Too many options, no clear answer
You know finance admin is eating time. You have searched for tools and found hundreds: invoice capture apps, expense apps, reconciliation tools, workflow platforms, and now a wave of products describing themselves as AI agents for finance. Each demo looks good. None of them obviously fits the way your business actually works, and you are wary of buying three subscriptions that do not talk to each other.
The question you actually want answered is simpler: for the specific jobs that take up our time, what is the right kind of tool?
Why the choice is confusing
Most tool marketing describes capabilities, not fit. Every product can 'automate invoices'. What differs is where each one stops: which systems it connects to, which exceptions it handles, whether it understands your coding rules, and what happens when it is unsure.
AI agents add another layer of confusion. The term covers everything from a chatbot with access to your ledger to a system that reads documents, makes decisions and posts transactions. For finance, the useful question is not whether something is an agent, but which decisions it makes and how those decisions are checked.
What choosing wrong costs
- Subscriptions for tools that each solve part of a process, with a person still carrying data between them.
- Automations built in Zapier that nobody understands once the person who built them leaves.
- AI tools given write access to the ledger without clear limits, producing errors that are hard to trace.
- A large custom build for a problem that a setting in Xero would have solved.
The last one matters. Part of choosing well is recognising when the answer is to use what you already have properly.
How we match tools to processes
When we look at your finance processes, we sort each one into a tier and recommend the lightest option that will actually work.
| Kind of tool | Good for | Where it stops |
|---|---|---|
| Built-in features of Xero, QuickBooks or Sage | Bank rules, recurring invoices, basic reminders | Anything needing judgement or other systems |
| Marketplace apps (capture, expenses, payments) | One well-defined job, done well | Your specific rules, and hand-offs between apps |
| Zapier, Make or n8n | Moving data between systems on a trigger | Complex logic, error handling, auditability at volume |
| AI models or agents | Reading documents and emails, classifying, drafting, proposing matches | Final decisions on money without checks |
| Custom build | Processes across several systems with your own rules | Simple problems a tool already solves |
For the processes that need more than an off-the-shelf app, what we build usually combines these layers:
- Connectors to your ledger, bank, CRM, job system and inboxes through their APIs.
- AI steps for the parts that involve unstructured input, such as reading an invoice, a remittance or a customer email, using models from OpenAI or Anthropic Claude with outputs forced into a fixed structure.
- Deterministic rules for the parts that must be exact: VAT arithmetic, tolerances, approval limits.
- An exception queue where people handle anything the system is not confident about, with the reason shown.
- Logging of every automated decision so your accountant or auditor can see what happened and why.
AI agents in this design are given narrow tasks and limited permissions. An agent can propose that a payment matches three invoices. Whether it is allowed to post that match without a person depends on rules you set.
What you end up with
A clear map of your finance processes, each with a tool choice you understand and a reason for it. Some processes will be solved by configuring software you already own. Some by a single app. The ones that remain, usually the messiest and most time-consuming, get a build that fits how your business works, with people kept in the loop where it matters.
Where you might be now
- You have trialled several finance tools and none quite fitted.
- You pay for apps that still need someone to move data between them.
- You have Zapier automations nobody fully understands.
- You are interested in AI for finance but unsure what is safe to automate.
- You want an outside view of what is worth building and what is not.