The Treasury Mentor · Node Zero

Build your own
AI Finance Desk

Node Zero is 378 finance tasks across 8 departments, built the way a treasurer or controller actually works, not a prompt library. Each one reads your real policies, covenant definitions, and chart of accounts before it does anything, so the output is built on your numbers and your rules.

Most AI in finance is a chatbot bolted onto the same broken process. You still gather the inputs, still remember the covenant math, still catch what is missing. Node Zero doesn't guess. If a number isn't in your context base, it says so instead of inventing one.

It is not a replacement for what you run today. It is a layer on top of it. It works against the systems and structure you already have, so it scales with you: add an entity, a covenant, a reporting line, and every task touching it inherits the update.

 
tasks
8
departments
 
branches
17
context files
One square below is one task. All of them read the same 17 context files before they run.

Already doing the work, not just reading about it.

Kaneo Capital
Cuddlo Clothing Co.
Why this one is different

Built by someone who has to sign it.

There is a lot of AI content aimed at finance right now. Almost none of it was written by anyone who has had to defend a number to a lender, a board, or an auditor. That is the whole difference here.

01 · Who built it

A finance executive, not a content account

Over eighteen years in senior finance roles, built in rooms with leaders who came out of Tesla, thyssenkrupp, Ford, General Motors and Blue Cross Blue Shield. That is the standard these were written to. Every task exists because the work behind it landed on my desk first.

02 · What it is built on

Real deliverables, not imagined ones

Built from the outputs a finance function genuinely produces. Close packages and provision workpapers. Budgets, reforecasts and long range plans. Pricing and margin analysis. M&A screening, valuation and diligence support. Board and strategy material. Cash and liquidity forecasting. Structured the way each one is really assembled.

03 · Where the logic comes from

Public standards, practice, and a lot of miles

Published accounting and tax standards, established treasury and FP&A practice, and hands on work across North America, Europe, Latin America and Asia, where the same close has to satisfy four sets of local expectations at once. Nothing proprietary sits in these files. No employer data, no client data, nothing confidential.

These are not skills an influencer posted last week after reading a changelog. Each one was written with the logic finance actually needs, which is why they read your rules before they run and stop when something is missing.

click to go deeper · click the core to zoom back out
Every task, by name

Not categories. The actual work.

Nobody sells you a list this specific, because nobody has built it. This is a live slice of the library. Hover over it to hold it still.

Governance, not vibes

It reads the rules first. Then it refuses to guess.

This is the part that separates a finance agent from a chatbot. In treasury and accounting, being confidently wrong has consequences. A covenant computed from a textbook definition instead of your credit agreement can show a pass that is not there.

Step 01 · Read

Governing rules load before anything runs

Covenant definitions captured clause by clause from your agreement. Approval thresholds. Investment policy limits. Your KPI builds, not generic ones. The task reads them as constraints, not as suggestions.

›
Step 02 · Check

A missing input is named, never filled

The prime directive of the whole system is that it never assumes an input. If a figure is not in your base, the task names it, tells you which file it belongs in, and points at the task that can produce it, rather than handing back a clean looking number nobody can trace.

›
Step 03 · Write back

Output becomes context for the next run

Work products land in your workpapers and update the current state, so the next task starts from what the last one produced. Conflicts between files get surfaced rather than quietly resolved.

Sample behaviour · Quarterly covenant testIllustrative figures
Reads first Debt & Covenants Reporting Conventions Historical Baseline Entities
INTEREST COVERAGE RATIO
EBITDA ÷ interest expense
4.81xProduced
Covenant minimum 3.00x · 60% headroom
0.00xcovenant floor 3.00x6.00x
LTM EBITDA128.4
LTM interest expense26.7
Add-backs applied7 of 7
DEBT SERVICE COVERAGE RATIO
CFADS ÷ total debt service
2.91xProduced
Covenant minimum 1.25x · 133% headroom
0.00xcovenant floor 1.25x4.00x
CFADS112.6
Interest component26.7
Scheduled principal, next 4 quarters12.0
Total debt service38.7
Before the base existed

Debt service coverage was the one nobody wanted. The amortisation profile lived in a lender spreadsheet, the EBITDA build lived in someone's head, and the ratio got reassembled from scratch every quarter. Two analysts, three days, and a number that changed depending on who built it.

›
After the base existed

The amortisation schedule and the add-back caps were captured once, during the interview. Now both ratios compute from the same source in one pass, they agree with each other because they read the same figures, and the workpaper shows every input it used.

Both ratios are produced here because everything they need was captured. That is the point of the base. If something is missing, the task says so and names the file it belongs in, rather than assuming a straight line amortisation and handing back a ratio that looks fine. A number you cannot trace is worse than no number at all, and in a covenant test it is the difference between a pass and a default.
Where the line sits

The assembly gets automated. The judgment stays with you.

Break any recurring deliverable into the steps it actually takes. Most of them are assembly: pulling, matching, computing, formatting. A few are judgment. Node Zero takes the first kind and hands you the second, faster and with the work already shown.

The task runs it
Stays with you

Quarterly covenant package

8 steps · 2 stay with you
TaskRead facility terms
TaskBuild EBITDA to the clause
TaskApply add-back caps
TaskCompute the ratios
TaskTest forward step-downs
YouJudge the ambiguous clause
TaskDraft the certificate
YouSign it

13-week cash forecast

7 steps · 2 stay with you
TaskPull and reconcile balances
TaskWeekly IS per entity
TaskAR / AP / inventory rollforward
TaskBuild the 13 weeks
TaskConsolidate and translate FX
YouChallenge the assumptions
YouOwn it in the room

Month-end close package

7 steps · 2 stay with you
TaskRun the readiness check
TaskDraft and trend accruals
TaskMatch JE support
TaskDetect the flux
YouExplain the real driver
TaskAssemble the package
YouApprove and release

Lender reporting

6 steps · 2 stay with you
TaskTrack what's owed and when
TaskPull the reporting figures
TaskPopulate the forms
TaskTie out to the statements
YouReview the exceptions
YouSubmit to the agent

Every dark block is a judgment call. That was always the real job. The rest was overhead you were carrying because nobody had taken it off you.

The latest branch, inside Treasury

Capturing the future of finance. Digital assets, built in.

Most finance AI libraries stop at the tasks everyone already knows how to do. Node Zero doesn't. Digital assets are already a working branch here, not a slide for later: 14 tasks covering the full lifecycle, from first policy draft to incident response, built the same way as everything else in the library, reading your real rules instead of guessing at them.

3tasks
Policy & strategy
Digital Asset Treasury Policy Builder Stablecoin Settlement Business Case Builder 24/7 Settlement Liquidity Policy Designer
3tasks
Due diligence & custody
Stablecoin Issuer and Reserve Due Diligence Digital Asset Custodian Selection and Due Diligence Tokenized Money Market Fund Eligibility Assessor
4tasks
Controls & reconciliation
On-Chain Payment Control Designer Wallet Address Sanctions Screening Procedure On-Chain to Subledger to GL Reconciliation Designer Tokenized Deposit Evaluator
4tasks
Risk & readiness
Cross-Border Digital Asset Corridor Readiness Screener Digital Asset Pilot Designer and Kill-Criteria Manager Digital Asset Counterparty and Concentration Risk Assessor Digital Asset Incident Response Playbook
And what it reports into

The tasks run. The dashboards show it.

Every task writes back, so the picture stays current instead of being rebuilt for the meeting. Three views a CFO actually asks for, and they tie to each other, because they read the same base.

Liquidity & 13-Week Cash
Consolidated · all entities · FX at policy rate · week ending 12 Sep
Refreshed weekly
Opening cash
48.2
reconciled to bank
Week-13 close
41.7
−6.5 over horizon
Trough balance
37.9
week 9
Revolver drawn
25.0
of 80.0 committed
Total liquidity
96.7
cash + undrawn
Min-cash cushion
7.9
above 30.0 policy
Closing cash by week
$m · weeks 1 to 13 · trough flagged
1
2
3
4
5
6
7
8
37.99
10
11
12
13
Where the liquidity sits
$m · at week 13
Cash on hand41.7
Undrawn revolver55.0
Minimum cash policy(30.0)
Trapped / restricted(6.4)
Freely deployable60.3
Working Capital
LTM revenue 1,445.4 · LTM COGS 1,264.0 · days on a 365-day basis
Refreshed monthly
DSO
54.2
+2.1 vs prior
DIO
48.6
+0.7 vs prior
DPO
61.8
+3.4 vs prior
Cash conv. cycle
41.0
−0.6 vs prior
Net working capital
168.9
AR + inv − AP
NWC / revenue
11.7%
−0.1pt
Cash conversion cycle, how it builds
days · DSO + DIO − DPO
54.2
DSO
+
48.6
DIO
−
61.8
DPO
=
41.0
CCC
The cycle came down 0.6 days, but not for a good reason. Receivables slipped 2.1 days and inventory added 0.7, so the operating side got worse by 2.8. The entire improvement came from stretching payables 3.4 days. The commentary says exactly that, because the drivers are computed rather than eyeballed.
Balance sheet components
$m · period end
Trade receivables214.6
Inventory168.3
Trade payables(214.0)
Net working capital168.9
Overdue > 60 days18.7
Debt & Covenant Position
Total debt 247.0 · cash 41.7 · LTM EBITDA 128.4 · tested quarterly
Refreshed quarterly
Net debt
205.3
247.0 gross − cash
Net leverage
1.60x
vs 3.50x max
Interest cover
4.81x
vs 3.00x min
DSCR
2.91x
vs 1.25x min
Undrawn
55.0
revolver
WA cost of debt
6.4%
blended
Maturity wall
$m · contractual repayments by year · total 247.0
12.02026
18.02027
25.02028
192.02029
78% of the stack lands in a single year. That's the refinancing conversation, and it's visible three years early.
Covenant headroom
against tested thresholds
Net leverage · 1.60x of 3.50x54% clear
Interest cover · 4.81x of 3.00x60% above
DSCR · 2.91x of 1.25x133% above
Next step-down · Q2 20273.25x
All three ratios read the same EBITDA build, the same net debt figure and the same amortisation schedule, so they agree with each other by construction. Three covenants that disagree is a reconciliation problem you should never have to have.

Illustrative figures. Internally consistent, but not a real company.

Start here · everything else depends on it

Built on your inputs, tailored to how you work.

378 tasks are worth very little if they have to guess at how you work. So Node Zero is not the tasks. Node Zero is the base underneath them, and building it is the first thing that happens. Five guided modules, 25 files, one field at a time, nothing invented on your behalf. Nothing is uploaded anywhere and nothing is shared. The answers become files that sit on your side, and you decide how much detail goes into them.

It asks. You answer.

One field per turn, in plain language, in whatever order suits you. Where a field does not apply you mark it N/A deliberately, so a later task knows it was considered rather than skipped. Stop mid-way and pick it up next week; it resumes where you left off and shows you what's still open.

Answers become files, immediately.

Every confirmed answer is written into the relevant context file as you go: your entities, chart of accounts, banking, facilities, policies, close calendar, current state. You can open any of them in a text editor and read exactly what it knows about you. No black box, and no database somewhere else holding it.

Then every task reads it.

From that point the base is the thing that makes the library yours rather than generic. The test that it worked is not a score. It is this: a finance professional who has never seen your operation could read that folder and act correctly for you tomorrow.

01 Identity
02 Systems & Data
03 Financial Structure
04 Policy
05 State
Node Zero intake · state & history · building the cash forecast base
Node ZeroFor the 13-week forecast I need to know how your receivables actually convert, not a blended DSO. Do you have a collection curve by entity? Roughly what share lands inside 30 days, then 31 to 60, then 61 to 90?
YouUS, Germany and Mexico all behave about the same. Call it 62% inside 30 days, 28% in the 31 to 60 bucket, 8% in 61 to 90. Romania is much slower and I would have to pull the detail.
Node ZeroCaptured for US, Germany and Mexico. That adds to 98%, so I am carrying the remaining 2% into 90 plus unless you tell me otherwise. Romania stays open. The forecast will still run, and it will show Romania as unmodelled rather than applying the US curve to it.
Written to Historical Baseline · AR collection curves · 3 entities set · 1 open · US curve NOT applied to Romania
YouThat's right. Romania usually runs past 90 days but I don't want to guess the split.
Node ZeroThen I will leave it open rather than write "90 plus" as though it were measured. It goes on the open items list, so it surfaces every time you run the forecast until you close it.
The questions people actually ask

Where it runs, and who sees your numbers.

Does this only work with Claude?

No. The library is plain text. There is no software to install and nothing to lock you in. It runs wherever your AI can read a folder of files you hand it, which today means Claude, ChatGPT, Gemini or Copilot in the workspace tiers your company has already approved. Claude is what I build and test on, so that is the smoothest path, but the files are not tied to it.

Does the model learn from my data?

That is governed by your contract with your AI provider, not by these files. On business, enterprise and API tiers the major providers exclude customer content from training. Anthropic's commercial terms say so explicitly, and OpenAI and Google draw the same line. On personal consumer plans it is a setting you control. Worth a five minute check with your IT team before you load anything real.

What do you see?

Nothing. There is no platform here, no account, and no server of mine anywhere in the path. You get files, they sit in your environment, and your numbers never leave it. That is a deliberate choice. A finance professional should not have to hand their numbers to a vendor to get value out of them.

Most people run the interview themselves, and it is built to be self serve. If you would rather not, I run it with you. A working session where I ask the questions and your base gets built properly the first time, by someone who has filled one of these in anger.

What you actually get

Everything in the box.

No tiers to decode and no feature grid. One library, one base, one set of guardrails, and a person on the other end of it.

Day 1
378agents
Every agent, on day one

No phased rollout. All 378 agents across 8 departments and 60 branches are active in your workspace as soon as you're in.

17context files
The base underneath it

Built with you through a five module guided intake. Every agent reads it before it runs, which is what makes the output yours rather than generic.

Every agent
Governance and data hygiene, by default

Every agent reads the rules first, never assumes an input, and writes the result back. Data hygiene isn't a setting you turn on, it's how the base is built.

Ongoing
Monthly
A monthly masterclass

A live session on what's actually changing in treasury, tax, and finance AI, and how to use it. A working session, not a webinar funnel.

Monthly
New agents, built from real work

New agents added from continuous training on real engagements and real-life workflows, not a fixed catalog that stops growing the day you buy it.

Direct line
A direct line to me

Standards changes tracked, questions answered by a finance executive, not a support queue.

Treasury 2.0: Future-Proofing Finance with AI
The book, included

Treasury 2.0: Future-Proofing Finance with AI

Bojan Belejkovski

The execution playbook this whole library grew out of, written before any of it existed. It is built on the D.A.S.H. framework, and every member gets it in full.

D Digitize A Automate S Streamline H Harness AI
After you have it

You get the library. Then what?

A file library that never changes goes stale the first time a standard moves or your structure does. Here's the rhythm after day one.

Week one
First close
Every month
Ongoing
Run the interviewFive modules, at your pace
Node Zero populated17 context files, gaps visible
Pick three tasksStart where the pain is
Close packageAssembled against your calendar
Flux commentaryYour thresholds, your voice
Workpapers written backNext run starts warmer
Library updateWhat changed, and why
New tasks addedBuilt from real engagements
Standards trackedAccounting and tax changes
Your base stays yoursIt never leaves your environment
Structure changes absorbedNew entity, new facility, new covenant
Direct line to meFinance executive, not a support queue

Two ways in.

The founding cohort is full. The next one opens soon, and the custom build is open right now.

The Treasury Mentor Network

Founding cohort closed.

The first 100 seats filled. The next cohort opens soon, and I'll reach out directly when it does.

  • 378 agents across 8 departments, active from day one
  • Your own context base, built through a guided intake
  • Monthly masterclass, new agents added from real engagements
Join the waitlist
No spam, no drip sequence, just a note when the next cohort opens.
Done for you

Custom Implementation

For teams who want this handled end to end. I audit how your finance function actually runs today, then build the agents and dashboards directly into your own systems, on the same governance guardrails as everything else here.

  • Full audit of how your finance function runs today, before anything gets built
  • Bespoke agents built directly in your stack, not templates adapted after the fact
  • Live dashboards and a command center view built around what you actually track
  • Direct, ongoing access to me as the build evolves
Book a strategy call
Best for teams who want full delegation, not a self-serve library.