A deep read of A Step-by-Step Guide to Revenue Growth (31:49) from SaaStr Annual, given by Mark Roberge — Senior Lecturer at Harvard Business School, former CRO of HubSpot, author of The Sales Acceleration Formula, and founding partner at Stage 2 Capital.
Why this analysis exists
This is one of the source layers behind the BuildOS cold-email-icp-signal-design child skill. It supplies the leading indicator of retention (LIR) formula, the segment-tier view of product-market fit, and the Quality × Engagement lead grid that the skill uses to decide which accounts get the manual cold-outreach treatment vs. which get inbound-only handling vs. which get disqualified.
Core thesis
"75% of series-C startups fail. We're screwing up the scale. The first slide of your board deck should not be revenue growth — it should be revenue retention."
Most startups misread product-market fit. They measure new ARR and growth rate and feel validated; they raise a Series C; they scale headcount; revenue actually goes down because they were scaling on top of an unfinished foundation. The fix is to measure retention as a leading indicator before scaling sales.
Three checkpoints, in order:
- Product-market fit measured by customer value creation (leading indicator of retention).
- Go-to-market fit measured by unit economics (LTV/CAC > 3, payback < 12 months).
- Scale — added at a pace, not all at once.
Cold outreach (and SDR hiring) lives in checkpoint 3. Running it before checkpoint 1 is the Series-C failure pattern.
The Leading Indicator of Retention (LIR) formula
Churn is a lagging indicator — by the time it shows up in the data, you have already wasted a year. The fix is a single sentence that defines the leading indicator:
"P% of customers achieve E event(s) every T days."
Three slots:
- P = the percentage threshold (e.g., 70%).
- E = the activation event(s) that correlate with long-term value (Slack: team sends 2,000 messages; Dropbox: user adds a file in a folder on a device; HubSpot: customer uses 5 of 25 features).
- T = the window in which the event must happen (typically 30 or 60 days from signup).
Roberge's claim: this is the slide that should open the board deck. If the LIR is hitting target, the segment has product-market fit. If it is not, no amount of sales effort fixes the gap.
Green / Yellow / Red — segment-level PMF
The bigger insight in the talk is that PMF is never uniform across segments. Roberge walks through a six-box grid: small business / mid-market / enterprise, each split by inbound and outbound.
"We only had product-market and go-to-market fit in the top center. We really only knew how to sell mid-market companies through inbound demand gen."
The pattern he sees in nearly every company:
| Segment × Motion | LIR | LTV/CAC | Verdict |
|---|---|---|---|
| Mid-market × Inbound | 70% | 5M / 8mo payback / 6% logo churn | Green — scale this. |
| Enterprise × Inbound | Lower | Payback too long | Yellow — experiment, do not scale. |
| Mid-market × Outbound | Mixed | SDR efficiency unproven | Yellow — experiment, do not scale. |
| SMB × Anything | LIR low | Churn too high | Red — disqualify until product changes. |
The Series-C failure pattern: take the Series C money raised on Mid-market × Inbound performance and spend it adding outbound SDRs and enterprise reps in segments that have not earned PMF. The right play is to scale only Green segments and run small cross-functional teams to find PMF in the Yellow boxes.
For cold outreach: Green = manual outreach worth doing. Yellow = experiment only. Red = do not write.
The Quality × Engagement lead grid
Roberge's HubSpot example for connecting marketing and sales. Two axes:
- Quality (company-side, A/B/C by company size, revenue, or fit). 10,000+ employees → A; 100–10,000 → B; under 100 → C.
- Engagement (action-side, A/B/C by signal strength). Demo request → A; ebook download → B; blog signup → C.
Marketing gets paid not on lead count but on lead value, with prices attached to each cell of the 3×3 grid. An A/A lead might be worth $100 of credit; a C/C lead $10. Marketing's quota becomes a revenue-credit number, not a lead-count number.
Two implications for cold outreach:
- Inbound-converted demo requests are higher-quality than blog-driven leads even when both come from the same account. Outbound prioritization should treat inbound engagement as a signal, not a downstream consequence of outbound.
- The 3×3 grid is the disqualifier rubric. C/C leads are not "lower-priority outreach" — they are out of ICP until the product or pricing changes. Treating them as a stretch tier is what produces unprofitable outbound.
The hiring formula — coachability is the surprise
The other usable piece for the skill: when Roberge ran a regression on what predicted HubSpot rep success across 100+ hires, the top correlated trait was not closing ability or objection handling (both negatively correlated) but:
"Coachability was the one. It was those reps who checked all the other boxes — huge success — but they showed up on the first day and said, 'Mark, thank you for the training but I'd been selling for five or 10 years, I'll just be in my cubicle.' And that was the issue."
Negatively correlated traits in his data: closing ability, convincing-ness, objection handling. Positively correlated: preparation, domain experience, intelligence, coachability, curiosity.
For the outreach skill this is a tangent — but it lands one durable point: the skills that look like sales (closing, convincing) are anti-predictive of success in a consultative B2B motion. Cold outreach that sounds like sales is doing the same thing in writing.
What this contributes to the BuildOS ICP and Signal Design child skill
- LIR formula as ICP validation. The skill's segment definition includes an LIR target field: "P% of customers achieve E events every T days." If the segment cannot define one, the ICP is unverified — outreach is exploratory, not scaled.
- Green/Yellow/Red tiering for outreach prioritization. The skill's segment scorecard outputs one of three tiers. Green segments get manual, high-effort outreach. Yellow segments get experimental small-batch tests with a one-variable change. Red segments are explicitly excluded from campaigns.
- Quality × Engagement grid as the inbound-signal layer. When a target account has inbound signals (demo request, pricing-page visit, trial signup), the skill upgrades it on the Engagement axis. A C/A combination (small company, demo requested) is treated as a stronger signal than an A/C combination (big company, blog signup).
- The "we only have PMF in a slice" check. The skill's pre-outreach review asks explicitly: which slice of the ICP currently has measured PMF? Outbound runs only against that slice unless the campaign is explicitly a PMF experiment.
- The 75% Series-C failure rate as scope guardrail. When a user asks for a high-volume campaign in a segment without LIR or unit-economics data, the skill returns a refusal note pointing to the missing measurement.
Caveats
- 2019 talk, 31 minutes. Roberge's later work — The Science of Scaling book (2024) and the Stage 2 Capital podcast — formalizes more of this. The 2019 SaaStr talk has the cleanest single-slide articulation of LIR and the Green/Yellow/Red grid.
- HubSpot-scale data. The hiring-formula regression had 100+ rep observations. The PMF-by-segment pattern is observation across 25+ companies he has worked with, not formal research.
- HubSpot-context bias. The Quality × Engagement grid is built for inbound-heavy SaaS. For founder-led, recruiting, investor, or PR outreach, the inbound signal is weaker and the grid simplifies.
Source
- Transcript:
2026-05-15-mark-roberge-science-of-scaling-hbs.md(in-repo) - Video: A Step-by-Step Guide to Revenue Growth (SaaStr AI, 2019-02-26, 31:49)
- Book: The Science of Scaling — Stage 2 Capital (book page)
- Earlier book: The Sales Acceleration Formula — Mark Roberge
- Podcast: Science of Scaling