A governed AI accountant for NetSuite. Connected to NetSuite, email, Slack and your document stores. Designed so a conservative AI policy can say yes.
It drafts journals, bills, rec exceptions and controller packs from your email, Slack and document stores.
Consumer chat products are never the system of record.
September rent JE
Waiting for Approve / Edit / Reject
A competent NetSuite accountant costs $6,500–$10,000 a month loaded. A fractional specialist costs $3,000–$8,000. Close software still leaves the labour — and the AI-policy risk — with you.
Pointing a consumer chatbot at the general ledger is cheaper. It is also the reason your AI policy exists.
The agent prepares. A named human approves. NetSuite posts only after that approval. The same identity cannot draft and approve the same transaction.
It is not a Controller. It is not a CFO. It does not close the period, release a bank file, or sign a tax opinion.
Bills
Journals
Reconciliations
The rec starts from exceptions, not a blank sheet. We do not claim the bot closed the bank rec.
Prepare is about 80–90% of the human minutes. Decide is about 10–20%.
Human
24 min prepare + 3 min decide = 27 min person-time
Booksquire
Agent prepares. Human 3 min = 3 min person-time
RPA
No lease-file script. 27 min still human
Creating the journal is the job. Approving it is the signature.
| Work | Human | Booksquire | RPA |
|---|---|---|---|
| Watch finance@ and Slack all dayCoverage, not a 90/10 split. | Owns |
Owns — agent is the watch |
Cannot |
| Turn an invoice PDF into a coded draft bill | Prepares |
Prepares |
Rules if template, else fallback |
| Duplicate-bill and vendor match | Prepares |
Prepares |
Rules |
| Post the bill or journal in NetSuiteThe decide bar is short in every column. The saving already happened on the draft row above. | Decide only |
Cannot |
Rules |
| Propose bank-rec matches | Prepares |
Prepares |
Rules |
| Finish the rec and lock it | Owns |
Cannot |
Cannot |
| Draft “September rent” from the lease folder | Prepares |
Prepares |
Cannot |
| Flag dirty vendor / customer / item records | Prepares |
Prepares |
Rules |
| Weekly controller brief | Prepares |
Prepares |
Cannot |
| Payment run / bank-file release | Owns |
Cannot |
Rules / should not |
| Tax opinion, payroll, statutory audit | Owns |
Cannot |
Cannot |
| Immutable audit of who approved what | Owns |
Owns |
Rules |
Illustrative mix for one mid-market NetSuite entity. Not a timesheet.
Every item below is a draft. A named human approves before NetSuite posts. The agent does not post.
Balanced draft from source documents or last month’s pattern — period open, debit equals credit. You approve, edit or reject.
A coded draft from the PDF or email: header and lines extracted, vendor or item matched, duplicate check run, chart of accounts applied. You confirm vendor, tax and coding. On Approve, NetSuite posts the bill under the AP-draft role.
Unmatched bank lines versus unmatched NetSuite transactions, with proposed matches, evidence, an exception list and a draft clearing journal. You accept matches and finish the rec in NetSuite.
A nightly draft scan for duplicates, missing fields, bank-detail changes, and inactive records on open transactions. You approve any change — bank-detail changes always dual-control — and the scan does not post on its own.
A weekly draft brief and ad-hoc questions, every figure cited to a NetSuite report, saved search or record ID. Official numbers run through NetSuite reports; we will not ship a P&L that is valid SQL and wrong accounting.
The audit pack stores the source, the model version, the approver and the NetSuite internal ID. Every material action writes an exportable audit record of the draft and the named human’s decision.
Your data does not train Grok, Claude or OpenAI.
Production traffic uses enterprise APIs with Zero Data Retention — not consumer chat apps.
You choose the allowed models. The router fail-closes if a vendor is off the list.
Nothing posts without a named human approval.
NetSuite access is a custom least-privilege MCP role. Oracle already blocks Administrator from the Connector.
Email, Slack, SharePoint and Drive use delegated OAuth and allow-lists — not tenant-wide admin tokens.
Every material action writes an exportable audit record.
One kill-switch disables a model or freezes all writes.
Full wording lives in the Architecture & Trust Pack.
$1,250 / month
Read-only insights and exception flags. No posts. The safe start while Legal reviews.
$2,000 / month
Draft plus human-approved posts. One entity. Named reviewer included.
$3,500 / month
Full month-end pack, flux narratives, auditor PBC support.
Onboarding $3,500 one-time. Extra legal entity +$600 / month. The two-week sandbox is credited against the $3,500 onboarding fee if you convert.
| Included in the $2,000 Accountant tier | Excluded |
|---|---|
|
|
Default envelope: one legal entity. Up to approximately 500–1,000 operational tickets per month. Standard write workflows: journals, AP bills, AR invoices, rec exceptions, master-data flags. Month-end pack targeted for business-day 5 to 7, depending on client close calendar.
Custom MCP role. One mailbox or library. A controller pack and a rec exception list. Nothing writes.
Journals and bills appear as drafts. Your reviewer accepts or rejects. Nothing posts unless they click Approve.
Exit: CISO signs the eight commitments. Controller signs that drafts are usable. Then the $2,000 seat — or stay on Monitor.
Write to info@booksquire.co.
No. The agent prepares. A named human approves. NetSuite posts only after that approval. L2 is the approved payload only, and the identity that drafted cannot be the identity that approved.
No. Your data does not train Grok, Claude or OpenAI. Production traffic uses enterprise APIs with Zero Data Retention — not consumer chat apps.
NetSuite stays the book of record. The agent queries it live, rather than keeping bulk exports. Working files and what the agent is taught live in a folder you own in SharePoint or Google Drive. Standing rules live in your AI workspace project. Where your data lives.
Yes. You choose the allowed models. The router fail-closes if a vendor is off the list. One kill-switch disables a model or freezes all writes. You name the allowed models and pin the region. If residency cannot be verified, the call does not go out.
A custom least-privilege MCP role. Administrator cannot use the connector. Oracle already blocks that. Write tools refuse unless they carry an approval token issued by our control plane. The path is Oracle’s official AI Connector Service: Model Context Protocol, OAuth 2.0, user-delegated.
A named Booksquire reviewer reviews every engagement. A named reviewer on our side prepares; your controller still approves the post. The same identity cannot draft and approve the same transaction. Slack sends your named accountant a card: Approve, Edit or Reject.
Payment initiation or bank-file release, period close, tax opinion or statutory audit sign-off, payroll processing, deleting master data, Administrator-role access to NetSuite, any solely-automated decision with legal or substantial effect on a person (including credit-like vendor onboarding), unsupervised posting of journals, invoices, bills or master-data changes, consumer ChatGPT, Claude.ai or grok.com as the system of record, and training any model on your books. It is not a Controller. It is not a CFO. It does not close the period, release a bank file, or sign a tax opinion. What it does not do.
Week 1 is read-only. Custom MCP role. One mailbox or library. A controller pack and a rec exception list. Nothing writes. Week 2 is draft-only. Journals and bills appear as drafts. Your reviewer accepts or rejects. Nothing posts unless they click Approve. Exit: CISO signs the eight commitments. Controller signs that drafts are usable. Then the $2,000 seat — or stay on Monitor.
The Accountant seat: draft plus human-approved posts. One entity. Named reviewer included. Included in the $2,000 Accountant tier: always-on Accountant Agent on Slack and email; NetSuite MCP connection with custom least-privilege roles; inbox, Drive and SharePoint intake for invoices, statements and supporting documents; draft journals, invoice / bill coding and reconciliation exception queues; master-data review flags (vendors, items, customers); weekly insights brief and month-end exception pack; named human reviewer with dual-control on all posts; immutable audit log, exportable to the client; quarterly access and role review; client-selectable model policy (Grok / Claude / OpenAI / mix). Onboarding is $3,500 one-time. An extra legal entity is +$600 / month. The two-week sandbox is credited against the $3,500 onboarding fee if you convert.
A named Booksquire reviewer reviews every engagement. A named reviewer on our side prepares; your controller still approves the post.