The AI-ready finance function: a maturity model for SMEs
Five levels, from paper and spreadsheets to a finance function where AI does the reading and people make the decisions. How to tell where your business stands, and the order in which to move up.
By Qëndrim Uka
Founder and Managing Partner
- Published
- Reading time
- 11 min read
Almost every small and mid-sized business we speak to wants to use AI in finance. Fewer can say what it would change in their month, and fewer still have the foundations it needs. AI does not fix a finance function that runs on email attachments and spreadsheets. It amplifies whatever is already there: clean, structured, timely data becomes faster and more useful, and messy data becomes messy faster. This report sets out a five-level maturity model we use with clients to decide what to do first.
Why a maturity model
Finance teams rarely fail at AI because the technology is weak. They fail because they try to automate a step that has not yet been standardised. A model cannot read invoices reliably if they arrive through five channels, and it cannot reconcile accounts if half the transactions are booked weeks late. The levels below describe the foundations each step depends on, so that investment goes where it pays back.
The five levels
| Level | Documents and data | Close and reconciliation | Reporting | Role of people |
|---|---|---|---|---|
| 1. Manual | Paper, email attachments, messaging apps | Typed by hand, reconciled when there is time | Built by hand in spreadsheets, often late | Collecting and typing |
| 2. Digitised | Scanned or PDF documents, cloud accounting software | Bank feeds exist, matching is still manual | Standard reports from the software | Typing and checking |
| 3. Connected | One channel for all documents, consistent naming and periods | Fixed close calendar, reconciliations every month | A monthly pack with the same structure every time | Reviewing and reconciling |
| 4. Automated | AI reads documents, rules check them, exceptions are flagged | Matching and routine entries proposed automatically, approved by an accountant | Dashboards refresh as the books update | Handling exceptions and approving |
| 5. Intelligent | Structured history used for forecasting and anomaly detection | Continuous close, issues found during the month | Assistants answer questions on the company's own data, with sources | Advising and deciding |
Level 1: Manual
Documents arrive however the sender prefers. Someone collects them, types them into the books and reconciles the bank when there is time. The month closes late, and the numbers are trusted mostly because nobody has time to question them. The priority here is not AI. It is getting every document into digital form and every transaction into the books in the right month.
Level 2: Digitised
The business uses accounting software, often in the cloud, and documents exist as PDFs or scans. Bank feeds may already be connected. But entry and matching are still manual, and every person files documents their own way. Most SMEs we meet are here. The next step is standardisation, not new software.
Level 3: Connected
All documents arrive through one channel, such as a client portal, and are linked to the right period. The chart of accounts is stable. There is a close calendar with named owners and fixed dates, anchored in Kosovo to the statutory rhythm: wage tax and pension by the 15th, VAT by the 20th. Reconciliations are done every month, not at year end. This is the level at which AI becomes worth deploying, because the inputs are now predictable.
Level 4: Automated
Documents are read automatically and checked against rules: arithmetic, known VAT rates (18%, 8% or zero for domestic invoices), duplicates and periods. Low-confidence fields and anything unusual are flagged for a person. Bank transactions are matched to invoices automatically, and an accountant approves the proposals rather than creating every entry. Management dashboards update as the books do. People spend their time on exceptions, judgement and approval.
Level 5: Intelligent
With clean, structured history, the finance function can look forward: cash forecasts, alerts when a cost line moves unexpectedly, and assistants that answer questions such as what the VAT position will be on the 20th, citing the documents and entries they used. The close becomes continuous because issues are found during the month. People spend most of their time advising the business. Very few SMEs need to be here in every area; it is worth aiming for where it matters most to the business.
How to tell where you stand
- How many channels do documents arrive through, and does anyone know the full list?
- On which working day of the following month are the books closed, and is it the same every month?
- Are bank accounts reconciled monthly, and is the reconciliation reviewed by a second person?
- How much time each month goes into typing information that already exists in a document?
- When management asks a question about the numbers, is the answer found in a report or rebuilt in a spreadsheet?
Answer honestly and score each area separately. Most businesses sit at different levels in different areas, for example level 3 for payroll and level 1 for purchase invoices.
Move up one level at a time
The most common mistake is to jump from level 1 or 2 straight to level 4 by buying a tool. Automation on top of inconsistent inputs produces confident errors at speed. The order matters: standardise the inputs, fix the calendar, reconcile every month, then automate, and only then build intelligence on the history that automation creates.
A 90-day plan to reach level 3
- Days 1 to 30: choose one channel for all documents, agree a close calendar with named owners, and clean up the chart of accounts.
- Days 31 to 60: connect bank feeds where possible, reconcile every account monthly, and introduce a standard monthly reporting pack.
- Days 61 to 90: pilot AI document reading on one document type, such as purchase invoices, with a person reviewing every result, and measure how often corrections are needed.
Governance does not wait for level 5
From the first automated step, a few rules should be fixed. Nothing is posted to the books without a person's approval. Every automated entry keeps a link to its source document. Confidence scores and exceptions are visible, not hidden. Financial documents are stored privately, with access limited to the people who need it. And the people who approve understand what the system does, so that they can explain any number to an auditor or to the tax administration.
What to measure
| Measure | Why it matters |
|---|---|
| Working days to close | Shows whether the calendar holds and how fresh the numbers are |
| Share of documents processed without correction | Shows whether automation can be trusted with more |
| Filings made on time | The minimum standard every level must keep |
| Hours spent on manual entry | The capacity freed for review and advice |
An AI-ready finance function is not one that has bought AI. It is one whose data is clean enough that AI can be trusted with the typing.
About the author
Qëndrim Uka
Founder and Managing Partner
Qëndrim Uka founded Ura Advisory after a career that has covered every side of the tax and finance table. He began as a Tax Inspector in the Tax Audit department of the Kosovo Tax Administration, then worked as a Senior Tax and Finance Consultant at EY Kosovo in accounting and global compliance reporting, before serving as CFO and Senior HR Officer of an international technology company and advising businesses as a management consultant. He is an accounting graduate of the University of Prishtina.