Suggested Slices — Grow by Subtraction.

Most book-optimization conversations in this industry start with growth — write more, retain more, cross-sell more. Suggested Slices works in the opposite direction. The mechanic is divestiture: identify the small, scattered segments that compound management overhead without contributing meaningful revenue, and carve them out so the agency can focus capacity on the segments that actually drive the multiple. Reagan's phrase for this pattern is "Grow by Subtraction." The pipeline ships exactly that mechanic.

§ 01 · What the pipeline actually doesFringe detection, four dimensions, daily.

Suggested Slices runs as a daily dbt pipeline (10:00 AM UTC, orchestrated through Dagster) operating four fringe-area detection macros — one per Slice dimension. Each macro identifies portions of the book where a single dimension represents 5% or less of total book premium. The output is a per-book inventory of fringe segments that meet the threshold and pass MVT.

The four macros.

• fringe_carrier — surfaces carriers representing ≤5% of book premium. Use case: a carrier relationship that has shrunk to a few stragglers; rarely worth maintaining contingency tracking or appointment compliance for it.
• fringe_lob — surfaces lines of business at ≤5% of book premium. Use case: a small Personal Lines tail on a commercially-focused agency, or a Life & Health vestige that no longer matches the agency's positioning.
• fringe_state — surfaces states at ≤5% of book premium. Use case: a few accounts in a state where you maintain licensing and carrier appointments without commensurate revenue.
• fringe_type_of_insurance — surfaces insurance types at ≤5% of book premium. Use case: a specialty type (umbrella, cyber, professional liability) the agency does not actively specialize in.

The MVT gate.

Each macro applies the Minimum Viability Thresholds before generating a Suggested Slice. The candidate segment must have at least 5 policies, at least $10,000 in premium, and represent at least 0.01% of total book premium. The first two conditions are surfaced in the seller UI; the third is enforced in code as a guardrail against degenerate Slices that don't have enough material to attract a buyer.

§ 02 · Why fringe segments cost more than they earnThe operational math underneath.

The conceptual case for divestiture rests on a math that operators recognize but rarely quantify. Every carrier relationship has fixed overhead: appointment compliance, contingency tracking, system configuration, producer training, carrier-portal management. Every LOB has its own renewal cadence, claim handling pattern, and E&O exposure. Every state has its own licensing, CE requirements, and regulatory follow-up. Every insurance type has its own underwriting model and operational details.

The cost of those overheads does not scale linearly with revenue. A carrier at 2% of premium incurs roughly the same compliance overhead as a carrier at 25% — but generates one-tenth the revenue to amortize it against. A state at 1.5% of premium requires the same licensing maintenance as a state at 30%. The fringe segment is, structurally, the worst-margin segment of the book — and removing it does not subtract proportionally from the bottom line. It usually adds.

"Grow by Subtraction" works because the fringe segment is the worst-margin segment of the book, not the smallest. Divesting it raises the average. The agency becomes more focused, more profitable, and more attractive to buyers — without writing a single new policy.

§ 03 · How to use a Suggested SliceThe seller-side workflow.

The pipeline does the surfacing. The seller does the deciding. The workflow:

One — review daily.

The Suggested Slices feed updates daily. Review it weekly, not daily — the value is in trend recognition, not per-day reaction. The same segments will surface day after day; the change you're watching for is which dimensions are accumulating fringe over time.

Two — filter to the dimension that matters.

If the agency is operationally focused on cleaning up carrier overhead, filter to fringe_carrier. If the cleanup is geographic, filter to fringe_state. Mixing dimensions in the same review session typically produces decision fatigue without action.

Three — verify against operational reality.

The pipeline surfaces what the data says is fringe. The seller is the only person who knows whether a particular fringe segment is strategic-fringe (a beachhead in a growing geography, or a producer specialty being built up) versus accretive-fringe (a vestige that should be divested). Strategic-fringe stays; accretive-fringe is the candidate for a Slice.

Four — publish.

Click through from the Suggested Slice to a draft listing. The Slice definition is pre-populated; you review, accept the listing agreement (Step 4 of 5), and publish. The Slice goes live immediately, anonymously, matched against active buyer profiles. Same mechanics as a manually-defined Slice — just upstream by the pipeline rather than upstream by the seller.

§ 04 · What's coming soonHotspot, and what to watch for.

The defensive (fringe-area) half of Suggested Slices ships. The offensive (Hotspot) half is coming soon. The distinction is worth understanding because it changes how the pipeline will eventually be used.

The offensive Hotspot logic.

Hotspot detection identifies book segments where demand from active buyers is concentrated — the inverse direction from fringe. Instead of asking "which segments are too small to keep," Hotspot asks "which segments are buyers actively seeking that you happen to hold." The model exists (int_book_appetite_score.sql in the production data warehouse); the wiring into the Suggested Slices recommendation surface is in progress. When it ships, the pipeline will produce dual-mission output: defensive fringe Slices (divest the underperforming) and offensive Hotspot Slices (monetize the demand-aligned).

What to plan for now.

Today, use Suggested Slices as the fringe-area pipeline it is. The Hotspot capability is real, in progress, and reasonably close — but the responsible seller does not plan around a feature that has not shipped. When Hotspot wires through, the pipeline becomes considerably more powerful; the fringe-area work you do today remains foundational either way.

§ 05 · What does not shipThe honest list.

Three claims occasionally surface in older Milly Books marketing literature that do not match the production code today:

• "AI Co-Pilot" framing. The pipeline is deterministic, rule-based, threshold-gated. No AI, no ML, no learned models. The recommendations are reproducible exactly from the data inputs.
• Demand-sensitive EBITDA multiples per Slice category. The Book Valuation Engine does not produce different multiples for "Hotspot" versus "Fringe" segments. It is a deterministic revenue-multiple formula. The 5–12× EBITDA tables you may have seen in older content do not reflect how it works.
• "Demand Metrics" dashboard. No Relative Buying Interest Score, no heat map, no real-time capital-volume counter ships today. What ships is the Buyer Connect Directory — browsable buyer profiles with their published acquisition criteria. That's the demand-transparency mechanic that exists.

The pipeline does what it does honestly. "Grow by Subtraction" is real, mechanical, and operationally useful. The marketing literature occasionally overstates it; the Tactical above is the version that maps to the production code today.