AI Advisory for Private Equity Portfolios in B2B SaaS.

Support headcount scales linearly with ARR if nothing changes. A $40M ARR portco runs 8 to 14 support FTE depending on product complexity. Tier 1 routine tickets (password resets, status, billing) make up 40 to 60% of volume. The standard chatbot-wrapping-OpenAI approach deflects 5 to 15% and erodes CSAT.

The category leaders (Decagon, Sierra, Ada, Intercom Fin) ship a retrieval-grounded agent that reads the help center, the product docs, and the past ticket history, and resolves Tier 1 plus a meaningful share of Tier 2 with the full transcript and the citation trail attached. Mindbody (Vista portco) saves $1.25M annually on this pattern. The trap: optimize for deflection on Tier 1, optimize for proactive intervention on Tier 2/3.

Sales cycles in mid-market B2B SaaS stretched from 60 to 90 days in 2022 to 90 to 140 days in 2025 as buying committees expanded. SE bench gets stretched thin on technical discovery calls and 40-hour RFP responses. Pipeline coverage looks fine; conversion is where the cycle dies.

A copilot trained on the firm's past RFP responses, product docs, and win/loss data drafts the response, surfaces the technical discovery questions the SE should ask, and produces a first-draft 80% complete in 4 hours instead of 40. Avalara (Vista portco) reports 65% faster response time using a similar pattern. Build-on-top with Glean or Notion AI works at the SE-tooling layer; custom RAG on the firm's own corpus works once the document library is structured.

Churn shows up in usage data 60 to 120 days before the renewal call. Logins drop 18%. Feature adoption narrows. Power-user activity falls off. By the time the CSM gets the renewal-risk flag from the standard health-score dashboard, the customer has already decided. Reactive saves are 30 to 40% effective. Proactive plays are 65 to 80%.

A model scoring accounts on usage pattern shifts, surfacing the top 30 at risk weekly, and routing them to the right CSM with a one-paragraph "here's what's going on" plus a segment-specific intervention playbook recovers 60 to 80% of at-risk revenue. Build is the right answer here for most portcos; the model runs on usage data the SaaS already owns. Catalyst and ChurnZero ship the workflow side if the in-house team doesn't have capacity.

BDR teams spend 40 to 60% of their day researching accounts before outreach. Most of that research surfaces information that's already public (recent funding, leadership changes, tech stack signals) but takes 20 to 40 minutes per account by hand. SDR productivity caps at the rate they can research-then-write, not the rate they can sell.

An agent that pulls 10-K filings, recent press, LinkedIn intelligence, and tech stack signals, then drafts the outbound message with the specific hook, takes the research-to-outreach cycle from 30 minutes to 3. Clay, Apollo, and Common Room ship this with various depth-of-research options. The build version works on the firm's own ICP definition and CRM data.

Product roadmaps at PE-backed SaaS portcos drift between two failure modes: half feature requests from the loudest 10 customers, half competitor mirror. The actual signal (what 200+ customers said in interviews and support tickets and CSM notes) lives in unstructured data nobody clusters systematically.

A clustering agent that ingests call transcripts (Gong, Chorus), support tickets, and CSM notes, then surfaces themed customer needs ranked by ARR weight and segment, gives the product team a roadmap input grounded in actual customer signal. Gong ships parts of this; Dovetail ships the research-ops side. Custom builds on top of the firm's call recordings work for portcos with engineering bench.