How to build a 90-day enterprise territory plan with AI (when you have 40 named accounts)

TL;DR: Score accounts on evidence you already hold rather than researching all of them equally, spend real research effort only on the top tier, and convert the output into a weekly commitment you can actually keep. A plan that names six accounts and one action each beats a beautiful forty-account document nobody opens.

The problem: the research burden is impossible

Carrying 20-50 named accounts is standard in enterprise sales, and the honest arithmetic is that deep research across all of them is not achievable — the burden is simply too high to maintain. So most reps do what is rational under the constraint: they work the three accounts that are already talking to them and hope something inbound happens with the rest.

The result is a territory plan written in January to satisfy a manager, and a year spent reacting. The fix is not more research. It is triage: decide fast which accounts deserve effort, then go deep only there.

What you need

• Your account list with whatever your CRM holds — last activity, historical spend, open opportunities.
• Any install base or existing-service data if you sell into existing customers.
• Public signals you can gather quickly: funding, hiring, leadership changes, expansion, regulatory events.
• Claude or ChatGPT, plus Perplexity or another search-connected tool for the research pass — I use the split described in the competitor analysis workflow: search tool gathers with citations, reasoning model synthesises.

Step 1: Score before you research (30 minutes)

Do not research forty accounts. Score them on what you already have.

“Here is my account list with CRM data. Score each account 1-5 on four dimensions, using only the data provided:

FIT: how closely they match the profile of deals we have historically won (I will paste that profile below).
TIMING: evidence of a trigger — contract end date approaching, recent leadership change, expansion, or a project mentioned in past notes.
ACCESS: do we have any warm relationship at all, including dormant ones.
VALUE: realistic deal size band based on their size and our comparable deals.

Output a table sorted by combined score. Where you lack data for a dimension, mark it UNKNOWN rather than guessing — do not average around missing data. At the end, list which specific missing data points would most change the ranking if I filled them in.”

That last question is the highest-value output in this entire workflow. It tells you the five facts worth an hour of research, instead of researching everything shallowly.

Step 2: Segment into three tiers, with different rules

• Tier 1 — Pursue (5-8 accounts). Deep research, a named plan, multithreading, weekly action. These get real time.
• Tier 2 — Develop (10-15 accounts). One trigger to watch, one relationship to build, a monthly touch. No deep research until a trigger fires.
• Tier 3 — Monitor (everything else). Alerts only. No proactive time. Being honest that these get nothing is what makes tiers 1 and 2 real.

Most territory plans fail because they refuse to write down that some accounts get no effort. Then attention spreads evenly and nothing gets enough.

Step 3: Deep research on tier 1 only (2-3 hours total)

For each of the 5-8, run a structured research pass with a search-connected tool. Separate searches beat one mega-prompt:

1. “What has [company] announced publicly in the last 12 months regarding technology, infrastructure, expansion, or restructuring? Cite sources with dates.”
2. “Who currently leads technology, infrastructure or IT at [company]? Cite the source and its date.”
3. “What is [company] hiring for in technical and operations roles right now? What does the pattern suggest they are building?”
4. “What regulatory, competitive or market pressures currently affect [company]’s industry in [region]?”

Then verify the people. Leadership data goes stale fast and AI search results routinely surface an executive who left eighteen months ago. Confirm on the company’s own site or the person’s own profile before you use a name in an email. Getting this wrong in a first approach is unrecoverable.

Step 4: Convert research into a hypothesis, not a summary

“Based on the verified research on [account], write: (1) a one-sentence hypothesis about a business problem they plausibly have that we can address, stated as a hypothesis rather than a fact; (2) the evidence supporting it, with sources; (3) the evidence against it or what would disprove it; (4) which specific role would own this problem; (5) the single question I would ask to test the hypothesis in a first conversation.

If the evidence does not support any credible hypothesis, say so — ‘no clear angle’ is a valid and useful answer.”

Permitting “no clear angle” is important. Otherwise you get a manufactured rationale for every account and lose the ability to distinguish genuine opportunities from wishful ones.

Step 5: Turn the plan into a calendar

A plan that is not on your calendar is a document. Convert it:

“Turn this account plan into a 13-week schedule. For each week, specify: which tier 1 accounts get an action and what specifically, which tier 2 accounts get a touch, and one hour of research time on a named topic. Constraints: no more than 5 proactive actions per week, every action must be something I can complete in under 45 minutes, and each must have a defined output. Front-load the accounts with the nearest timing trigger.”

Five actions a week is roughly what survives contact with an actual sales week of meetings and firefighting. Plans that assume twenty get abandoned in week three, which is worse than a modest plan that holds.

A worked example of the scoring output

Three rows from a 38-account list, to show what useful output looks like:

Account A — combined 17/20. Fit 5 (manufacturing, multi-site, matches our four largest wins). Timing 5 (CRM note from a year ago: incumbent contract expires Q1). Access 3 (dormant relationship — a former colleague sold them something in 2023). Value 4. Verdict: tier 1, and the timing trigger means this is a this-quarter account, not a someday account.

Account B — combined 11/20. Fit 4. Timing UNKNOWN (no contract data at all). Access 1. Value 5 (large — which is why it has been on the list for two years). Verdict: tier 2. The high value is seductive but there is no trigger and no access. Watch for a trigger; do not spend Q3 on it.

Account C — combined 13/20. Fit 3. Timing 4 (announced expansion into two new sites). Access 5 (we already serve them at three locations). Value 3. Verdict: tier 1 despite the middling total — existing customer plus a live expansion is the cheapest pipeline available to anyone.

And the most useful line the model produced: “Contract end dates are missing for 24 of 38 accounts. This is the single field that would most change the ranking — timing currently drives placement for the 14 accounts where it is known, and those accounts may simply be the ones with better record-keeping rather than better opportunity.”

That is a genuinely uncomfortable observation, and it is correct. Half a day spent filling in contract end dates would improve this territory plan more than any amount of additional research on the accounts you already know about.

Step 6: Review fortnightly, replan quarterly

Every two weeks, ten minutes:

“Here is what I committed to and what I completed. Which accounts have had no progress for three weeks or more? Which of my hypotheses have been tested, and what was the result — supported, disproved, or still untested? Based on this, which accounts should move tier, in either direction? Be willing to recommend demoting an account I have invested in.”

Demotion is the discipline that makes this work. Sunk cost is the strongest force in territory management: the account you have worked for six months feels like it deserves more time precisely because you have spent so much already.

Honest limits

• Scoring reflects your history, not the future. If your historical wins came from one industry, the model will rank that industry highest and you will never discover an adjacent one. Manually promote one or two wildcard accounts each quarter.
• Public signals favour large companies. Mid-market firms generate far less public information, so they score low on timing evidence and get systematically underweighted. Compensate deliberately.
• It cannot know what your colleagues know. The account someone lost badly three years ago, the executive with a history with your company — that lives in people’s heads. Ask before you plan.
• Stale leadership data is the recurring failure. Verify every name.

Once tier 1 accounts start engaging, the next problem is making sure you are not single-threaded in any of them — that is the stakeholder mapping workflow.