Field Notes - Jan 10, '26

Executive Signals

• Missed gross is the new heat map: budget by modeled lift, not clicks
• Incrementality beats attribution: strategy-level ROI drives promote/pause, not click credit
• v0 is the leverage: prototypes price vendors and compress timelines
• Memory compounds retention: cross-context identity turns one-offs into recurring usage
• Agents are the new crew: code throughput rises with measured guardrails
• Margin is the constraint: re-rank under ops limits, not channel cuts

CEO

Build v0 to Improve Vendor Leverage

A live v0 changes the negotiation. It gives you a rubric to evaluate vendors, reveals hidden integration costs, and improves pricing and timelines. If you acquire later, the discovery, APIs, and UX carry forward. Keep velocity by treating your internal lead as an advisor while an external team owns delivery.

• Timebox a 2–4 week v0 on the core conversational loop; make it demoable
• Tell suitors the build is underway; set a dated “we stop only if it’s worth it” gate
• Lock auth + 3–5 endpoint surface; external ships v0, internal secures data and rollout

Optimize Under Margin and Ops Constraints

When a channel is margin-constrained, treat margin and operational capacity as hard limits. Re-rank strategies and frequency within those limits instead of cutting profitable channels or spawning micro-segmentation. Protect the team from workflow sprawl and watch hours-per-dollar of gross as a guardrail.

• Recommend at most three changes monthly; avoid adding production SKUs
• Enforce no net headcount creep; track hours per $ gross
• Trim lowest-ROI subsegments before adding new workflows

Marketing

Replace Heat Maps with a Missed Gross Calculator

Heat maps describe, they don’t decide. Build a per-strategy lift score that predicts incremental gross and surfaces the top three actions with dollar impact and one-click budget allocation. Train on events, transactions, vehicles-in-operation, and socio-demographics; keep statistics hidden and show a single number plus next step.

• Gate to a monthly cadence; surface only actions above a clear lift threshold
• Render inside the core platform reps already use; avoid another dashboard
• Backtest across accounts and show missed gross for strategies not running

Demographics Belong in the Model, Not the Deck

Use socio-demographics to re-rank who to target and which strategy to run, not as sales theater. Only surface variables with proven causal influence on performance, and shift spend only when predicted lift clears a threshold.

• Add socio-demographics to features while preserving geography constraints
• Expose only variables with causal weight on strategy performance
• Move budget only when predicted lift exceeds a set threshold

Sales

Attribution That AEs Can Sell

“All matches” or “last match” teaches nothing. Move to modeled incrementality and ship a simple, defensible output: strategy-level ROI, missed gross, and a promote/pause recommendation. Keep the science inside; present one action and the financial impact.

• Run always-on strategy/geo holdouts; compute rolling lift over a fixed window
• Promote/pause only when lift deltas clear thresholds with minimum sample sizes
• Hide p-values; show a single recommendation with dollars attached

Programmatic Upsell via Dynamic Video

Expand share-of-wallet with automated, data-driven videos. For accounts using Product A but not B, auto-generate a 30–60 second clip with their metrics and predicted lift, then push via CRM sequences. Make it continuous and template-driven, not rep-dependent.

• Define product-gap triggers; refresh creatives monthly from live metrics
• Graduate creatives that 2x demo rate or clear a strong reply-rate bar
• Keep ops light: reuse templates; one video per use case per month

Sales Intel Beats Cold Opens

Ship monthly “what to say now” briefs by market: competitor moves, OEM trends, and local signals like weather or seasonality distilled into talk tracks. Treat as messaging intel tied to territories, not as target list changes.

• Deliver briefs inside CRM mapped to account lists
• Include only signals with repeatable lift across markets
• Update monthly to match field cadence

Product

Memory as the Retention Moat

Assistants with memory win repeat usage. Persist user context across channels and sites (with consent) so answers improve and switching costs rise. Make memory scope configurable to satisfy partner sensitivities, but ship with remembrance assumed.

• Implement user-scoped identity and consent; default to remembered preferences
• Offer per-tenant controls: on/off, channel-only, cross-channel
• Include an admin kill switch to de-risk rollout

Engineering

Default-to-Ship for v0 Conversational AI

Optimize for time-to-learning: default to open source when viable; rent now, own later; prefer variable over fixed. Keep payment founder-controlled to bypass procurement and wrap the system as a black box tightly coupled to the core platform.

• Use vendors for telephony/SMS now; document a migration plan before launch
• Keep early spend on a single card; formalize contracts after usage stabilizes
• Define a narrow API boundary (3–5 endpoints) to isolate integration work

Build the Software Factory for Agents

Design the pipeline around coding agents and agentic workflows. Pair a senior engineer with an agentic, CLI-first loop, persist agent memory over repos, and measure throughput and quality to guide broader adoption.

• Stand up an agent dev loop; pair it with a senior engineer
• Persist agent memory on repos; plan for model portability
• Publish internal metrics (lead time, defects, $/feature) to steer rollout

Model Costs Where It Matters

Early on, channel fees are rounding errors next to inference. Track LLM spend per resolved conversation and optimize there. Keep everything else simple and variable; migrate to fixed contracts only after volume steadies. Separate P&L lines to keep debates focused.

• Track cost per resolved conversation; set a pre-GTM ceiling
• Use usage-based pricing in v0; move to fixed only after stability
• Split inference vs channel fees; automate the latter, scrutinize the former