Why Business Data Stays Unused

The systems inside a small business normally arrive one at a time. Accounting runs in Xero or MYOB. Sales happen through an online store, a point-of-sale system, wholesale channels or a combination of all three. Customer relationships are recorded in a CRM, while operational detail often remains in spreadsheets.

Each system may be fit for purpose. The difficulty begins when management needs an answer that crosses all of them. A retailer may know total online sales and total store sales, yet still be unable to calculate a product's contribution after payment fees, shipping, returns and wholesale discounts. A hospitality business may have POS data, weather data, wastage records, web traffic and wholesale orders, but no dependable way to compare the signals before ordering stock or planning staffing.

Alastair McLeod, CEO of Wallace Corporation, describes the issue as a management problem before it becomes a technology project. "A business does not need every possible data point in one place. It needs the records that explain an important decision to be consistent enough to trust."

Data consolidation means extracting selected records from operational systems, standardising them and maintaining a common analytical layer. It does not mean replacing every existing system with one large platform.

Xero continues to run the ledger. Shopify continues to run the storefront. A POS system continues to record sales at the counter. The analytical layer makes it possible to reconcile the shared concepts between them: customer, product, order, payment, refund, supplier cost and time period.

That distinction matters. Many failed dashboard projects begin by visualising the first data export that is available. The result can look polished while still comparing incompatible totals. A warehouse or equivalent consolidated dataset gives the business a defined answer to questions such as: which sales channels are profitable after their fees; which product groups carry the highest wastage; which customers are becoming more valuable; and which demand signals should influence the next order.

For a Wallace Corporation client with financial, e-commerce and wholesale platforms, this work may begin with scheduled synchronisation and a reconciliation process. Matching records is often the labour-heavy part. The same customer can be named differently across systems, an order can be split across channels, and a payment may not have enough detail to match cleanly. A controlled AI-assisted matching workflow can resolve high-confidence matches and present ambiguous cases to a person with the competing records and the reason for uncertainty. That is a practical use of AI: it reduces repetitive comparison without inventing a financial answer.

For a prospect that also has POS, weather, wastage, web-traffic and social-channel data, the longer-term opportunity is a predictive model that combines the signals. The model is not the first deliverable. The first deliverable is a shared, inspected analytical layer that management, analysts and future agents can rely on.

Google's BigQuery, Snowflake, Redshift and Microsoft Fabric or Azure services can all provide capable analytical infrastructure. Wallace Corporation selects the platform according to the existing systems, data volume, skills, operating cost, security requirements and the decision the business is trying to improve. 

The platform choice is secondary to five design decisions:

1. What management decision needs better evidence?
2. Which systems hold the relevant facts?
3. Who owns each record when systems disagree?
4. How often must the data update?
5. Which people and systems require access?

This prevents a warehouse becoming an expensive archive. A business with a high volume of customer interactions may begin with customer and order reconciliation. A business with complex supplier costs may begin with margin. A business with variable demand may start by testing whether weather, promotions, wholesale orders and web traffic improve forecasting.

Cloud security is strong infrastructure, not a substitute for governance

Large cloud providers invest heavily in infrastructure security, but that does not remove a business's own responsibilities. The relevant question is not whether a cloud platform is secure in the abstract. It is whether access, retention, integrations and data-sharing rules have been designed properly for the information being stored.

Wallace Corporation's default position is need-to-know access. Staff and systems should receive only the data and permissions required for their role. AI tools used by privileged staff should normally begin with read-only access to the approved dataset, clear logging and an explicit human approval step before an action affects a customer, payment or financial record.

That approach is also more practical than a blanket ban on AI access. A controlled, read-only agent can help an operations or finance team identify anomalies, prepare a report or explain a movement in margin. It does not need the authority to alter the ledger, pay a supplier or contact a customer to provide that value.

From reporting to decision support

A connected dataset can support ordinary management work before any advanced model is introduced. It can provide a consistent exception list: orders with unusual discounts, products with worsening margin, invoices that do not match an underlying job, or customer groups whose repeat purchasing has changed.

The next step is prediction, where there is enough history and a measurable outcome. In the hospitality example, management could test whether POS patterns, weather, wastage, wholesale orders and digital traffic improve demand forecasting. The result should be judged against a baseline planning process, not assumed to be useful because it uses machine learning.

Tyrell's operational view is deliberately conservative. "A dashboard is useful when it tells the team what needs checking today, and lets them trace the result back to the records. If it only repeats last month's totals in brighter colours, it has not changed the work."

When consolidation is worth doing

Not every business needs a warehouse. A small operation with one or two reliable systems and simple reporting may be better served by a disciplined export process and a small number of reconciled reports.

The case becomes stronger when manual data assembly is recurring, when customer or transaction volume is high, when several systems need to agree before a decision is made, or when the business wants to apply forecasting or AI analysis to operational data. The conclusion is direct: agents and advanced analysis should not be the first layer placed over fragmented records. Establish the clean analytical layer first, use it to reconcile and govern the data, then give people and agents controlled access to evidence they can trust.