// USE CASE · FINANCE

AI finance agents that close the booksand explain every variance.

Most finance teams spend the first ten days of every month chasing reconciliations, then explain a missed forecast in the next board pack with no real attribution. Melaya lets you build a finance crew that reconciles ledgers, attributes variance line by line, monitors treasury concentration, and drafts the board narrative. The agents handle the legwork, controllers approve every posting, and the audit log keeps the receipts.

01
// What breaks today

Manual workflows cost more than the agent does.

Three pains every sales and BD team hits weekly. Each one is what your reps actually complain about, not what a feature page would call them.

  1. 01

    Month-end close drags into week three because controllers chase twenty-plus reconciliations across NetSuite, Stripe, the bank, and exchange statements by hand.

  2. 02

    Board pack lands with an 11 percent forecast miss and no root cause until the next cycle, so the same gap repeats next quarter.

  3. 03

    Audit prep eats a controller's December because no one tracked which NAV breaks were closed and which still sit open at year-end.

02
// Pipelines you can build

Agent workflows: compose, approve, replay.

Every pipeline below is a shape you wire on the canvas using the crew and tools further down. Not a feature we ship for you, a pattern you configure.

P01

Run the monthly close on a calendar

FinancialController reconciles bank, Stripe, ledger, and exchange positions every morning, with a HITL gate on any break above one thousand dollars. The CFO sees a break list, not a backlog.

P02

Attribute forecast variance line by line

FPnLAnalyst decomposes actual versus plan into volume, mix, price, fees, and carry, flags any line over fifteen percent off forecast, and writes the root cause. Cross-run memory tracks which drivers repeat month over month.

P03

Watch treasury runway and concentration

TreasuryManager scores cash by bank, custodian, and stablecoin issuer, flags any single-venue concentration over thirty percent, and projects months of runway. Scoped tool reads only the treasury database, never the ledger.

P04

Draft the board pack with cited numbers

ChiefFinancialSynth assembles a one-page Financial Health Score from each persona's section, cites the source agent on every number, and stages the deck for review. HITL gate blocks export until the CFO signs off.

P05

Build the year-end audit binder

ComplianceOfficer and FinancialController bundle journal entries, trade confirmations, and reconciliation evidence into a single audit folder. Replay reproduces every NAV break and the resolution, so the auditor walks a timestamped trail.

P06

Send the LP and investor letter

InvestorRelations writes the one-page monthly letter with benchmark in sentence one and attribution in three bullets. HITL on benchmark numbers and on the send, so no draft leaves the building without sign-off.

03
// The multi-agent crew

Finance & CFO office crew

Real personas from the finance_team crew. Each ships with a tuned system prompt and a default tool allowlist. Swap models per persona on the canvas.

Financial Controller

FinancialController

Owns NAV, daily reconciliation, and audit readiness. Flags every variance over one hundred dollars with a documented root cause inside twenty-four hours.

FP and L Analyst

FPnLAnalyst

Decomposes results into drivers (gross, transaction costs, funding, slippage, carry) and explains any line that drifts more than fifteen percent off plan.

Treasury Manager

TreasuryManager

Tracks cash across banks, custodians, and stablecoin issuers. Enforces concentration limits and keeps six months of operating expense in liquid reserves.

Risk Manager

RiskManager

Calculates VaR and CVaR, runs stress scenarios, and enforces non-negotiable circuit breakers on rolling drawdown.

Compliance Officer

ComplianceOfficer

Watches trading and treasury activity for wash patterns, velocity anomalies, and audit-trail gaps. Tracks license and filing deadlines per jurisdiction.

$

Tax Strategist

TaxStrategist

Models tax position under FIFO, HIFO, and Specific ID, surfaces harvesting opportunities before year-end, and maps jurisdiction exposure for each entity.

Investor Relations

InvestorRelations

Writes the one-page LP letter with benchmarks in line one. States underperformance plainly and benchmarks against the buy-and-hold baseline every time.

$

Chief Financial Officer

CFO

Owns capital allocation, budget review, and a ninety-day rolling cash flow forecast. Approves every deployment over one hundred thousand dollars.

Chief Financial Synthesis

ChiefFinancialSynth

Synthesizes every workstream into a board-ready one-pager with a Financial Health Score and a single strategic action.

04
// Scoped tools

Tool allowlists: only the actions you grant.

Every tool below is a real shared tool from the Melaya bundle. Allowlist per agent; HITL-gate the writes; revoke any of them in one click.

shared/tools/finance/

Pull equities, FX, and macro context for variance attribution and treasury hedging notes. Read-only, no HITL gate needed because nothing writes.

alphavantage_stock_pricealphavantage_stock_historyalphavantage_stock_indicatorsfx_ratealphavantage_macro_dataalphavantage_sector_performance
shared/tools/treasury_tools/

Reach US Treasury fiscal data for runway and cash positioning models. Read-only by design.

treasury_datasettreasury_exchange_ratestreasury_debt_to_penny
shared/tools/database/

Read from the ledger, billing, and treasury databases for close, variance, and audit work. sql_query is scoped per agent and DSN-gated, never granted to drafting personas.

sql_querysql_schemasql_export_csv
shared/tools/msoffice/

Read trial balances and budget workbooks, write the close pack, and produce the board deck. Every write to a shared path is HITL-gated by default.

excel_read_sheetexcel_write_dataexcel_createword_createpptx_create
shared/tools/knowledge/

Build the per-workflow knowledge store from chart of accounts, policy memos, prior board packs, and auditor PBC lists. Powers all three knowledge layers.

build_knowledgebuild_knowledge_from_textbuild_knowledge_from_file
shared/tools/email/

Read auditor and lender threads, prepare LP letters and bank confirmations, and stage every send for the CFO. gmail_send is HITL by default for this crew.

gmail_sendgmail_readgmail_list_labels
shared/tools/core/

General research, structured fetch, and Slack drops for the Monday close digest. pause_for_human is the explicit HITL primitive any workflow can call.

web_searchhttp_requestslack_post_textcalculatepause_for_human
05
// Three knowledge layers

The crew reads what you give it.

Every pipeline ships with three layers of knowledge access. Mix and match per agent on the canvas. No shared vector space with another tenant, no surprise reads, no opaque retrieval.

L1

Static context

includeContext

Per-pipeline documents appended to specific agents' input on every run. The ICP brief, playbook, pricing sheet, or won-deal email corpus. Whatever needs to be there before the agent thinks. You pick which personas get which docs.

L2

RAG retrieval tool

rag_retrieve

A scoped tool granted per-agent. When the agent decides it needs more depth, it queries the workflow's vector store on demand. Same knowledge base as Static context, accessed only when the model asks for it.

L3

Cross-run memory

pipeline_memory

Pipeline-level state that carries from one run to the next. Yesterday's research is in scope for today's follow-up. The crew remembers what it already prospected, what got approved, what was sent. The audit log is the second-order knowledge base.

07
// FAQ

AI agent questions we get every week.

Will the agents post journal entries on their own?

No, not by default. The finance crew ships with HITL on every ledger write, every send, and every export. The agents stage the journal entry; a controller approves the posting. You can lift the gate per template once you trust the pipeline.

Can the agents reason over our ledger, bank, and Stripe data?

Live, via scoped connector tools. Stripe is stripe_get_charges, stripe_get_subscriptions, stripe_create_refund. Ledger and bank statements ride the Postgres or MySQL connectors against your warehouse, read-only by default. HITL gates every write call (refund, journal post, transfer). On top of that, the per-workflow knowledge store covers chart of accounts, close calendar, policy memos, and prior board packs through three layers: static context, the rag_retrieve tool, and cross-run memory so yesterday's open breaks are in scope for today's close.

Does this replace Anaplan, Pigment, Mosaic, or Causal?

No, it sits next to them. Melaya is the reasoning and drafting layer; you keep your planning tool as the model of record. Anaplan, Pigment, Mosaic, or Causal hold the cube, the agents reach in through scoped SQL or CSV exports and write the variance narrative on top.

How do we keep the board pack from sounding like AI?

ChiefFinancialSynth cites the source agent on every number, including the section and the run id. Reviewers can require a citation on every line as a HITL pre-check. Wording reuses your prior board packs loaded into the knowledge store.

Which models can we run this crew on?

Any. Claude on FPnLAnalyst and ChiefFinancialSynth where reasoning quality earns the cost, GPT on InvestorRelations for the LP letter, a local Ollama on ComplianceOfficer when the audit data cannot leave your network. Each agent picks its own.

How fast can finance get the first pipeline running?

With the ledger and treasury DBs connected and a Gmail or Slack target authorized, the close-pack workflow is a four-node canvas: read trial balance, reconcile breaks, draft variance section, request approval. Most teams ship it inside a working session and have the first reviewed close pack the same day.

How does this handle audit evidence and SOX controls?

Every run logs every read, every tool call, every approver, and every approval reason. Replay reproduces any close, any NAV calculation, and any variance attribution from the source data. The audit log is the control evidence, exportable as a binder.

Can I audit exactly what the agent did and why?

Yes. Open any run, see every step in order with the prompt, the tool call, the model output, and the approver decision. Replay it on the same inputs and get the same trace. Auditors get a timestamped trail per posting, per filing, per send.

Does this work for crypto funds and traditional finance teams?

Both. The crew is institutional finance with crypto fluency baked in (exchange NAV, on-chain treasury, MiCA, IRC 1256). For a SaaS or services team, the same personas reconcile Stripe, NetSuite, and bank statements, attribute SaaS variance, and draft the lender or board pack. The skills generalize, the tools change.

Why use Melaya instead of n8n or Zapier for the monthly close?

n8n, Zapier, and Make execute predefined trigger-action steps, so the close stalls the moment a reconciliation breaks off script. Melaya, the visual AI agent builder, runs autonomous multi-agent workflows: the Financial Controller persona investigates the break through the database bundle, and every posting waits for human approval. Zapier and Make win on connector breadth for simple linear automations.

How do we measure whether the finance crew is worth running?

Melaya reports cost per accepted result, so the CFO sees what each approved close pack or board deck actually costs. Full run traces record every step, model call, and typed failure reason, and deterministic-first evaluation runs rule-based checks, such as totals that tie, before any model-graded review. Nothing counts as done until a controller accepts the output.

Can the treasury workflow read a banking app that has no API?

Yes. Melaya's Device Control operates a real Android phone: the agent opens the allowed banking app, reads the screen, and returns balances to the Treasury Manager persona for concentration checks. The phone taps and types only inside allowed apps and pauses for on-device approval before anything publishes. API-backed sources still run through the treasury_tools and database bundles.

Can our ledger data stay on our own infrastructure?

Yes. Melaya runs in the cloud or on your local runner, so sql_query calls from the database bundle execute on a machine you control and only approved outputs leave it. Per-step model routing across 23 providers lets sensitive close steps use a local model while drafting personas such as Investor Relations route to a cloud model.

What happens when the crew finds a variance it cannot explain?

The FP and L Analyst persona decomposes the line into gross, transaction costs, funding, slippage, and carry, then flags any residual instead of inventing a cause. The workflow calls pause_for_human from the core bundle, so a controller reviews the open item before the board pack drafts. The run trace records what was checked and why the item stayed open.

Build finance & cfo office pipelines on Melaya.

Sandbox tier is free with no card. Join the waitlist and we will email you the moment a slot opens.

← Back to every use case
Join the community
// Cookies
Melaya uses a small set of first-party cookies that are strictly necessary to authenticate you, maintain your session, and protect the platform from abuse. We do not use advertising cookies, cross-site trackers, or third-party analytics by default. The full cookie list is in our Privacy Policy.