch item, both have Seasoned Overextended super users with declining competency, and neither is in active escalation — meaning they're at exactly the moment where a light-touch call can prevent a harder conversation later. Dragonfly gets the edge on urgency due to the 92-day renewal window.
Stack Overview
Shepherd Labs runs on a lightweight, modern web stack with no backend database.
- → React + Vite for the client
- → Node/Express as a thin API proxy: no business logic server-side
- → Claude API (claude-sonnet-4-6)
- → Synthetic data engine built in JavaScript using a seeded random number generator
- → Hosted on Cloudflare Pages
How It Works
Shepherd Labs has no traditional backend database. All 75 accounts are defined in a constants file and processed by a custom data engine on page load, producing a full dataset including 22 weeks of signal history per account. That dataset is compressed into a structured snapshot and passed as context to the Claude API on every query. The agent reasons over the entire portfolio in a single call rather than querying a database: a design choice that keeps the architecture simple, eliminates query latency, and makes prompt caching highly effective. The data engine generates all account data on page load using a seeded random number generator, meaning the dataset is identical every time the app loads, with no database required. In a production deployment, the static data engine would be replaced by a server-side pipeline pulling transformed data from a warehouse like Snowflake, while the agent layer remains unchanged.
AI Agent Design
The agent reasons over a full portfolio snapshot rather than querying a database.
- → Agent receives a compressed CSV snapshot of all 75 accounts on every query
- → Responses stream word by word
- → Prompt caching enabled at the API level: up to 90% reduction in input token costs within active sessions
- → Chat state persists across tab navigation without a page reload
- → Per-message and full conversation export: copy as rich text or download as markdown
Cost Architecture
At current scale (75 accounts), each query costs approximately $0.024, with a full session of 6 to 8 queries running between $0.14 and $0.19. Prompt caching brings input token costs down by up to 90% within active sessions, keeping the AI layer lean. In a production deployment, the heavier costs sit upstream: ETL pipelines, data warehouse infrastructure, and signal transformation logic. Signal selection follows a three-question framework: does the signal have predictive value, is the underlying data available, and what is the fully loaded cost to produce it?