The Territory Carve Playbook: Scored Universe to Balanced Rep Books

This playbook covers how to split a deduplicated target account universe into rep books for the next period. It is written for founders handing accounts to their first AEs, sales leaders and the RevOps people who run the carve. The result is one assignment table. In it, every book fits under a selling-hours ceiling, every segment passes a variance band you named in advance, and every contested row points to a written rule rather than a negotiation.

What a finished territory carve produces

A finished carve is one assignment table, with one row per account and one summary row per rep. Every row names the rule that put it there. When a rep asks why an account is not theirs, the answer is a cell in that table, not a meeting.

The rest of this guide uses four terms:

• Book: the set of accounts one rep owns for the period.
• Carve: the act of splitting the account universe into books.
• Keeper: an account that stays with its current rep through the carve because a written rule says so.
• In-flight: an open opportunity that already exists on the day the carve is announced.

The table is done when it passes four tests. Someone who was not in the planning meeting can check each one:

1. Hours. The hours each book needs add up to no more than the rep's selling-hours ceiling.
2. Variance. Within each segment, the books fall inside a potential-variance band that was written down before the carve started.
3. Locks. Keeper and in-flight rows are marked as locked, and their hours count against the owning rep.
4. Rule trail. Every row involving a split parent or a territory boundary cites a named tiebreak rule.

This guide starts where building the universe, ICP gating and queue ranking stop. It assumes you already have a deduplicated universe with one row per company. It ends with the books loaded into the CRM.

ACCOUNT ROWS
account_id | ultimate_parent_id | buys_independently (Y/N) | segment | tier | potential_score | hours_per_year | rep | lock (keeper / in-flight / none) | rule_applied | tiebreak_id

REP SUMMARY ROWS
rep | segment | accounts | locked_hours | assigned_hours | total_hours | ceiling | headroom | potential_total | segment_mean | pct_from_mean

SEGMENT CHECK ROWS
segment | variance_metric | sd_formula (population / sample) | band | result | pass / fail

The territory carve procedure, step by step

The carve has nine steps, run in order: resolve hierarchies, score, set the ceiling, lock keepers and in-flight deals, assign the rest, check variance, write tiebreaks, cut over and monitor. Each step has a done condition, so you know when to move on.

Two notes on the order.

First, published sources disagree on whether the rules or the boundaries come first. A territory-carving glossary and SPOTIO both put the in-flight rule ahead of the carve. Map My Customers puts boundary drafting first. I side with publishing the rules first. Reps argue less about lines when they can see the rules that drew them. And a rule written before any boundary exists cannot be bent to fit one.

Second, no published source says how long any of these steps takes. Treat any duration you see quoted as an estimate. On your first run, time each step and write the times into the plan. Next period you will have real numbers instead of a guess.

Before step 1, decide which model you are running. Gradient Works contrasts dynamic books, where the rep's capacity stays fixed and the accounts in the book refresh, with static territories, where the account list stays fixed for the period. This guide builds a static carve. The hours math is the same for both models. A dynamic book simply reruns steps 5 and 6 more often.

How many accounts per sales rep: the hours ceiling

A rep can own as many accounts as their selling hours, divided by the hours each account needs, allow. On the common 1,500-hour selling year, that is 2 to 6 global enterprise accounts or 50 to 150 non-named accounts. An account-count benchmark is only valid if your hours per account match the assumption behind it.

The standard bottom-up formula assumes about 2,000 working hours a year, with 75% spent selling. That gives 1,500 selling hours a year, or 125 a month. Divide that by the hours each account needs and you get the number of accounts a rep can work at once. At 3 hours per account per month, 125 / 3 = 42 accounts.

Jon Miller publishes hour bands per account per year for each tier. Dividing 1,500 by those bands gives exactly the account ranges Winning by Design publishes by deal size:

That match is the useful finding: the account-count ranges are hour math. Better Sales Comp makes the same point the other way round, with 750, 250 and 30 hours per account a year giving books of 2, 6 or 50 accounts. If your team spends more hours per account than the band assumes, the published count overstates what a rep can carry.

Which selling year to use

Two bases are published. Gradient Works and Jon Miller use 1,500 selling hours. ORM uses 1,100, paired with 22, 8 and 3 hours per account per year. The 1,100 base shrinks every book by about 27% compared with 1,500. Neither is right in general. Pick the one that matches how much of your reps' time actually goes to selling, write it down, and use it for every rep in the segment.

When capacity and quota disagree

Gradient Works gives a quota-based example: a $45,000 monthly quota and a $15,000 average selling price (ASP) need 160 accounts per month per rep. The same source's capacity math gives 42 accounts at 3 hours each. Covering 160 accounts in 125 hours leaves about 0.78 hours per account. Both cannot be true at once. Either the touch model changes, or the quota does. Settle this before the carve. A carve that satisfies the quota math and breaks the hours math produces books nobody can work.

Setting the ceiling per book

• Book hours = the sum of hours per account across every account in the book, including locked keepers and in-flight accounts.
• Ceiling = the selling year you chose.
• Headroom = ceiling minus book hours.

If headroom is negative, the book fails, however fair its potential looks. What the ceiling proves: the book is workable on paper. When it misleads: when the hours-per-account figures were borrowed from a benchmark rather than measured. Before trusting a band, check last period's activity records for a sample of accounts in each tier.

Territory planning account scoring: potential from public signals

Score every account on three things: fit, timing and reachability. Write the weights down before the carve, then map each score onto the same tiers you used for hours. No public source has tested weights for territory balancing, so the weights are a decision you own, not a benchmark you can look up.

The published weightings contradict each other:

• An intent-led model weights 0.4 fit and 0.6 intent, for short sales cycles.
• A fit-led model weights 0.6 fit and 0.4 intent, on the grounds that "intent data at low volume is mostly noise".
• A three-term model weights 50% fit (firmographic match), 30% in-market (intent plus triggers) and 20% reachability (effective coverage of the decision-maker).

For carving, I start from the three-term model. Reachability is what separates two accounts with identical firmographics when one has a reachable buyer and the other does not. A book full of high-fit accounts with no reachable buyer is high potential on paper and dead in practice.

Signals: what each proves and how it misleads

• Firmographic fit (industry, headcount, stage). Proves: you can sell there. Misleads: when the record is stale. One account-scoring publisher estimates that 20-30% of firmographic records go stale within 12 months, so re-pull the data at carve time instead of reusing last year's scores.
• Hiring velocity, banded as flat headcount, 5-10% growth or 15%+ growth (Salesmotion's bands). Proves: budget is moving. Misleads: when the growth is in functions that never buy what you sell. Count hires in the buying function, not company-wide.
• Intent and triggers. Proves: timing. Misleads: at low volume, where it is mostly noise.
• Reachability (buyer seats): the number of people at the account who hold a title that signs off on or champions the purchase. Proves: there is someone to sell to. Misleads: when the title does not match the job. In Refolk's index, each 25-profile sample contained 1 headline in 25 (4%) that was not the target title, so read the headlines before counting a seat.

Counting buyer seats by hand across a few thousand accounts is the slowest part of scoring, and it is usually the term teams quietly drop. Refolk takes your definition of the seat as a plain-English query and returns the people from LinkedIn, GitHub and the open web. That makes reachability a column you fill in at carve time rather than a guess.

From score to tier

Map each score onto the four tiers in the table above. The tier label then carries both potential and hours, so one row tells you what the account is worth and what it costs to work. Do not let a high score push an account into a tier whose hour band does not match how you would actually work it. That is how a book can balance on potential and blow through its ceiling.

Which variance band to hold each segment to

Hold each segment's books to one named metric and one band. The most commonly published healthy line is a coefficient of variation below 15%, with a CV above 25% as the trigger to redesign. Choosing the metric is a policy decision, because the same set of books can pass on one metric and fail on another.

The coefficient of variation (CV) is the standard deviation of potential across books divided by the average, expressed as a percentage. Range over mean is the gap between the largest and smallest book divided by the average. Published bands:

• Map My Customers: potential within 10-15% variance.
• ChapsVision: CV held below 15 to 20%.
• Maptive: CV below 0.15 is healthy, and redesign is triggered when it crosses 0.25. Within a peer group, the top rep should hold no more than roughly twice the bottom rep.
• Maptive: a Gini coefficient on territory potential above 0.35 means a split is needed. The Gini coefficient measures how unevenly potential is spread, from 0 for perfectly even upward.

You will also see a 1.5x max-to-min tolerance quoted, but it is not established in any public source. If you want it, adopt it as your own policy and label it that way.

Here is one set of four books, using Boogieboard's published values of $4.0M, $4.2M, $4.8M and $5.0M, scored against each band:

The books are the same and the verdicts differ. Range over mean looks only at the two extreme books, so one outlier fails the carve. CV averages across all the books, so one outlier gets diluted. Even the standard-deviation formula matters: the same books score 9.2% or 10.6% depending on which function the spreadsheet uses.

Compare books only within a segment. An enterprise book and an SMB book are meant to differ, and putting them in one CV calculation makes a good carve look broken.

What a pass proves: the books are fair on the score. What it does not prove: that the score is right. If the potential score is wrong, a balanced carve just spreads the error evenly.

Keeper, in-flight, and parent-child rules

Apply the rules in a fixed order, from most specific to most generic: keepers, then in-flight deals, then subsidiaries that buy independently, then ultimate-parent routing, then segment and geography rules. SAP's documented precedence for territory rules works the same way: specific rules go at the top of the list and generic ones toward the end.

Keepers

Louie Bernstein's founder rule is to keep "only the accounts where losing them would hurt the year, or where the relationship truly runs through you." The test works for any rep, not just founders. No public source sets a numeric cap on keepers. Set your own cap and write it down. Then count every keeper against the keeping rep's hours ceiling, so a rep who keeps several large accounts takes on fewer new ones. Keepers are where carves turn into negotiations. A written test, plus counting keepers against hours, removes the incentive to hoard.

In-flight deals

Two versions are published:

• Most common: open opportunities stay with the original owner until they close, while the account and any new opportunities move to the new owner immediately.
• SPOTIO's variant: the originating rep gets full commission on any deal already at proposal stage, there is a defined cutover date for new pipeline, and reps get at least 2-4 weeks notice.

The second version is stricter about stage. That helps if you suspect early-stage opportunities get created just before a carve to hold on to accounts.

Parent-child families

The default is to route an account to the rep who owns its ultimate parent, the top company in its ownership chain, even when the lead or account matches to a subsidiary. The documented exception is that "an independently buying subsidiary can retain its own territory while remaining linked to the corporate family." A company only counts as a parent with majority ownership: 50% or more of voting power.

The buys-independently flag is the one reps will push hardest on. A separate address does not prove it. The evidence I accept is a separate prior purchase, or a named buyer seat inside the subsidiary with its own budget. Without one of those, the subsidiary routes to the parent's owner.

Boundary straddlers

No public source gives a tiebreak for multi-location accounts that straddle a boundary, for example HQ address versus largest site. So write one. Pick a basis, give it a rule ID, and apply it to every case. Reps rarely contest the choice itself. They contest inconsistency, when the HQ rule wins for one rep and the largest-site rule wins for another.

How a territory carve goes wrong

Most failed carves fail in one of eight ways. Each has a check you can run before the announcement. This section matters more than any other, because every one of these failures looks fair until a rep finds it.

1. Balancing on account count. Two reps with 100 accounts each look even. Then one turns out to hold 20 high-quality accounts and the other 100 weak ones (Gradient Works' Jane and Joe example). Check: report the potential CV next to the counts, and never publish counts alone.
2. Choosing the metric that passes. The four books in the variance table above fail on range over mean and pass on CV. Check: the metric, the formula and the band are in a document dated before the carve.
3. Ignoring workload. "A territory can be perfectly balanced on potential and completely uncoverable." Check: hours per book against the ceiling, for every book.
4. False parents. Investors get recorded as parent companies. "A startup that took Series B funding from Sequoia Capital is not a subsidiary of Sequoia." Check: apply the 50% voting test to every parent link before routing.
5. Stale signals. The potential score rests on records that go 20-30% stale within 12 months. Check: re-pull headcount and hiring data at carve time.
6. Activity that games coverage. A note logged in the CRM resets the reclaim clock. Meticulosity requires that "activity" is defined narrowly enough that a note does not reset the clock. Check: write down what counts as a meaningful touch.
7. Deciding in-flight after the fact. "Retrospective decisions about commission damage trust disproportionately." Check: confirm the in-flight rule was published before the announcement date.
8. Override flags left on. In SAP, manually assigned accounts "are skipped from Realignment Runs, as long as the field Override Territory is flagged." Salesforce has an account field called "Exclude from territory assignment rules" that has the same effect. Check: audit both flags before the run and clear any that have no written reason.

Failures 1 and 3 interact, and the clearest way to see it is to plot every book on potential and hours at once:

Balanced on potential and uncoverable on hours is still a broken book.

Tiebreaks and the dispute log

Decide every contested row with a named rule, record it in a dispute log, and send it to one named decider whose decision stands for the period. Bernstein's template puts it in one line: "Disputes go to [the Fractional Sales Leader]." Replace the bracket with a real person's name.

The log does two jobs. During the carve it stops the same argument from being reopened. Next period it shows you which rules generated the most disputes, and those are the rules to rewrite. If one rule keeps coming up, the problem is the rule, not the reps.

RULES OF ENGAGEMENT (published [date], effective [cutover date])
R1 Keepers: A rep keeps an account only if losing it would hurt the year or the relationship runs through that rep. Cap: [n] keepers per rep. Keepers count against the rep's hours ceiling.
R2 In-flight: Open opportunities stay with the current owner until they close. The account and any new opportunities move on [cutover date].
R3 Independent subsidiaries: A subsidiary flagged buys_independently = Y, on evidence of a separate prior purchase or its own buyer seat, keeps its own territory and stays linked to the family.
R4 Families: All other accounts route to the owner of the ultimate parent, defined by 50%+ voting power. Investors are never parents.
R5 Boundary tiebreak: An account with sites in more than one territory goes to the territory of its [HQ address / largest site], applied to every case.
R6 Precedence: R1, then R2, then R3, then R4, then segment and geography rules.
R7 Variance: Segment books are held to [metric], [population / sample] SD, band [value].
R8 Disputes go to [named decider]. Decisions are final for the period and recorded below.

DISPUTE LOG
date | account_id | raised_by | claim | rule_cited | decision | decided_by | table_changed (Y/N)

Every row in the assignment table that a rule decided carries that rule's ID in the tiebreak_id column. When a rep disputes an account, the decider reads the cited rule, checks the row against it and records the decision. The decider does not reopen the rule itself mid-period. Rule changes wait for the next carve.

Keeping the carve current after cutover

Announce with 2 to 4 weeks notice and load the final table into the CRM, because a carve kept in a spreadsheet drifts: "the CRM and the plan diverge within weeks." After that, track coverage, take back neglected accounts on a written threshold, and recalculate variance every quarter.

Coverage and reclaim

SPOTIO defines the coverage ratio as the percentage of accounts contacted in the last 90 days, with a target of 80%+ of high-priority accounts each quarter. Reclaim means taking an inactive account back from its owner. Published reclaim thresholds vary:

• Gradient Works' rules-of-engagement template reclaims "after 60 days of inactivity", down from 90 days in an earlier version.
• Amplify uses no meaningful engagement within three to six months.

The thresholds follow sales-cycle length, not a standard, so set yours to match your cycle and write it down. Pair it with a narrow definition of a touch. On one HubSpot portal, roughly 25,000 accounts had no activity in 90 days. They were flagged as pooled, with ownership left untouched. That is a reasonable middle path when a hard reclaim would cause more disruption than it fixes.

Rebalance triggers

Recalculate the named variance metric every quarter, as Maptive suggests, and rebalance when one of these fires:

• CV above 0.25 in any segment.
• Gini coefficient on territory potential above 0.35.
• Revenue potential shifting by more than 20%.

Vacant books

A carve sometimes produces a book with no owner. How hard it is to fill depends on the segment and the country. In Refolk's index of professional profiles:

This is an inference from talent supply, not a documented time to hire. Mid-market AEs were not counted, and the counts carry the same 4% off-title noise noted earlier. If the carve leaves a UK enterprise book open, start the search before the announcement rather than after it. A Refolk query such as "Enterprise account executives in London currently at Series B to D SaaS companies" returns that shortlist in the same session you use to finalise the books.