Use AI where a document answers the question. Build one pack per building holding the lease clauses, budget and actuals, let an assistant draft service-charge replies that quote the clause and the budget line, extract lease events into a sheet a person verifies, and draft owner-report commentary from a locked extract. Check every date, share and total against the source.
The pattern is the same for all three jobs: the model works only from documents you hand it, says where each statement came from, and a person verifies before anything leaves the office. The risk is a confident statement about what a lease says that nobody opened the lease to confirm. Service charge, meaning the tenants' share of the cost of running a building, is called CAM or operating-expense recovery in some markets, and the method is identical.
Start with a building pack, not a tool
Every useful answer in this job lives in a handful of documents, and the assistant is only as good as what it can see. Assemble one pack per building and keep it in a folder with a fixed naming rule (building, document, date). It should hold:
- each lease, with every deed of variation, side letter and licence for alterations, since the lease as signed is rarely the lease as it stands;
- the current service-charge budget, last year's reconciliation and the year-to-date actuals;
- a one-page summary of each supplier contract with its renewal date and rate;
- the apportionment table, showing each unit's area and share;
- previous tenant queries and the replies a manager approved, as a tone and policy reference.
The worked example below uses an invented eight-unit office building with 40,000 sq ft of lettable area. Costs are shared by area, so each tenant's share is its area divided by 40,000.
| Unit | Area (sq ft) | Share |
|---|---|---|
| 1 | 6,000 | 15% |
| 2 | 5,000 | 12.5% |
| 3 | 5,000 | 12.5% |
| 4 | 4,000 | 10% |
| 5 | 6,000 | 15% |
| 6 | 4,000 | 10% |
| 7 | 6,000 | 15% |
| 8 | 4,000 | 10% |
| Total | 40,000 | 100% |
The year's service-charge budget is $400,000, against $351,000 last year. The second table shows the lines, and it is the table every later check in this tutorial refers back to.
| Budget line | Last year | This year | Change |
|---|---|---|---|
| Cleaning | $66,000 | $72,000 | +$6,000 |
| Security | $56,000 | $64,000 | +$8,000 |
| Lift maintenance | $26,000 | $28,000 | +$2,000 |
| Common-parts utilities | $44,000 | $60,000 | +$16,000 |
| Repairs and maintenance | $70,000 | $76,000 | +$6,000 |
| Insurance | $44,000 | $48,000 | +$4,000 |
| Management fee | $22,000 | $24,000 | +$2,000 |
| Waste and landscaping | $11,000 | $12,000 | +$1,000 |
| Sinking fund | $12,000 | $16,000 | +$4,000 |
| Total | $351,000 | $400,000 | +$49,000 |
The Complete Excel Course with AI Integration
From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.
By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner
Buy nowService-charge questions: a reply drafted from the pack
The Unit 4 tenant emails: "Our service-charge demand has gone up by nearly 14%. What is driving it, and what is the sinking fund line?" Without a pack, a manager hunts through the lease, the budget and last year's reconciliation, which can take twenty minutes or more. With one, an assistant can draft the reply in a minute, provided the prompt stops it from improvising. Suppose the pack's lease says the tenant pays "the Proportion" under clause 7.2 and Schedule 3 (10% for Unit 4), clause 7.6 lets the landlord collect a reasonable contribution towards replacing plant, and the budget notes explain the three biggest movements.
You are drafting a reply for a property manager to review. Reply to the
tenant's email using ONLY the attached building pack: lease clauses, budget,
last year's reconciliation, supplier summaries and earlier approved replies.
Rules:
- State the tenant's share and cite the lease clause and schedule it comes from.
- Quote budget figures exactly as they appear. Do not calculate. If a figure
is not in the pack, write [FIGURE NEEDED].
- Give a reason for a change only where a budget note or supplier summary
gives it, and name that document.
- Do not say a cost is or is not recoverable unless a lease clause says so in
terms. If the clause is unclear, write "for the manager to confirm".
- Do not promise refunds, credits or reductions. Do not mention other tenants.
- Under the draft, list: documents used, clauses quoted, anything you could
not answer.
Tone: polite and plain.This is the kind of draft that comes back, with invented figures and notes throughout:
Dear [first name],
Thank you for your email. Your share of the service charge for Unit 4 is 10%, as set out in clause 7.2 and Schedule 3 of your lease (4,000 of the building's 40,000 sq ft).
The budget for the year is $400,000, against $351,000 last year, an increase of $49,000. Your share therefore moves from $35,100 to $40,000. The largest increases are common-parts utilities (up $16,000, because the energy supply contract was renewed at a higher unit rate, per the supplier summary), security (up $8,000, reflecting a second night patrol added from February, per the budget notes), and cleaning and repairs (each up $6,000).
The sinking fund line is $16,000, up from $12,000. Clause 7.6 allows the landlord to collect a reasonable contribution towards the future replacement of plant, and the budget notes link this year's figure to the lift controller replacement planned in three years' time. Whether this contribution is reasonable is for [manager name] to confirm.
Kind regards,
[your name]
Before it goes anywhere, the manager runs five checks, and each is quick because the draft names its sources:
- The share. 4,000 divided by 40,000 is 10%, and clause 7.2 and Schedule 3 really do say that. Open the lease at the cited clause; a model can cite a clause number that does not exist.
- The demand. 10% of $400,000 is $40,000, 10% of $351,000 is $35,100, and the $4,900 difference is 10% of the $49,000 increase.
- The reasons. Each is traced to a named document. A reason with no document is struck out.
- The percentage. The model left out the tenant's "nearly 14%" because the pack has no row for it. The manager adds the figure from the sheet: $49,000 divided by $351,000 is 14.0%.
- The recoverability sentence. It stays as "for the manager to confirm" unless the manager, or the landlord's lawyer, has decided.
Some queries should never get a draft at all. Route these straight to a person: anything in dispute, any mention of legal action, a request to inspect the accounts, a cost that isn't in the budget, and any query about a clause the model flagged as unclear. The tutorial on service-charge queries and repairs in block management software covers the residential end of the same problem, including how platforms log the queries.
Data checks that run in the sheet, never in the chat
Models write fluent sentences and can slip on arithmetic, and a tenant will spot a wrong total faster than a wrong adjective. Keep the sums in the spreadsheet and let the assistant write the words. These are the checks to build once per building and re-run whenever the budget changes, shown with the example building's numbers:
| Check | How it works | Example result |
|---|---|---|
| Areas add up | Unit areas sum to the lettable total | 6,000 + 5,000 + 5,000 + 4,000 + 6,000 + 4,000 + 6,000 + 4,000 = 40,000 |
| Shares add up | Apportionment column sums to 100%, or to the figure the leases require | 15 + 12.5 + 12.5 + 10 + 15 + 10 + 15 + 10 = 100 |
| Budget lines add up | Lines sum to the stated budget | Nine lines sum to $400,000 |
| Demand equals share times budget | Each tenant's demand recomputed independently | 10% of $400,000 = $40,000 for Unit 4 |
| Big movers have a note | Any line up more than 10% on last year needs a documented reason | Utilities +36.4%, sinking fund +33.3% and security +14.3% need notes; the other six lines are all under 10% |
| Year-to-date against pro rata | Actuals compared with the budget's share of the year elapsed | Repairs: $52,000 spent at month six against $38,000 pro rata, so $14,000 over |
The "big movers" check is the one that pays for itself. It tells you before the tenant does which three lines will draw questions, so the budget notes can be written first and the reply drafted from them. Each of the three has a note in the pack; if one did not, the draft above could only have said "for the manager to confirm".
The same sheet handles the year-end reconciliation. Suppose actual costs come in at $412,500, which is $12,500 over the $400,000 budget. Unit 4's 10% share is $41,250 against the $40,000 it paid on account, so the balancing charge is $1,250. An assistant can then draft the covering note from the reconciliation table, explaining which lines ran over and citing the notes, while the demand figure itself comes from the sheet. If exports from your accounting system arrive with inconsistent unit names or merged cells, the guide to cleaning messy data is worth a read before you build the checks on top.
Lease events: a dated sheet that a person has verified
A missed break date or rent review can cost an owner real money, and the dates sit in documents nobody reopens until a tenant calls. The job for AI here is extraction, not interpretation. Run one lease at a time, with its variations attached, through a prompt like this:
Extract the lease events from the attached lease and every variation.
For each event give: event type, date or rule, clause and page, the exact words
quoted, and the notice required.
Event types: term start; expiry; break option (who may exercise it); rent
review (dates, basis, upward-only or not); rent-free or stepped rent end;
renewal option; service-charge year end; insurance duty; assignment and
sub-letting consent rules.
If a date is not stated, write "not stated". If a date is derived from the
term, mark it "derived". Do not work out notice deadlines.
End with: documents read, documents you could not read, and anything that
differs between the lease and a variation.Here is an illustrative extract for Unit 6, a ten-year lease from 1 March 2023 with one deed of variation:
| Event | Date or rule | Clause, page | Notice |
|---|---|---|---|
| Expiry | 28 February 2033 (derived from "a term of ten years beginning on 1 March 2023") | 1.1, page 2 | None stated |
| Break option | Tenant only, 1 March 2028 | 14.1, page 21 | "Not less than six months' prior written notice" |
| Rent review | Fifth anniversary, open market, upward only | 5.3, page 9 | None stated |
| Rent review (variation) | Third anniversary, in place of the fifth | Variation 1, clause 2, page 3 | None stated |
| Sub-letting | Landlord's consent, not to be unreasonably withheld | 12.2, page 17 | Written application |
The fourth row is the one the closing note should flag, and it is the reason the prompt asks for variations: the lease and the deed disagree, and the abstract must show the variation, not the original. A person then verifies every expiry, break, review and notice field against the cited page and signs the row with a date. Unverified rows stay out of the calendar, because a reminder built on a wrong date is worse than no reminder.
The sheet works out reminder dates, not the model. For the break above, six months before 1 March 2028 is 1 September 2027, so the sheet sets internal reminders at 12, nine and seven months before the break date and names the manager responsible. Treat that 1 September date as a working figure. Where a lease counts "clear days" or deems notice served on a later day, the real deadline moves, and the landlord's lawyer should confirm the counting rule for any break or review that matters.
Where lease abstracts go wrong, and a test to measure it
Expect the same handful of failures, and design the verification step around them:
- The amendment never made it into the pack. The abstract is faithful to a lease that no longer applies. The sign is a lease file with no variations at all for a building that has had tenants for years.
- A scan misread. A break date reads 2026 instead of 2028 because the page was skewed. The sign is a date earlier than the term's start or a clause quote with odd characters in it.
- Who holds the option. A break that only the landlord may use is recorded as mutual, or the reverse. The sign is a quoted clause that names the party and an abstract that doesn't.
- Counting rules left out. The abstract records "six months' notice" and drops the clause on how notice must be served. The sign is a notice field without a serving address or form.
- Two leases blended. Run several leases in one chat and details drift between units. Run one lease per chat, every time.
Measure it before you rely on it. Take five leases you already know well, abstract each with twelve fields, and mark every field right, wrong or wrongly "not stated". An invented but plausible result is 52 right out of 60, with five wrong and three wrongly empty. Two errors came from a misread scan, two from a variation that wasn't in the folder and one from the option holder, and all three empties were on the scanned lease. That is not good enough to skip verification, but it shows where to spend it: scans and variations first. The tutorial on abstracting a commercial lease with page citations goes deeper on the prompt and the citation habit.
Owner reports: commentary from a locked extract
The monthly owner report is mostly numbers, plus a paragraph or two that explains them. Export the figures first, freeze them, and let the assistant write only the paragraph. The extract needs five blocks: the rent roll and what was billed and collected, the arrears ageing, occupancy, the lease events in the next eighteen months, and the service-charge position against budget, plus any notes from the manager on what happened.
Verify the numbers in the sheet before any words are written. For the example building these are the checks, with invented figures:
| Check | Working | Result |
|---|---|---|
| Arrears movement | Billed $130,000 less collected $118,400 adds $11,600 to last month's arrears of $30,000 | $41,600 |
| Ageing buckets add up | Current $18,000, 30 to 60 days $11,600, 60 to 90 days $7,000, over 90 days $5,000 | $41,600, matching the movement |
| Occupancy | Occupied 36,000 sq ft of 40,000 (Unit 8 is vacant) | 90.0% by area |
| Weighted average lease term | Area in thousands of sq ft times unexpired years to expiry, at a report date of 30 November 2026: 6 x 4 + 5 x 2.5 + 5 x 6 + 4 x 1.5 + 6 x 3 + 4 x 6.25 + 6 x 7 = 157.5, divided by 36 occupied (Unit 6 has 6.25 years left on the lease above) | 4.4 years to expiry, weighted by area |
| Service charge to date | Repairs and maintenance $52,000 against $38,000 pro rata | $14,000 over, which is 36.8%, so a note is required |
State the basis of the lease-term figure in the report. Many owners expect it weighted by rent rather than by area, and the two can differ a lot when one big tenant pays a low rent. Others want it measured to the first break rather than to expiry. Counting Unit 6 only to its March 2028 break, 1.25 years away, turns 4 x 6.25 into 4 x 1.25, so the total drops to 137.5 and the figure to 3.8 years. Put both numbers in the report when a break is close, so the owner sees the risk. With the checks passed, the assistant gets the extract and the manager's note and writes the commentary. An illustrative result:
Rent billed in the month was $130,000 and $118,400 was collected, leaving arrears of $41,600 at month-end, of which $5,000 is more than 90 days old. Occupancy is 90.0% by area, with Unit 8 (4,000 sq ft) vacant. The weighted average unexpired lease term is 4.4 years to expiry by area, or 3.8 years to first break. Repairs and maintenance are $14,000 over the pro-rata budget, which the manager's note attributes to a burst water main in March.
What the manager still adds: whether the over-90-day balance is in dispute, what is being done about the vacancy, and any break date that falls inside the next twelve months, taken from the verified sheet and not from the model. Tagging each figure to a source row, as in the workflow for factsheet commentary with a named approver, works equally well here if your owners are particular. The tutorial on AI for real estate asset management reports and lease abstracts covers larger portfolios.
Four kinds of tool and what to ask each vendor
You do not need all four. Most small managers start with the first and the last, and add a platform or a data tool when volume justifies it.
| Type | What it does for this job | Named examples | Cost shown |
|---|---|---|---|
| General assistant on a business plan | Drafts replies, abstracts and report text from the pack you attach | ChatGPT, Claude, Gemini, Microsoft Copilot | ChatGPT Business Standard $20 a user a month billed annually ($25 monthly); Claude Team $20 annual ($25 monthly); Microsoft 365 Copilot Business $21 annual ($25.20 monthly). ChatGPT Business and Claude Team need two seats at minimum |
| Commercial property platform with built-in AI | Works on data and documents already in the system | Re-Leased's Credia: Action reads emails and suggests tasks, reminders, notes and work orders, acting only "with your approval"; Extract pulls key information from leases and invoices; Advise answers lease questions "with direct references to the source". Action comes with the Core and Pro packages, while Extract and Advise are a Credia Plus add-on | Not published; pricing on request |
| Lease-data platform | Turns lease documents into structured data, with expert review behind the AI | Prophia, whose page lists lease abstraction, critical dates and CAM review and describes itself as built for enterprise teams | Quote only: a per-tenant annual subscription, with no figures shown |
| Reminder automation | Pushes verified dates into calendars or task lists | Calendar rules in Google or Outlook; Zapier | Zapier Free is 100 tasks a month with two-step Zaps only; Professional starts at $19.99 a month billed annually ($29.99 monthly) for 750 tasks |
Put the same questions to any vendor in the middle two rows before a demo turns into a purchase:
- Where are our documents stored, and are they used to train anyone's models?
- Can I click from any extracted field to the clause and page it came from?
- Can a person verify a field and lock it, with a record of who and when?
- How are variations and side letters linked to the lease they change?
- What does it cost per lease at our volume, and what happens to the data if we leave?
- How does it handle apportionments where a vacant unit's share sits with the landlord?
The third question matters most. A platform that shows its source for every field makes verification a click; one that doesn't makes it a second abstraction. Tool prices change often, so use the cost column as a starting point and read each vendor's current page.
Three decisions that stay with a person
Drafts and sheets speed up the preparation. They should never make these calls:
- Whether a disputed cost is recoverable. The lease decides, and ambiguous clauses go to the manager and, where money is at stake, the landlord's lawyer.
- Serving or answering a notice. Break notices, review triggers and counter-notices carry formalities that a draft can't guarantee. A person serves them, using the lawyer's wording.
- Agreeing a rent review outcome or an arrears plan. The assistant can summarise comparables or draft the letter. The owner and the manager decide the number.
A thirty-day start for a small portfolio
Pick the building with the most tenant queries and the messiest paperwork, because that is where the saving shows. In week one, assemble the building pack and write the apportionment and budget tables. In week two, abstract each lease, verify every date against the page, and load the verified rows into the calendar. In week three, take last quarter's real tenant queries, draft replies from the pack, and compare them with what was actually sent, noting every difference in facts. In week four, produce one owner report with the extract-and-check routine and compare the time against the last manual report.
Put rough numbers on it first. A portfolio of 40 leases, abstracted by hand at 90 minutes each, takes 60 hours. If a verified AI-assisted abstract takes 35 minutes including the check, the same job takes about 23 hours, and the 37 hours saved are worth about $1,850 at $50 an hour. For queries, assume 15 a quarter take 25 minutes each to answer from scattered documents, and 10 minutes from a pack. That saves about four hours a quarter. These are assumptions, so use your own times, and remember that verification is the time you must not trade away. For the wider picture of where else AI fits in a manager's week, AI landlord updates and owner statements covers the owner-communication side.
Questions property managers ask before they start
Can an AI assistant tell a tenant what they owe?
No. The demand comes from your accounting system and the lease. An assistant can explain how a figure was reached, quoting the clause and the budget line from your building pack, but a person checks the explanation before it is sent. Amounts, due dates and anything in dispute should never come from a model's memory.
Is it safe to put leases into a chat assistant?
Leases hold commercial terms that tenants and owners expect you to protect. Use a business plan, where ChatGPT Business, Claude Team and Microsoft 365 Copilot don't train on your content by default, and read your management agreement for confidentiality terms. Leave out bank details and personal data the task doesn't need.
How accurate are AI lease abstracts?
The vendor pages checked for this tutorial give no accuracy figure, so measure it. Take five leases you already know well, have the tool abstract them, and compare field by field: dates, rent, review basis, break terms, notice rules. Track errors by field. Clean text scans well, and poor scans, side letters and amendments cause most of the misses.
Do we need lease abstraction software or will a chat assistant do?
For a small number of leases, a chat assistant plus a verification sheet usually does the job, because a person checks each date anyway. Specialist platforms start to earn their cost when the portfolio is large, when leases arrive constantly or when you need data feeding other systems. Ask for the cost per lease at your volume and compare it with your verification time.
Further reads
- AI for Commercial Appraisers: Lease Abstracts and Income Prep — Turn abstracts into income figures for valuation work
- AI Assistant for Property Management: A Resident FAQ Bot From Leases — Build a question-answering bot from lease documents
- Dilapidations Schedule: Draft Items From Lease Clauses With AI — Draft end-of-lease schedules from the clauses
- How Property Managers Use AI to Triage Maintenance Requests — Sort incoming repair requests by urgency
- ChatGPT for Property Management: Resident and Owner Message Templates — Message templates for owners and occupiers
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Re-Leased home, AI and pricing pages, Prophia home and pricing pages, and the pricing pages for ChatGPT Business, Claude Team, Microsoft 365 Copilot Business and Zapier, with Microsoft's Copilot data and privacy page, all read on 4 November 2026.