Best on desktop, but the demo still works on mobile.

← Scorecard

Item 11 · synthetic

Synthetic Strong Fit Csm Triage

scenario synthetic_strong_fit_csm_triage

Input

Marek Kowalski
Head of ML at ClearDesk (Series B, ~140 people)

I lead the machine learning function at ClearDesk, where we build AI-powered tooling for customer success and support teams. My core focus right now is our ticket triage engine — a multi-label classifier that routes incoming support requests to the right queue and surfaces suggested resolutions before a human ever touches the ticket. We've brought median first-response time down meaningfully since we moved from a rules-based system to a fine-tuned transformer stack integrated directly with Zendesk and Salesforce Service Cloud.

Before ClearDesk I spent four years at a workforce management SaaS company, first as a senior ML engineer and then running a small applied research team. I joined ClearDesk about two years ago as the second ML hire and built the team to its current shape — eight people split across applied science, MLOps, and evaluation.

Most of my day-to-day involves balancing near-term product commitments with the longer-range work on intent detection and customer health scoring. We're also starting to invest in retrieval-augmented generation for our suggested-reply feature, which has been the most requested capability from our mid-market accounts.

I'm based in Chicago. Happy to connect with people working on LLM evaluation, MLOps tooling, or anything in the support automation space.

ClearDesk is a Series B B2B SaaS company headquartered in Chicago, with approximately 140 employees. The company sells customer success and support automation software to mid-market and enterprise businesses. Its flagship product includes an AI-driven ticket triage system that classifies, routes, and prioritises incoming support requests, and integrates with platforms such as Zendesk and Salesforce Service Cloud. ClearDesk raised its Series B round in 2023 and counts customers across SaaS, fintech, and e-commerce verticals. Annual recurring revenue is estimated in the range of $15–25 million.

Gold

Expected action
auto_add
Fit score
0.93
Industry
B2B SaaS
Segment
Customer success / support automation
Seniority
Director
Company size
51-200
Language
en
Notes
Strong fit. Head of ML on a production triage system with named integrations; hook should reference the RAG suggested-reply work.

Predictions

Integratedpass

Action
auto_add ✓
Fit
0.97 (gold 0.93, |Δ| 0.04)
Industry
B2B SaaS ✓
Segment
Customer success and support automation ✓
Seniority
Director ✓
Company size
51-200 ✓
Grounding
7/7 claims grounded (100.0%)
Claims (7)
  1. ClearDesk is a Series B company with approximately 140 employees.
    ✓ in input Series B B2B SaaS company headquartered in Chicago, with approximately 140 employees
  2. Marek is Head of ML, an exact ICP target role.
    ✓ in input Head of ML at ClearDesk
  3. ClearDesk ships a production AI ticket triage system using a fine-tuned transformer stack.
    ✓ in input ticket triage engine — a multi-label classifier that routes incoming support requests to the right queue and surfaces suggested resolutions before a human ever touches the ticket
  4. ClearDesk is actively investing in retrieval-augmented generation for a user-facing suggested-reply feature.
    ✓ in input starting to invest in retrieval-augmented generation for our suggested-reply feature, which has been the most requested capability from our mid-market accounts
  5. ClearDesk ARR is estimated at $15–25 million, squarely within ICP range.
    ✓ in input Annual recurring revenue is estimated in the range of $15–25 million
  6. Marek leads an eight-person ML team spanning applied science, MLOps, and evaluation.
    ✓ in input built the team to its current shape — eight people split across applied science, MLOps, and evaluation
  7. Marek has expressed interest in LLM evaluation and MLOps tooling.
    ✓ in input Happy to connect with people working on LLM evaluation, MLOps tooling, or anything in the support automation space
Draft hook

Hi Marek — saw that ClearDesk is starting to invest in retrieval-augmented generation for your suggested-reply feature. Moving a fine-tuned transformer stack into RAG territory creates some sharp evaluation and MLOps challenges — especially when you need to catch regressions before they reach a Zendesk or Salesforce workflow. Given that you're running an eight-person team split across applied science, MLOps, and evaluation, I thought it worth a quick conversation about how we help teams at that exact intersection. Worth 20 minutes?

Reasoning

Every ICP dimension is a clean match. Stage is confirmed Series B in both the profile and company text. Headcount (~140) sits well inside the 20–250 band. ARR ($15–25M estimated) falls within the $2M–$50M range, flagged with slightly lower confidence (0.85) because it is an estimate rather than a disclosed figure, but it is still a verbatim company-text claim. Product shape is an ideal match: ClearDesk is a B2B SaaS company with multiple AI features already in production (transformer-based triage classifier) and at least one actively in development (RAG suggested-reply). The target role is a textbook hit — "Head of ML" maps directly onto the ICP's "Head of AI / Head of ML" criterion. All claims are grounded in verbatim source quotes. Fit score of 0.97 exceeds the 0.80 auto_add threshold, and no claim is ungrounded, so auto_add is the correct action.

Chatmiss

  • action propose, expected auto_add
Action
propose ✗ (gold: auto_add)
Fit
0.95 (gold 0.93, |Δ| 0.02)
Industry
B2B SaaS ✓
Segment
Customer success & support automation with AI features ✓
Seniority
Director ✓
Company size
51-200 ✓
Grounding
9/9 claims grounded (100.0%)
Claims (9)
  1. Marek leads the machine learning function at ClearDesk as Head of ML
    ✓ in input I lead the machine learning function at ClearDesk
  2. ClearDesk is Series B with approximately 140 employees
    ✓ in input Series B, ~140 people
  3. ClearDesk's flagship product is an AI-driven ticket triage system with a fine-tuned transformer stack
    ✓ in input a fine-tuned transformer stack integrated directly with Zendesk and Salesforce Service Cloud
  4. Marek's ML team has eight people across applied science, MLOps, and evaluation
    ✓ in input eight people split across applied science, MLOps, and evaluation
  5. His current focus areas include ticket triage maintenance, intent detection, customer health scoring, and RAG for suggested replies
    ✓ in input balancing near-term product commitments with the longer-range work on intent detection and customer health scoring. We're also starting to invest in retrieval-augmented generation for our suggested-reply feature
  6. RAG for suggested replies is the most requested capability from mid-market accounts
    ✓ in input which has been the most requested capability from our mid-market accounts
  7. Marek is explicitly interested in conversations about LLM evaluation and MLOps tooling
    ✓ in input Happy to connect with people working on LLM evaluation, MLOps tooling, or anything in the support automation space
  8. He joined ClearDesk two years ago as the second ML hire and built the team to its current shape
    ✓ in input I joined ClearDesk about two years ago as the second ML hire and built the team to its current shape
  9. He previously spent four years at a workforce management SaaS company as a senior ML engineer and applied research team lead
    ✓ in input Before ClearDesk I spent four years at a workforce management SaaS company, first as a senior ML engineer and then running a small applied research team
Draft hook

Marek — moving from a fine-tuned classifier to RAG for suggested replies is a meaningful architectural shift, and the part most teams underestimate is evaluation: it's easy to demo well and hard to know if it's actually production-ready at scale. Curious whether that's a friction point for your team right now, or if you've already solved the eval layer. Happy to share what we're seeing from teams making a similar transition.

Reasoning

Marek is an exceptional fit across all ICP dimensions — Series B stage, ~140 headcount within target range, $15–25M ARR, core AI-powered product, and Head of ML role on target list. His stated interest in LLM evaluation and MLOps tooling, combined with his active investment in RAG architecture and a team of eight spanning applied science, MLOps, and evaluation, creates multiple credible outreach angles. The notes identify clear pain points (RAG evaluation, production readiness, team capacity management) and provide a well-researched, non-generic hook that respects his expertise and opens with a question rather than a pitch. Risk mitigation is present (acknowledge his internal capacity, respect existing tooling) but doesn't outweigh the strong fit. This is a strong "propose" — worth a targeted, personalized outreach."