Finance & AP
Finance / AP Track
Close faster, audit harder, and let the numbers carry their own proof — without ever trusting a figure you didn't trace.
Mike Hart · VP Finance
Laura Dominguez · AP / Finance
Vincent Lo · Finance
Anyone touching close, AP, or reporting
You're on this track if
- You own or support month-end close, AP, expense review, or KPI reporting
- You re-key the same numbers between RentManager, Sheets, and a report every month
- You spend hours reconciling, chasing receipts, or writing variance commentary by hand
- You want AI to do the grunt work — but every figure still has to be defensible
Your path through the 4 Levels
Everyone follows the same spine. Here's what each level means specifically for finance work.
L1
Better Conversations
Read the AUP, learn to prompt for reasoning instead of answers.
For you: this is where you build the verification habit — "show me how you got that number" — before you trust anything AI hands back.L2
Your Workspace
Set up your Claude Teams project and "interview your AI" so it knows your context.
For you: load the Company State + Glossary and your chart-of-accounts conventions once, so you stop re-explaining GL codes every chat.L3
Run the Workflows
Use the sanctioned Workflow Skills — your monthly cycle, on rails.
For you: this is the payoff level. F1 KPI Cycle, F2 Expense Audit, and Smart Bills + AP all live here.L4
Build New Workflows
Spot the next repetitive task and co-build a new skill for it.
For you: CAM reconciliation, owner statements, AP coding rules — if you do it every month by hand, it's a Level 4 candidate.Your Skills
Yours = a workflow you own and run. Expert = knowledge you summon by topic. Both live in the Skills Catalog.
Pulls last month's KPI dashboard from RentManager, validates against the prior snapshot, writes property-level variance commentary into the KPI Sheet.
Reviews Divvy + VentureX card transactions against receipts, flags missing receipts and policy outliers, produces a management-grade audit memo.
Pulls queued bills, validates against POs and budget, flags coding anomalies, preps the batch for your review before it hits RentManager.
Month-end close, bank rec, variance analysis, intercompany, accruals, trial balance, consolidation.
GL codes, chart of accounts, PM fees, late-fee mechanics, loss-to-lease, vacancy loss, write-offs, owner statements.
Why a metric moved — variance explanation, benchmarking, occupancy/revenue/expense-ratio trends, same-store.
Distribution waterfalls, preferred return, capital accounts, K-1s, management/acquisition fees, NAV, multi-entity consolidation.
Prompts to try in your voice
Start here, then make them yours. These run in your Claude Teams project with your data loaded.
"Here's the October trial balance and last month's. Walk me through every account that moved more than $1,000 or 10%, and tell me which ones look like timing vs. real variance."
Turns a 2-hour variance read into a first draft you sanity-check — instead of building it from scratch.
"Draft the variance commentary for the KPI Sheet for these 12 properties. Flag anything you're inferring vs. reading directly from the data, and don't smooth over the ones that look bad."
Asks for the honest version, not the flattering one — and makes the AI mark its own guesses.
"These 40 card transactions are missing receipts. Group them by cardholder and vendor, and tell me which are likely recurring SaaS vs. one-off purchases I should chase first."
Prioritizes the receipt chase instead of treating all 40 as equal.
"This AP batch has 18 bills. Compare each to its PO and budget line. List only the ones where the coding or amount looks off, with the specific reason."
Surfaces the exceptions so your review time goes to the 3 that matter, not all 18.
What NOT to do (Finance edition)
- Never trust a number you didn't trace. AI output is a draft, not a source. Every figure that lands in a financial report, board packet, or owner statement gets cross-referenced to RentManager / the GL / the bank — by you. The AI's confidence is not evidence.
- Don't discuss pre-disclosure financials in an open chat. Current-quarter actuals, fund-specific capital, and the named acquisition pipeline only go into a sanctioned Claude Teams project with data you pasted — never a personal AI account, never "from memory."
- Don't paste resident PII you don't need. Bank account numbers, SSNs, full payment histories — strip them. If a task needs an account identifier, use the resident ID, not the SSN.
- No tenant or applicant scoring. Collections and bad-debt work stays at the portfolio/aggregate level. Do not ask AI to approve, deny, or score an individual applicant or tenant — that's prohibited (AUP §2.2) and routes to a human. Per-resident risk scoring already runs through the sanctioned, human-gated delinquency model; don't recreate it in an open chat.
- Collections can't segment by protected class — or its proxies. No targeting delinquency outreach by income source (Section 8 / SSDI), family composition, age, or neighborhood. If a prompt drifts that way, stop and flag it to Carlos. The Fair Housing skill is the reference.
Your first 30 days
A realistic ramp. You don't have to do it all in week one.
Week 1
Read the AUP and do Personal AI Setup. 30 minutes total. Then use Claude for five real finance tasks — a quick reconciliation question, reading a dashboard, drafting an email. Build the reflex.
Week 2
Set up your Claude Teams workspace. Load the Company State, Glossary, and your GL/chart-of-accounts conventions once. Interview your AI so it stops asking you what a "lot rent" line is.
Week 3
Shadow your Workflow Skill. Watch Carlos run it end-to-end on this month's real cycle (Mike → F1 KPI · Laura → Smart Bills · expense audit → F2). Ask every "why did it do that" question now.
Week 4
Run it solo, once. You drive, Carlos is on standby. That solo run starts your 30-day unassisted clock — the Rock 2 finish line. From here it's reps, not training.
Note for review: The prompt examples and 30-day ramp on this page are a first draft authored from the finance workflows we already run. Mike / Laura / Vincent should gut-check the specifics — real task names, actual pain points, whether the week-by-week pacing is right — during the Cohort 1 walkthrough before this is treated as final.