From Paper to ERP—and Everything in Between: Where Can AI Help?

Artificial Intelligence. Super Intelligence. Generative AI. AI agents.

Whatever we eventually call it, the business question is much simpler: what can it actually do for you?

For a finance function, I would not start by asking which AI product to buy.

I would start with the business as it operates today.

Where is information still entered manually? Where do errors occur? Which reconciliations take hours? Which reports require repeated spreadsheet work? What questions can management not answer easily?

And, increasingly important:

What can your existing accounting system already do that you may not be using?

Only then should we ask where AI could improve accuracy, control, productivity or management information.

That applies whether a business still handles much of its paperwork manually, depends heavily on Excel, uses an older accounting package, operates modern cloud accounting, has several specialist systems or has invested in a full ERP.

AI can potentially help at every stage.

But what it should do is different at each one. And some improvements may not require AI at all.

At a Glance

Some of these opportunities may be addressed through existing software or conventional automation without introducing AI.

There is another important change across all six stages.

Traditionally, management receives reports that have been designed in advance.

AI may increasingly allow an authorised manager to ask a question that nobody thought to build into a standard report.

Instead of:

“Please prepare another report.”

the question becomes:

“Explain what is happening.”

That does not remove the need for reliable data, controls or structured financial reporting. It simply changes how management may be able to use the information already available.

1. Mostly paper and manual records

For some businesses, the largest opportunity is also the most basic: getting useful information off paper.

Supplier invoices, LPOs, goods-received notes, receipts and payment approvals may still pass physically between employees.

The problems are familiar:

• information is entered more than once;
• documents go missing;
• input errors occur;
• approvals are difficult to track;
• discrepancies are discovered late;
• finding an old transaction can consume considerable time.

AI-supported document processing can potentially read scanned documents, extract the relevant information and convert it into structured data.

It could compare an invoice with an LPO and GRN, identify missing documents, highlight apparent quantity or price differences and flag possible duplicates.

What could you ask?

• Which supplier invoices do not have supporting GRNs?
• Which invoices exceed their approved LPOs?
• Which transactions are waiting for approval?
• Have we received two invoices with the same number?
• What have we purchased from this supplier during the last three months?
• Which documents contain discrepancies requiring investigation?

Possible outcome

Paper becomes searchable and checkable information.

At this stage, AI does not have to produce sophisticated forecasts to create value.

Reducing document handling, re-entry, missing paperwork and basic errors may already make a substantial difference.

2. Manual processes supported by Excel

Excel remains an extremely useful finance tool.

The difficulty arises when important processes depend on repeated copying and pasting, multiple versions of worksheets, complex formulas or the knowledge of one employee who understands how everything fits together.

Finance staff can spend more time preparing information than analysing it.

AI may assist with consolidation, reconciliation, classification, identifying missing or duplicate transactions and highlighting unusual movements.

It may also help with some compliance-related checking.

For a Kenyan finance team, this can have direct financial consequences.

KRA’s VAT return system uses electronic invoice records, and eligible input VAT claims generally depend on valid, transmitted TIMS/eTIMS invoices or other permitted supporting records. Missing or invalid records can delay, or in some cases prevent, VAT recovery.

KRA also validates income and expenses declared in income tax returns against eTIMS, withholding-tax and customs information, subject to applicable exceptions.

Automated checks could help identify purchase invoices with missing or mismatched eTIMS records before returns are filed, giving the finance team an opportunity to investigate and resolve discrepancies.

This would not replace professional tax judgement or statutory responsibility. It could reduce manual checking and identify exceptions earlier.

What could you ask?

• Why are expenses above budget this month?
• Which costs increased most?
• Which customers caused the increase in receivables?
• Which products have declining margins?
• Which accounts remain unreconciled?
• What are the largest unexplained variances?
• Which transactions do not agree with our VAT or eTIMS records?
• What are the main cash pressures over the next four weeks?

Possible outcome

Finance moves from compiling information towards analysing it.

And management is less restricted to questions somebody anticipated when the spreadsheet was originally designed.

3. Basic or legacy accounting systems

An older accounting package is not necessarily a bad accounting package.

I experienced this personally with Pastel before it became part of Sage.

Pastel performed its accounting role satisfactorily. Around it we developed separate Excel/VBA applications for management information and controls, and used Excel and PowerPoint extensively for reporting.

We eventually moved to Oracle ERP — a major implementation involving considerable time, cost, process redesign, testing, training and management involvement.

With today’s technology, I would ask an additional question before assuming that a dependable accounting core needed replacing:

Which problems really require a new accounting system, and which could be solved around the existing one?

A company using an older version of QuickBooks, Pastel/Sage or another reliable package might potentially use AI to:

• check information before accounting entry;
• compare LPOs, GRNs and supplier invoices;
• reconcile accounting and stock records;
• identify unusual transactions;
• analyse exported ledger data;
• improve cash, margin and working-capital analysis.

What could you ask?

Instead of simply receiving an aged debtors report:

• Which customers are progressively slowing payment?
• Which overdue balances are becoming most significant?
• How much cash could be released if debtor days returned to last year’s level?

Instead of simply receiving a stock report:

• Which items have not moved for six months?
• Which products are accumulating stock while sales decline?
• Where is working capital being tied up unnecessarily?

Instead of merely looking at the P&L:

• Why did gross margin fall?
• Was it selling price, purchasing cost, volume or product mix?
• Which products or customers contributed most?

Possible outcome

Better management intelligence without automatically discarding a dependable accounting ledger.

The inefficiency may be around the system rather than inside it.

The same principle is explored in Don’t Invent It. Technify It.: sometimes the opportunity lies in applying new technology to an existing business process rather than reinventing the process itself.

4. Modern cloud accounting

Businesses using modern cloud accounting should start with a different question:

What can the system already do?

Current accounting products increasingly contain automation and AI of their own.

Xero’s Just Ask Xero (JAX) and QuickBooks’ Intuit Intelligence are examples. The capabilities available depend on the product, subscription, market and user permissions.

That matters because the cheapest and least disruptive improvement may already be included in the software licence.

The problem may simply be under-utilisation.

Employees continue with old manual routines. Reports are exported to Excel unnecessarily. New functionality introduced through software updates has never been configured or even investigated.

So the first opportunity is not necessarily to add another AI application.

It is to use the existing system properly.

Only after that should management identify the genuine gaps — perhaps deeper analysis, forecasting, combining operational and accounting information, or providing more flexible management enquiry.

What could you ask?

• What caused cash to fall this month?
• Which customers drove the revenue increase?
• Which costs moved unexpectedly?
• Which suppliers increased prices most?
• What are the three most important financial changes this month?
• Which accounts require investigation before month-end?
• What happens to cash if collections slow by ten days?

Possible outcome

More value from technology the business already owns before adding another system.

This should be a fundamental principle:

Before buying more AI, understand the automation and AI already available to you.

5. Accounting software plus specialist systems

This environment may be particularly familiar to established manufacturing, processing, FMCG, distribution and trading businesses.

Accounting sits in one system.

Inventory may be in another.

There may also be separate applications for payroll, purchasing, sales, production, warehousing or CRM — with Excel still filling some of the gaps.

Each system may work adequately.

The difficulty often lies between them.

Finance repeatedly exports information and reconciles different sources. Product codes may not match. Month-end relies on several departments supplying information.

Management can see accounting, stock and sales separately, but it is harder to understand how they interact.

AI may potentially help compare datasets, identify mismatches and analyse financial and operational information together.

What could you ask?

• Sales increased — why didn’t cash?
• Which high-revenue customers produce poor margins?
• Which products sell well but consume excessive working capital?
• Which purchase-price increases are eroding gross margin?
• Which stock shortages appear to have affected sales?
• Which branches have unusual stock adjustments?
• Which customers generate sales but poor cash conversion?
• Is overtime increasing without a corresponding rise in production?
• Where do warehouse quantities disagree with accounting stock?

Possible outcome

Management begins understanding relationships across the business rather than examining each system separately.

This is one situation where a focused AI layer may make particular sense.

Not because AI must replace the underlying systems, but because it may help extract more value from several systems that already work.

6. Full ERP

ERP does not remove the opportunity for AI.

It changes it.

The organisation may already have integrated finance, purchasing, sales, stock and operational information.

The problem may now be the sheer quantity of data available.

Hundreds of reports can exist. Managers still export information to Excel. Finance still spends substantial time explaining variances. Operational systems generate more alerts than managers can realistically investigate.

AI can potentially sit above the transactional strength of ERP and help users interrogate information, prioritise exceptions and explore underlying causes.

What could you ask?

• Why did operating profit fall despite higher sales?
• Which products caused the deterioration in margin?
• Which customers have become less profitable?
• Which purchase-price movements are having the greatest financial impact?
• Why is inventory increasing?
• Which suppliers are becoming less reliable?
• Where is working capital deteriorating?
• Which branches or departments require attention?
• Which operational exceptions are financially significant today?

And perhaps eventually the most useful question:

“What should I be looking at that I haven’t asked about?”

That is a very different role for AI.

Instead of merely producing information on request, it may help identify unusual relationships, exceptions or trends that management has not yet noticed.

Possible outcome

More intelligence from an ERP investment already made.

At this level, AI is less about replacing the system and more about helping managers make better use of the information it contains.

Reports will remain — but our relationship with them may change

Financial statements, reconciliations, control reports and regular management information will remain essential.

AI does not make accounting discipline unnecessary.

But not every business question fits neatly into a report designed months or years earlier.

If sales rise but cash falls, management may want to move immediately from revenue to receivables, inventory, purchasing and margins.

If margin falls, the next question may be about product mix, purchase prices or discounts.

AI potentially makes that investigation more conversational.

But there is an important limitation.

AI cannot analyse information that the business has never captured properly.

If stock movements are missing, documents are incomplete, accounting entries are wrong or controls are routinely bypassed, AI may help expose the weaknesses.

It cannot simply manufacture an accurate history.

Human review also remains essential for payments, accounting adjustments, tax treatment, policy decisions and material business decisions.

Before introducing AI tools, management should also establish what information may be shared externally, particularly supplier banking details, payroll and customer information.

There is an important distinction between a tool that proposes an accounting entry for approval and one authorised to post it automatically. Access controls, segregation of duties and audit trails must remain effective. Automation should not weaken accountability.

Start with the business problem

There is no single “AI solution for finance”.

A more practical approach is:

Current process → Pain point → Existing capability → AI opportunity → Business benefit → Recommendation

If I had to identify a starting point, I would generally favour finding exceptions that can be verified — unmatched invoices, unreconciled balances, duplicate transactions and unexplained differences.

These are tangible problems whose results can be checked against source documents.

But they do not necessarily require AI. Existing software and conventional rules-based automation may provide a simpler, more reliable and economical answer.

The more ambitious ability to ask complex business questions in natural language could be valuable. Its usefulness, however, depends on the quality, completeness and accessibility of the underlying information.

Before approving an investment, management should establish whether the proposed improvement:

• reduces processing time, errors or repetitive work;
• strengthens controls, information quality or decision-making;
• delivers benefits that justify implementation costs, recurring charges and staff training.

And the recommendation does not have to be “buy AI”.

It may be:

• use a feature already available in the accounting system;
• change an inefficient process;
• eliminate unnecessary manual work;
• introduce a focused AI tool;
• connect information from separate systems;
• or, where genuinely justified, replace an inadequate system.

The objective is not more AI.

It is a finance function that is more accurate, better controlled, more productive and more useful to management.

Could Your Business Get More from Its Information Systems?

An independent review of information systems and business processes can help management identify inefficiencies, improve information flows and assess opportunities to make better use of existing technology. The focus is on improving business operations and management decisions, rather than providing accounting or tax services.

The starting question is: “Which information or process problem is most limiting your business decisions today?”

Sources & References

• Kenya Revenue Authority — VAT Auto-Populated Return FAQs. Conditions for claiming input VAT and reconciling electronic invoices.
• Kenya Revenue Authority — Validation of Income and Expenses in Income Tax Returns. Public notice covering validation effective 1 January 2026.
• Xero — Just Ask Xero (JAX), official product overview. Official product guidance.
• QuickBooks — Introducing Intuit Intelligence. Official product capabilities and availability information.