Part 4 of 6 · The one-person finance team
What's the bare minimum finance stack a scaling startup needs?
The frameworks assumed a finance department. I was one person, so I fitted the principles to how the company ran.
In this part
- Textbook control frameworks assume a finance team. With one person, fix the principles and fit the mechanics to how the company works.
- The tools had three requirements: one platform for income and one for expenses, mature integration with the ledger, and software-driven controls with people on the exceptions.
- What not to buy yet was as deliberate as what to buy.
New to the series? It follows one finance lead rebuilding a startup's finance function with Claude. Start with Part 1.
The same week the accountants started on FY25, I opened a new thread with Claude. The ledger had been chosen, and now it was time to select the next layer of the functional stack.
A general ledger is the foundation. But for a startup planning to multiply its headcount and revenue several times over, it needed systems that could scale: part standardization, part workflow-automation.
The next rung was revenue and cost management: a system of record for what we bill, collect and recognize, and a system for controlling what we spend and how.
I asked Claude to:
Refer to a canonical, well reputed source on finance function requirements.
It came back with a framework I remembered from university, COSO's model for internal controls, and something newer: a framework tailored to Singapore-incorporated companies, covering GST, ACRA and CPF. Both assume a finance team to command, with separation of duties, layered approvals and a controller at the center.
But it was just me, and Hupo's growth team spent much of its time on the road with clients. Import the textbook wholesale, and finance becomes a blocker for the very people generating revenue. Import too little, and Series B diligence finds a company without visibility or controls.
So I flipped the question: what is the bare minimum a company like this needs? Is there software that handles over half of it? Can it handle ten times the volume, users and contracts?
1. Start from how the company actually works
To answer that, I needed to see how Hupo really ran. Coffee with the new RevOps lead led to a chat with the Director of Growth, which became an on-the-fly discovery session with the CSAM Director. Their questions were practical and urgent:
- Has the invoice I sent [client] been paid yet?
- How does my team check the commission they've earned?
- Can I have budget for [insert problem]?
The CEO wanted monthly cash forecasts for a lean plan and an aggressive one, how much contracted revenue was live, and net revenue retention for the past year.
Spending told the same story. Several people traveled across the region to meet clients, and every budget request went by Slack to the chief of staff.
Meanwhile, Claude interviewed me about how each process worked today, from billing and collections to spending, payroll and reporting. I gave it documentation on Xero and Claude's connector, and a blog on AI-augmented spend management platforms like Airwallex and Revolut. Claude, in turn, asked whether I'd considered B2B SaaS revenue management software. I hadn't, so it wrote me a brief on the new breed of companies: Measure, Sequence, Tabs and others.
The principles in the frameworks still mattered, but I fitted the mechanics to how the company did things on the ground:
| First principle | How the company operated | What I set up |
|---|---|---|
| Control spending before it happens | People on the road meeting clients, every budget request sent by Slack to the chief of staff | A budget per team, so only exceptions came to me; a card per budget owner and a virtual card per budget line, each with a monthly limit and mapped to a Xero account |
| The person who prepares a payment isn't the one who releases it | Too few people for a second approval on every payment | Built into a monthly working-capital approval: I estimate the month's spend, and the CEO releases it into a spend wallet holding one month of working capital |
| One source of truth for revenue | Contracts in folders, invoices from Stripe and Canva templates, and the commercial team asking whether their client had paid | Measure as the system of record for contracts and invoices, synced to Xero, starting with invoicing |
| People can see what they've earned | The sales team asking how to check their commission | Commission tracking in Measure, next to the contracts it's earned on |
From those, the tools had three requirements:
- One platform of record wherever possible: one for income, one for expenses.
- Mature integration with the general ledger, in our case Xero.
- Software-driven controls, with people stepping in only on exceptions: an AI agent applying the spend policy, say, instead of a chain of approval emails.
The gap between today and that stack was widest in two places: revenue, and spending.
2. Revenue: one tool, end to end
The brief for revenue was short. Become the system of record for contracts, automate invoices and reminders, and route payments into the right bank.
Claude started with ten vendors. For each one, it sent an agent through the vendor's own documentation, then filtered on those requirements: does it talk natively to our CRM (Attio) and to Xero, and can it run without anyone re-keying a contract? Our sales operations lead and I narrowed the list to two, Sequence and Measure, and Claude wrote each demo agenda around a few "kill questions" that would rule a vendor out on the spot.
| Considered | Why it didn't make it |
|---|---|
| Stripe Billing | A percentage fee, even on invoices paid by bank transfer; it can't read contracts |
| Orb | Built for usage-first billing; no quoting, and no native CRM or Xero link |
| HubSpot Revenue Hub | Meant replacing the CRM |
| Airwallex billing | Newly launched, with no revenue recognition yet |
| Sequence | Strong, but reporting and commissions needed an add-on, and its pricing scaled with revenue |
Measure won on coverage. Quoting, contracts, invoicing and commission tracking sit in one tool, connected natively to Attio and Xero. And the pricing was a flat platform fee: no percentage of revenue, no features held back for a higher tier.
It wasn't a free choice. In the sales team's review, Sequence looked strongest as a system that might carry us three or four years, while Measure perhaps only two.
Invoicing was the easy part. The rest of revenue would take longer.
3. Spending: put the control in the card
For costs, the criteria followed the third requirement: cards with limits set in software, expense capture and approvals in one place, multiple currencies, and somewhere to earn yield on idle dollars.
An expense-only tool would have done part of it. But the company already had an Airwallex account, dormant for a year and a half, that did all of it. I opted to consolidate. Wise stayed what it is: a payment rail, not a bank.
How the cards held up in practice is Part 6.
4. Decide what waits
The list of what we didn't buy was longer than what we did:
- A board metrics tool: later, once the numbers underneath were stable.
- A Singapore payroll tool and e-invoicing: later, when volume justified them.
- A usage-first billing engine: only if usage-based revenue came to dominate.
- A second bank: when there was a reason to borrow, not before.
Each one would wait for the pain to arrive.
Throughout, Claude worked as a systems analyst: it held the requirements and sent agents through vendor documentation in parallel.
Where it went wrong: The mistake was mine, not Claude's. I ran this research in Cowork. In Claude Code, the context and the vendor research would have lived as files in the repo, organized and easy to revisit. In Cowork they lived in a long chat, so when the market moved, as when a vendor shipped a feature we'd marked as missing, updating the analysis meant reconstructing it. It was the same lesson as Part 2, and it cost more this time: ledgers stay put, but markets move.
What I'd do differently
Not much. I'd have preferred to have the early team I'd asked for running day-to-day operations while I implemented the systems. Instead, I juggled both, albeit poorly.
Takeaway: Fix the principles, and fit the mechanics to how the company actually works.
The tools were chosen. But before anyone got a card, every team would have to justify its budget, from zero.