Why did burn rise in September?
Explains what moved, line by line: which costs rose, which receipts slipped and which budget lines ran over, each with its transactions.
Numbers from budgeting and forecasting software
AI CFO software · CX Cash
What does a $2M SAFE buy us, and what does it cost us? AI CFO software that shows its working
CX Cash's AI CFO answers questions about your company's numbers in plain English. The model reads the question and writes the reply; every figure in it comes from the same calculations as your forecast, plan, metrics and cap table, and links to the records behind it. It reads your books and never changes them.
Private beta. Free during the beta.
AI CFO Mon 12 Oct 2026
What does a $2M SAFE buy us, and what does it cost us?
A $2,000,000 SAFE closing in January buys about 11.6 months4: cash lasts until March 2029 instead of April 2028, on $3,100,000 of cash1 and net burn of $172,000 a month2. At a $16,000,000 post-money cap it converts to 12.5% of the company, so the founders' fully diluted stake falls from 66.7% to 58.3%3.
Assumes burn holds at its July to September average and no other money comes in before the SAFE converts. Change either and I'll rerun it.
Whether 11.6 months of runway is worth 8.3 points of the company, and whether to raise on a SAFE or wait for a priced round, is your call. Want me to draft a note to your lawyer with these terms?
The working $0.00 computed by CX Cash · plain text written by the model
- 1
Cash $3,100,000
Sum of 3 account balances at the 30 September month end
Mercury and Chase, read through Plaid · 3 balances →
- 2
Net burn $172,000 a month
Average of $168,400, $171,900, $175,700 (July to September)
QuickBooks Online · 1,214 ledger lines →
- 3
Founders' stake 66.7% → 58.3%
The SAFE converts to $2,000,000 ÷ $16,000,000 post-money cap = 12.5%, taken from every holder in proportion
Cap table imported from Carta, with the round scenario · 3 holder classes →
- 4
Runway 18.0 → 29.7 months
Cash run forward month by month at net burn, with the SAFE's $2,000,000 in January
Runway model · monthly projection →
Source: a hypothetical seed-stage company asking on Mon 12 Oct 2026; illustration with sample figures. The SAFE converts as a post-money SAFE does on our cap table software page. Select a citation number to light its working.
AI CFO software answers the questions a founder would put to a finance chief (can we afford this hire, why did burn rise, what does this round do to us) from the company's own books, bank, billing, payroll and cap table. Language models are good at reading those questions and writing the reply. They are unreliable with figures, and a founder who repeats a wrong figure to a board owns it.
So CX Cash splits the work. The model reads the question, chooses the calculations it needs and writes the answer; CX Cash's own calculations produce every figure in it, and each figure links to its records. Below, the answer above is taken apart in five steps, from the question to the one call the AI CFO leaves to the founder.
Step 01The question
It reads the question the way your finance lead would
"Buy" and "cost" are not numbers. Before anything is computed, the AI CFO decides what each means for this company: the months of runway the cash adds, and the share of the company the founders give up for it. It takes the definitions agreed at setup (runway on net burn over the trailing three months; ownership fully diluted, the count investors price on) and the round scenario saved in the cap table: a post-money SAFE at a $16,000,000 cap, with the cash arriving in January 2027.
Whatever the question leaves open goes into the answer as a stated assumption: burn holds at its July to September average, and no other money comes in before the SAFE converts. Had the cap table held no scenario for the round, the AI CFO would have asked for the cap before answering, because the cap decides the cost.
| In the question | Taken to mean | From |
|---|---|---|
| "buy us" | The months of runway the cash adds, and the month cash runs out | The runway definition set at setup: net burn, trailing three months |
| "a $2M SAFE" | A post-money SAFE for $2,000,000 at a $16,000,000 cap, the cash arriving in January 2027 | The round scenario saved in your cap table |
| "cost us" | The founders' share of the company given up, fully diluted | The ownership definition set at setup: fully diluted, as investors price it |
| Not in the question | Burn holds at its July to September average; no other money comes in before the SAFE converts | Stated in the answer, so you can change it |
Source: hypothetical company; definitions as agreed at its setup call.
In CX Cash
The AI CFO starts from your definitions, not a generic one.
- Your definitions: runway, burn, ownership and the rest are set with you at setup, so an answer uses the figure your board already sees.
- Stated assumptions: whatever the question leaves open appears in the answer, ready to change.
- Questions before guesses: when a figure the answer depends on is missing, it asks for it first.
The answer starts from what the question means for your company, and says what it assumed.
Step 02What it reads
It reads your systems and never writes to them
For this answer the AI CFO read three month-end balances through Plaid, 1,214 ledger lines in QuickBooks Online and the cap table imported from Carta, with the round scenario saved in it. The same connections serve every question: accounting, banks, billing, payroll, spend, CRM, cap table exports and spreadsheets.
Every connection is read-only. The AI CFO cannot move money, pay a bill, post a journal entry or recategorise a transaction, in any system. That is a design choice. CX Cash is analysis software, with no cards, bill pay or expense workflow behind it, so an answer can never set off an operation. Several AI finance tools are built the other way: Fuelfinance says its agents draft invoices, journal entries and payment runs and post them once a person approves (Fuelfinance), and Puzzle's agents categorise, reconcile and prepare the books (Puzzle).
| Source | Read for this answer | Access |
|---|---|---|
| Mercury · Operating | Balance at 30 Sep 2026: $1,840,000 | Read-only, through Plaid |
| Mercury · Treasury | Balance at 30 Sep 2026: $1,000,000 | Read-only, through Plaid |
| Chase · Payroll | Balance at 30 Sep 2026: $260,000 | Read-only, through Plaid |
| QuickBooks Online | 1,214 ledger lines, 1 July to 30 September 2026 | Read-only |
| Cap table (Carta export) | Founders 8,000,000; Seed investors (preferred) 3,000,000; Option pool, granted and unallocated 1,000,000; the round scenario | Imported; Carta itself is never written to |
| Written to any of them | Nothing | |
Source: hypothetical company; balances at the 30 September 2026 month end.
In CX Cash
One read-only view of the systems your numbers live in.
- Works with QuickBooks Online, Xero and NetSuite; banks through Plaid; Stripe and Chargebee; Gusto, Rippling and Deel; Ramp; HubSpot and Salesforce; Carta exports; Excel and Google Sheets.
- Read-only by design: connections ask for read access only, and you can revoke any of them at any time.
- US dollars: the AI CFO works on books kept in USD.
Nothing an answer does can change your books.
Step 03Who computes
The model writes the answer; the software computes every figure in it
Language models are better at choosing a calculation than at doing one. On Vals AI's Finance Agent benchmark, 927 expert-reviewed questions on company filings, the best model scored 65.40% as of 7 October 2026, and the best score on financial modelling questions was 34.5%; an answer with any wrong load-bearing number gets no credit, and the strongest models lean on a calculator tool (Vals AI, 2026). When researchers changed only the numbers in school maths problems, every model they tested did worse (Mirzadeh et al., 2025). OpenAI's own researchers argue that models guess because training and evaluation "reward guessing over acknowledging uncertainty" (Kalai et al., 2025).
So the AI CFO never asks a model for a number. The model reads the question and chooses the calculations; CX Cash runs them, with the same code as its forecast, plan, metrics and cap table, and the model places each result in the answer. The 12.5% is the cap table's own calculation: a post-money SAFE converts to its amount divided by its post-money cap, and every other holder shares the dilution in proportion, so the founders' 66.7% becomes 58.3%.
| Task | Done by | In this answer |
|---|---|---|
| Read the question | The model | "Buy" read as runway, "cost" as the founders' fully diluted stake |
| Choose the calculations | The model | Cash position, net burn, runway with the SAFE's cash, SAFE conversion |
| Compute every figure | CX Cash's calculations | $3,100,000 · $172,000 · 18.0 → 29.7 months · 66.7% → 58.3% |
| Apply the instrument's terms | CX Cash's calculations | Post-money SAFE: $2,000,000 ÷ $16,000,000 = 12.5%, taken from every existing holder in proportion |
| Write the answer around the figures | The model | The sentences, with each figure placed as computed and cited |
Source: hypothetical company; the SAFE arithmetic as on our cap table software page.
In CX Cash
One set of calculations behind every screen and every answer.
- The same figure everywhere: the runway in an answer is the runway on the forecast screen and in the board pack, to the dollar.
- Terms applied as written: SAFE caps and discounts, payroll tax rates and wage bases live in the calculations, not in a model's memory.
- Your choice of model: CX Cash works with several AI models and you choose which one writes your answers. The figures come from the same calculations whichever you pick.
Change the model and the words change; the numbers stay the same.
Step 04Where each figure leads
Every figure links to the records behind it
A founder who repeats a figure owns it, whoever produced it. When Air Canada argued that its website chatbot was responsible for its own answers, the tribunal called the argument "a remarkable submission" and held the airline liable (Moffatt v. Air Canada, 2024). Deloitte refunded more than A$97,000 to the Australian government over a report with AI-generated errors (CFO Dive, 2025). In KPMG's 2026 survey of 1,013 senior finance leaders, 48% named the accuracy of AI-generated financial outputs as a top concern (KPMG, 2026).
So each figure in an answer carries a citation. Select the net burn of $172,000 and the report behind it opens: $168,400 in July, $171,900 in August and $175,700 in September. Select a month and its ledger lines open, 1,214 of them for the quarter, by account and vendor. A founder can check a number before repeating it, and an accountant or an investor in diligence can follow the same trail.
In the answer
Net burn $172,000 a month 2
Report it opens
- July 2026$168,400
- August 2026$171,900
- September 2026$175,700
- Average$172,000
Records behind it
1,214 ledger lines in QuickBooks Online, 1 July to 30 September 2026, by month, account and vendor
Source: hypothetical company; QuickBooks Online ledger, 1 July to 30 September 2026.
In CX Cash
Every figure arrives with its trail.
- Numbered citations: each figure in an answer names its source system and the report it comes from.
- Down to the record: from the figure to the monthly report to the transaction, invoice, bill or pay run.
- The same trail for everyone: you, your accountant and an investor in diligence see the records the answer used.
A number you are about to repeat to your board is two clicks from its source.
Step 05What it leaves to you
It knows which calls are not its to make
Some questions have no figure for an answer: how much of a renewal belongs in this quarter, whether to reserve against a doubtful invoice, how an equity grant is taxed. Audit firms told PCAOB staff that a person who uses a generative AI tool "is still responsible for the results" (PCAOB, 2024), and OpenAI's GPT-5 system card notes that earlier models, missing an input, would "hallucinate answers or claim to have solved" the task (OpenAI, 2025).
When a question needs that kind of judgment, or the data to answer it is missing, the AI CFO says so, sets out what the decision depends on, and drafts an email to the person who can make it. The SAFE answer ends the same way: the figures are the AI CFO's, and whether 11.6 months of runway is worth 8.3 points of the company is the founder's call.
Can we recognise all of the Acme renewal this quarter?
That depends on the contract's terms, which I can't see: what Acme bought, when each part is delivered, and how the $84,000 is allocated between them. How much belongs in this quarter is a judgment for your accountant. Here is a note you can send, with the invoice attached.
ToYour accountant
SubjectAcme renewal: how much revenue in Q4?
Acme signed a 12-month renewal on 2 October for $84,000, invoiced in full (Stripe invoice attached). How much should we recognise in Q4, and do you need the signed contract to decide? We close the quarter on 31 December.
Source: hypothetical company and renewal.
In CX Cash
A clear line between what it computes and what you decide.
- It says when it cannot answer: missing data or a judgment call is named as such, never filled with a guess.
- It drafts the hand-off: an email to your accountant, lawyer, board or fractional CFO, with the figures and records attached.
- Decisions stay yours: hiring, raising and recognising revenue are left to you and your advisers.
The AI CFO does the arithmetic and the drafting; people make the calls.
Ask your own question about your own numbers.
Private beta. Free during the beta.
Method
The calculations behind every answer
The AI CFO uses the same formulas as the rest of CX Cash. These are the ones the sample answer used, with the page that defines each.
| Figure | How CX Cash computes it | Notes |
|---|---|---|
| Cash position | sum of account balances on the date asked | $1,840,000 + $1,000,000 + $260,000 = $3,100,000 at 30 September. |
| Net burn | cash out − cash in, for the month | July to September: $168,400, $171,900 and $175,700. |
| Trailing burn | average of the last 3 months' net burn | $516,000 ÷ 3 = $172,000. |
| Runway | months until cash, run forward at net burn, reaches zero | 18.0 months today; 29.7 with the SAFE's cash in January. |
| Zero-cash month | the month in which projected cash reaches zero | April 2028 today; March 2029 with the SAFE. |
| Post-money SAFE ownership | amount ÷ post-money valuation cap | $2,000,000 ÷ $16,000,000 = 12.5%. |
| Founders' fully diluted stake | founder shares ÷ fully diluted shares, then × (1 − SAFE ownership) | 8,000,000 ÷ 12,000,000 = 66.7%; × 87.5% = 58.3%. |
Source: formulas as used across CX Cash; worked figures from the hypothetical company above.
Beyond one question
Seven more jobs, with the same working
Each answer draws on the calculations behind one of CX Cash's feature pages, linked below, so its figures match the screens they come from.
What drove MRR this month?
Splits the change into new, expansion, contraction and churned MRR, and names the accounts behind each movement.
Numbers from SaaS metrics software
What if we hire three engineers in Q2 and close the round in June?
Builds the scenario from the sentence. The hires, their start dates and the round become inputs you can see and edit, and the scenario engine computes the result.
Numbers from budgeting and forecasting software
Tell me when something changes.
Speaks up on its own: when the forecast's lowest week moves, when spend runs above its usual range, or when the same charge appears twice.
Numbers from cash flow forecasting software Weekly brief by email or Slack: coming soon
Draft September's investor update.
Drafts the monthly investor update or the commentary for the board pack from the month's figures, for you to edit and send.
Numbers from SaaS metrics software
What would a Series A at a $32M pre-money leave the founders?
Models the round from its terms: the option pool top-up, the SAFEs converting and every holder's stake after, with the arithmetic written out.
Numbers from cap table software
Which portfolio companies have less than 12 months of runway?
For investors: answers across the numbers each portfolio company reports and the fund's own performance, and drafts the LP letter.
Numbers from VC portfolio management software
For investors
For investors, one question across the whole portfolio
The AI CFO is built first for founders, and it answers for the investors who back them too: across the numbers each portfolio company reports, and across the fund's own capital calls, distributions and marks. The figures come from the same fund calculations as our VC portfolio management software, with TVPI, DPI and IRR defined on their own pages.
- Which companies reported less than 12 months of runway last quarter?
- What is Fund II's net TVPI, and what moved it this quarter?
- Draft the Q3 letter to our LPs from the fund's figures.
Buyer's guide
How to judge AI CFO software: six tests for any demo
A demo shows the answers a vendor chose. These six tests show how a product behaves on your numbers, and each takes about a minute. CX Cash is not the only product built to pass some of them: Cube and Fuelfinance also say their figures come from calculations rather than a model (Cube, Fuelfinance). The comparison below shows where the products differ: what each covers, what it can change and what it costs.
- 01
Ask it a number you already know.
Last month's net burn, or cash at the month end. The answer should match your books to the dollar. A system that predicts figures can be close and wrong; one that calculates them is exact, or visibly broken.
- 02
Select a figure in the answer.
It should open the report and the records behind it. A figure you cannot trace is a figure you cannot check before you repeat it.
- 03
Ask something it cannot know.
How much of a renewal to recognise this quarter, or what next year's tax bill will be. A good answer says what it would need and who decides; a poor one fills the gap with a confident guess.
- 04
Ask what it can change.
Analysis software should read your systems and never write to them. An assistant built into a bookkeeping or spend platform can act on your books or your money, which is a different product with different risks.
- 05
Ask where your data goes.
Whether it trains models, where it is processed and how long chat history is kept. The main model providers' business terms exclude training on customer data by default (OpenAI, Anthropic), while consumer chat plans may train on chats unless the setting is off (OpenAI). Where data is processed depends on how a vendor reaches its models: Anthropic's own API offers US or global processing, while Amazon Bedrock and Google Cloud offer EU regions (Anthropic).
- 06
Ask which model answers.
You should know which AI model writes your answers, and be able to choose. If the figures come from calculations, changing the model should not change them.
| Test | CX Cash AI CFOAnalysis only, on your connected data | A general chatbotChatGPT or Claude with a spreadsheet export | AI accounting softwareDigits, Puzzle | FP&A platform with AICube | Software plus a finance teamFuelfinance | A fractional CFOA part-time person |
|---|---|---|---|---|---|---|
| Where the figures come from | CX Cash's calculations, never the model | The model, or code it writes; both vendors tell users to check outputs | Their own ledgers; Digits says to confirm details in the ledger | Governed data, traced to the general ledger | Deterministic steps run as code; its AI CFO is in beta | The CFO's own model |
| Trail from figure to record | Each figure to its report and its records | Back to the file you uploaded, at most | Puzzle shows the source of each proposed entry | Every insight to a general ledger transaction | Findings point to transaction rows | As far as they document it |
| What it can change | Nothing: read-only, analysis only | Nothing in your systems | Your books: they categorise, reconcile and close; Digits also pays bills | Consolidation and close work; writes plans back | Drafts invoices, journal entries and payment runs, posted after approval | Whatever access you give |
| What it covers | Cash, budgets, SaaS metrics, the cap table, and fund metrics for investors | Whatever you paste in | Bookkeeping and the close; Digits adds bill pay and invoicing | Planning, forecasting and reporting | FP&A, with a finance manager or CFO available | Strategy, fundraising and the board |
| Built for | Founders, and the investors who back them | Anyone | Puzzle: pre-seed to Series B; Digits: accounting firms and businesses | Companies with finance or FP&A teams | Startups and small businesses | Companies not ready for a full-time CFO |
| Price | Free during the private beta | Free tier with limits, and paid plans | Digits from $65 a month; Puzzle $30 to $360 a month | Quote only | Tailored, not published | About $3,000 to $15,000 a month |
Source: each vendor's own pages, read on 12 October 2026 and listed under Sources; every description of a product is its vendor's. Fractional CFO rates are vendor and marketplace figures, not an independent survey.
AI CFO or a person
AI CFO or fractional CFO: which jobs go where
A fractional CFO is a person, part-time, with judgment, relationships and a signature. Published rates run from about $3,000 to $15,000 a month on retainer, most often $5,000 to $7,500, or $175 to $450 an hour, on vendor and marketplace figures rather than an independent survey (Focused.Financial, 2026). Bessemer notes that software companies typically hire a full-time CFO only past $50 million of ARR (Bessemer Venture Partners). The AI CFO answers every day from the books, and the person makes the calls; the AI CFO's hand-off emails are written for that person. When to hire a fractional CFO covers the timing.
| The job | AI CFO | A CFO or fractional CFO |
|---|---|---|
| Answer a question about the numbers the day it comes up | In minutes, with every figure cited | When they next have hours on your account |
| Explain what moved this month | Line by line, with the transactions | With context from inside the business |
| Build a scenario | From a sentence, computed by the scenario engine | With judgment about which scenarios matter |
| Draft the investor update or board commentary | A first draft from the month's figures | Edits it and stands behind it |
| Revenue timing, reserves and tax positions | Names the judgment and drafts the email | Decides, with your accountant |
| Negotiate a round, a loan or a contract | Prepares the figures | Leads it |
| Sign or certify the numbers | Never | Yes, with your accountant or auditor |
Data and models
Your data, and your choice of model
The AI CFO reads more of your company's data than any other part of CX Cash, so the rules for that data are set out here plainly.
Never used to train AI models
Your questions, the answers and your company's data are never used to train AI models, ours or any provider's.
Processing region
Processing in the EU is possible on request.
Chat history you control
You set how many days chat history is kept, in the app's settings.
Read-only connections
Every connection reads and none writes. Revoke any of them at any time.
Your choice of model
CX Cash works with several AI models, and you choose which one answers. The figures come from the same calculations whichever you pick.
Who it suits, and when to use something else
Built for
- ✓Founders from first revenue to Series B who want answers about their own numbers between board meetings, with the working attached.
- ✓Finance leads who field the CEO's questions and would rather check a cited first draft than start from a blank page.
- ✓Investors who want one place to ask about their portfolio companies' numbers and their fund's performance.
Not the right fit if
- ✕You need the books kept. The AI CFO reads the ledger and never writes to it; bookkeeping and the month-end close belong to your accountant or to AI accounting software such as Digits or Puzzle.
- ✕You need cards, bill pay or spend controls. CX Cash is analysis only; spend platforms such as Ramp and Brex run those operations.
- ✕Your books are not kept in US dollars. The AI CFO works on USD books for now.
- ✕You want someone to make the call. The AI CFO computes, explains and drafts; decisions and signatures stay with you, your CFO and your accountant.
Free template
Investor Update Template + Cap Table & Dilution Calculator
A one-page monthly update template in six blocks (PDF) and a cap table sheet: enter the holders and a new round, and the price per share, new shares and each holder's dilution compute for you.
The preview is the file itself: live formulas and worked sample numbers, yours to keep whether or not you use CX Cash.
Private beta · Free during the beta
Get started with CX Cash
Start onboarding in the app. We set up each account by hand, one at a time. Free during the beta.
- Done-for-you setup. We connect your systems and build your first view with you on a 45-minute call. You do not map accounts or build a model.
- First results in 2 business days. Your first forecast, budget view or metrics pack is ready within 2 business days of the setup call.
- Spreadsheets welcome. Anything without a connection comes in from an Excel or CSV export, and your existing model can be the starting point.
Questions
Is the AI CFO available now? +
Yes, in CX Cash's private beta: it is the chat panel in the app. Start onboarding from this page. We set up each account by hand, one at a time.
What does it cost? +
CX Cash, the AI CFO included, is free during the beta. Pricing for after the beta has not been published yet.
What is AI CFO software? +
Software that answers the questions a founder would ask a finance chief, such as whether a hire is affordable, why burn rose or what a round does to the founders, from the company's own books, bank, billing, payroll and cap table. CX Cash's version computes every figure with its own calculations and links each one to the records behind it.
Can an AI CFO replace a CFO? +
No. It answers questions about the numbers, explains what moved and drafts the documents a CFO would draft. It does not make judgment calls, negotiate or sign anything. For many early companies it covers the questions that come up between a fractional CFO's hours.
Does the AI CFO ever make up a number? +
No. The model reads the question, chooses the calculations and writes the answer; every figure in it is computed by CX Cash's calculations and cites the records it came from. When a figure cannot be computed from your data, the answer says so.
Which AI model does it use? +
CX Cash works with several AI models, and you choose which one writes your answers. The figures come from CX Cash's calculations whichever model you choose.
Is my data used to train AI models? +
No. Your data is never used to train AI models. You set how many days chat history is kept, and processing in the EU is possible on request.
Can it change anything in QuickBooks, Xero or my bank? +
No. Every connection is read-only: it cannot move money, pay bills, post journal entries or recategorise transactions.
How is it different from AI accounting or FP&A software? +
AI accounting tools such as Digits and Puzzle keep the books, and FP&A platforms such as Cube also run consolidation and close work. CX Cash's AI CFO only reads your systems. It answers across cash, budgets, SaaS metrics, the cap table and, for investors, fund metrics, and it is free during the beta.
Does it work for venture funds? +
Yes. It is built first for founders, and investors can ask across the numbers their portfolio companies report and their fund's performance, including TVPI, DPI and IRR, and have it draft the LP letter.
Does it send alerts? +
It flags changes such as the forecast's lowest week moving, spend running above its usual range and charges that appear twice. A weekly brief by email or Slack is coming soon.
How long does setup take? +
On a 45-minute call we connect your systems with you, read-only, and agree the definitions your answers will use. Your first answers follow within 2 business days.
Sources
- Anthropic (2025). Commercial Terms of Service.
- Anthropic (undated). Data residency (Claude Platform documentation).
- Bessemer Venture Partners (2023). How to hire a CFO and build a finance team.
- CFO Dive (Sadovi, M. W.) (2025). Deloitte refunds over $60K for report with AI errors, Australian government says.
- Civil Resolution Tribunal of British Columbia (2024). Moffatt v. Air Canada, 2024 BCCRT 149.
- Cube (undated). AI at Cube.
- Cube (undated). The agentic FP&A platform.
- Cube (undated). Pricing.
- Digits (undated). Ask Digits (Help Center).
- Digits (undated). Pricing.
- Focused.Financial (2026). Fractional CFO rate report.
- Fuelfinance (undated). Fuel AI CFO.
- Fuelfinance (undated). Platform.
- Fuelfinance (undated). Pricing.
- Kalai, A. T., Nachum, O., Vempala, S. S. and Zhang, E. (2025). Why language models hallucinate (arXiv 2509.04664).
- KPMG LLP (2026). AI in finance 2026.
- Mirzadeh, I. et al. (2025). GSM-Symbolic: understanding the limitations of mathematical reasoning in large language models (arXiv 2410.05229).
- OpenAI (2025). GPT-5 system card.
- OpenAI (2026). Enterprise privacy.
- OpenAI Help Center (2026). File uploads FAQ.
- OpenAI Help Center (2026). How your data is used to improve model performance.
- Public Company Accounting Oversight Board (staff) (2024). Spotlight: observations on the use of generative AI in audits and financial reporting.
- Puzzle (undated). AI Suite.
- Puzzle (2026). Puzzle launches AI Suite.
- Puzzle (undated). Pricing.
- Vals AI (2026). Finance Agent benchmark (v2).
The company, its cash, its burn and its round are hypothetical, and figures without a citation are CX Cash's arithmetic on them. Other products are described from their own pages, read on 12 October 2026; their claims about their products are theirs, and prices may have changed since. Vals AI's leaderboard changes as models are added; its figures are as dated. Fractional CFO rates are vendor and marketplace figures, not an independent survey. This page is not legal, tax or investment advice.
Go deeper on FP&A, Reporting & Finance Systems
- FP&A, Reporting & Finance Systems
AI in Finance Pays Off Where a Person Can Check the Work
- FP&A, Reporting & Finance Systems
When Is a Fractional CFO Cheaper Than a Full-Time Hire?
- Startup Fundraising, Runway & Equity
Investor Update Template: 10 Lines to Send Every Month