Jeff Wilson exit-debrief — services pricing strategy + AI initiative ROI
Why this is in the vault
Two load-bearing insights from founder's exit conversation with Mammoth Growth's CEO. (a) The hourly-billing-as-wedge framing reframes a pricing model the broader services discourse treats as obsolete — turns out it functions as a retention + procurement-avoidance lever, not just a billing convenience. (b) Verified unit economics for an agentic-services R&D investment: a peer company in our market converted $500k + 3k hours into ~3x EBITDA expansion and 6-month payback. Both insights directly inform RDCO's services-as-software thesis and any future Sanity Check piece on AI-services unit economics.
⚠️ Confidentiality
Specific Mammoth financial data points below (EBITDA %, retention rates, R&D dollar amounts, expansion rates) are Mammoth-internal. Founder is the source of the disclosure post-exit, but the data belongs to Mammoth. Do NOT publish externally — no Sanity Check piece, no public Ray Data Co content, no HQ surface beyond the vault. If we ever cite the pattern publicly, use orders-of-magnitude framing without the specifics (e.g., "we've seen consultancies double EBITDA after agentic R&D" rather than "10% → 30%").
Conversation summary
Thread 1 — Hourly billing as wedge, not as liability
Industry-wide narrative pushes "outcome-based pricing" or "fixed-fee projects" as the modern services-pricing answer. Mammoth has stuck to hourly billing despite the discourse, which Ben had seen as a risk. Jeff's reframe:
- Hourly billing keeps the contract relationship open-ended. "Foot in the door" stays in the door.
- Once in, the agency can sell, extend, expand the next workstream organically.
- Fixed-fee projects that complete quickly force the agency back through enterprise procurement/provisioning each re-engagement. Procurement cycles are months and high-effort.
- Smoothing the project across time is therefore both a retention lever AND a cash-flow lever — recurring billable hours look like recurring revenue from the agency's cash-management perspective.
The takeaway is subtle and counterintuitive: hourly is not a worse pricing model intrinsically; it's a different shape that buys retention at the cost of perceived rigor on outcome ownership.
Thread 2 — Agentic R&D investment paid back in 6 months
Jeff's framing of Mammoth's AI initiative ROI:
- Investment: $500k + 3k hours of internal time on building out agentic workflows.
- EBITDA impact: 10% → 30% (3x expansion).
- Retention rate: 100% for the cohort of clients where Mammoth has been running the AI workflows.
- Expansion rate: doubled.
- Rates: increased.
- Payback period: ~6 months for the R&D build-out.
Jeff is clearly seeing the ROI. The investment thesis on agentic-services is paying off on a directly-measurable services-business KPI set.
Mapping against Ray Data Co
Three concrete RDCO-load-bearing maps:
Services-as-software thesis validated at peer-N scale. Vacca (ColdIQ, $7M ARR) and Mammoth (more mature, larger, longer-running) now both validate that agentic R&D drives services-business unit economics. The "Spine → Agents → Loop" sequencing from [[2026-05-19-alex-vacca-3-phases-ai-layer-services-as-software]] is the same shape Mammoth ran. This is corroborating evidence that the agentic-services bet is durable, not hype.
Pricing-shape framework for any RDCO services future. If RDCO ever offers paid services (current state: founder is the only operator; no paid services), the hourly-billing-as-wedge framing argues against the modern instinct to default to fixed-fee. Worth retaining as a default-pricing decision input. For the AI-COO bet specifically: hourly retainer with capacity caps may beat the natural urge to bundle into a flat monthly subscription.
Concrete ROI calibration for the Sanity Check positioning audience. Ben's audience (data engineering managers, ops leaders, services-shop operators) hears AI-ROI claims constantly but rarely sees verified unit economics from a peer-shape company. Mammoth's 6-month payback at $500k + 3k hours is the kind of concrete data point that grounds the discourse. Use the pattern (orders-of-magnitude) without the specifics in any public-facing Sanity Check work — see ⚠️ Confidentiality above.
Founder-specific carry-over: Ben spent 5+ years inside Mammoth. He carries this institutional knowledge as default-priors when thinking about RDCO. Worth recognizing that "hourly is bad" is a learned-from-discourse position he was holding against his own validated experience. The data point reframes that.
Open threads / follow-ups
- Synthesis-queue candidate: agentic-services unit economics convergence — Vacca ($7M ARR ColdIQ) + Mammoth (mature) + State of AI Dev 2026 (Claude > ChatGPT paid devs). Three independent data points triangulating on the same shape. Worth a concept-article synthesis when synthesis-work skill ships. Inputs: this note + [[2026-05-19-alex-vacca-3-phases-ai-layer-services-as-software]] + [[2026-05-19-state-of-ai-dev-2026]] + Mammoth's earlier vault notes.
- Forward question: does the hourly-billing-as-wedge framing collapse when the buyer is itself an AI-COO agent (not a human enterprise-procurement function)? At RDCO's L6+ trajectory, the procurement cycle that hourly billing exists to avoid may simply not exist. Watch for this.
Related
- [[2026-05-19-alex-vacca-3-phases-ai-layer-services-as-software]] — Vacca's spine/agents/loop sequencing, same shape Mammoth ran
- [[2026-05-19-state-of-ai-dev-2026]] — Claude paid-conversion data point (services-as-software market validation)
- [[2026-05-19-cloudflare-cyber-frontier-models]] — Glasswing harness pattern (peer architecture)
- [[2026-04-27-jeff-agentic-velocity-quantification-draft]] — earlier Jeff/Ben artifact on agentic-velocity measurement (if extant)
- [[../mammoth-growth/STRATEGY.md]] if extant — Mammoth project folder anchor