ACC 6070 Module 6 Final Project: Data Analytics and E-Commerce in Business Decisions Example

From Orders to Insight: Data Analytics and E-Commerce Across the Business Processes of a Montclair Coffee Roaster

Student Name

Master of Science in Accounting, Thomas Edison State University

ACC-6070: Accounting and Information Systems Design and Implementation

Mentor Name

Month Day, Year

From Orders to Insight: Data Analytics and E-Commerce Across the Business Processes of a Montclair Coffee Roaster

Introduction

E-commerce has changed what a small company knows about its customers. Every web order records who bought what, when, at what price, from which device and after which email. Analytics turns those records into decisions about pricing, purchasing, staffing and risk. For accountants, the change matters twice: analytics reshapes the business processes the AIS records, and it gives accountants new tools for control and assurance (Vasarhelyi et al., 2015). This paper asks how data analytics and e-commerce support decisions and processes at a small consumer business, what it takes to build the capability and what can go wrong.

Discussion: Data Analytics and Data Mining

Four Kinds of Analytics

Descriptive analytics reports what happened, such as sales by product and channel. Diagnostic analytics asks why, for example why subscription cancellations rose in March. Prediction estimates who is likely to cancel next month; prescription goes a step further and chooses the offer each of those subscribers should receive. Most small businesses work at the descriptive level; the value grows as they move toward prediction and prescription, but so do the data and skill requirements (Chen et al., 2012).

Data Mining Techniques

Data mining depends on data preparation. Romney and Steinbart (2018) stress the dependence of analysis on data quality, and in practice most of the effort goes into pulling, cleaning and matching records from separate systems before modeling starts.

Evidence on Value

Firms that base decisions on data tend to perform better. Brynjolfsson and McElheran (2016) report that data-driven decision making in U.S. manufacturing plants roughly tripled from 2005 to 2010, with adopting plants more productive. The evidence does not mean analytics pays everywhere; benefits depend on complementary changes in processes and skills.

Discussion: E-Commerce Models and Processes

Montclair Roasting sells through four e-commerce models. Its web store is business to consumer; its subscriptions are a recurring revenue model; its marketplace storefront sells through a third party that controls the customer relationship and charges fees; and its wholesale portal is business to business, letting cafés order on account. Each model changes the revenue cycle differently. The web store and subscriptions give the company full customer data and card payments processed instantly. The marketplace provides sales but little data, and it settles cash weekly net of fees. The wholesale portal replaces phone orders with structured data but requires credit approval and invoicing.

E-commerce also changes the expenditure cycle and controls. Packaging, shipping and payment processing become larger costs, platform and processor fees must be reconciled to sales and returns arrive through the mail rather than at a counter. The AIS must record each of these events with enough detail to analyze them, which means capturing order identifiers consistently across the store, the processor and the warehouse.

Analysis: The Company

Montclair Roasting has 42 employees, a roastery and two cafés. Its 2024 sales were $3.1 million through the web store, of which $1.9 million came from 7,400 subscribers, $0.8 million through the marketplace and $2.9 million from 160 wholesale accounts. Its systems are an e-commerce platform, a subscription billing application, a cloud accounting package and spreadsheets for roasting schedules and green coffee purchases. Three problems prompted this project: monthly subscription churn of 6.5 percent, stale roasted coffee written off at 9 percent of roasted inventory and card disputes rising toward the level at which card networks place merchants under monitoring.

Data quality is uneven across these systems. Customer records exist separately in the store and the subscription application, with no shared identifier, and wholesale accounts live only in the accounting package. About 12 percent of web customers also hold subscriptions under a different email address, so any analysis must first match records to a single customer.

Analysis: Analytics in the Revenue Cycle

Predicting Subscription Churn

Each subscriber pays an average of $21.40 a month, and about 55 percent of that, $11.77, is gross margin. At 6.5 percent monthly churn, about 480 subscribers cancel each month, and an average subscriber stays roughly 15 months. A classification model using skipped shipments, complaints, delivery delays, roast preferences and email engagement can score each subscriber's risk of canceling. If offers to pause, switch coffees or change frequency, sent to the highest-risk tenth each month, saved 15 percent of cancellations, about 72 subscribers a month would stay an average of 15 more months, worth roughly $150,000 of gross margin a year before the cost of the offers.

Customer Clusters and Product Pairs

Clustering web customers by recency, frequency, spending and product mix would likely reveal groups such as gift buyers in November and December, espresso enthusiasts who buy equipment and single-origin coffees and price-sensitive buyers of large bags. Each group warrants different emails and offers. Association rules on order lines would show which products are bought together, such as a grinder with a light roast sampler, supporting bundles and recommendations on the product page.

Wholesale Credit

For wholesale accounts, a regression of days to pay on order history, café age and payment method can help set credit limits and identify accounts whose payment is slowing before they become collection problems.

Pricing and Promotions

Regression on past promotions can estimate how sensitive each coffee's sales are to price, separating the effect of a discount from seasonal and email effects. If the analysis shows that a 15 percent discount on single-origin coffees mostly shifts purchases forward rather than adding new ones, the company can stop offering it, protecting margin. Testing prices on a random share of web visitors, rather than relying on history alone, would give cleaner evidence.

Analysis: Analytics in the Expenditure and Production Cycle

Green coffee is bought months ahead under contracts with importers, but roasted coffee is best within a few weeks. The company currently roasts to rough weekly estimates, and 9 percent of $2.4 million of roasted coffee, about $216,000, is written off or discounted as stale each year. A time series forecast by product, combining subscription commitments, which are known in advance, with seasonal web and wholesale patterns, would set roasting schedules and green coffee purchases. If the forecast cut stale write-offs to 5 percent, the saving would be about $96,000 a year, along with fewer stockouts of popular coffees.

Analytics can also improve supplier decisions. Combining cupping scores, prices and delivery records by importer would show which suppliers deliver consistent quality on time, informing which contracts to renew. Tracking roasting yields by batch would detect equipment problems early, since a falling yield often precedes a breakdown.

Analysis: Analytics for Control and Accounting

Order Fraud and Disputes

Anomaly detection on web orders can flag those unlike legitimate purchases: billing and shipping addresses far apart, several cards tried from one device, first orders for large quantities of equipment and expedited shipping to freight forwarders. Flagged orders can be held for review rather than shipped, cutting disputes, which cost the order value, a fee and the risk of losing card acceptance.

Continuous Monitoring

The same data supports accounting control. Daily reconciliations can match platform orders, payment processor settlements and ledger postings automatically, with exceptions routed for review, and marketplace settlements can be checked against expected fees. Vasarhelyi et al. (2015) argue that such data allows assurance to move from periodic samples toward continuous testing of full populations.

Revenue Recognition

Many subscribers prepay for three or six months. Under ASC 606, prepayments are a contract liability, recorded as deferred revenue and recognized as each shipment is delivered. The subscription application's data on prepaid months remaining supports the monthly entry, and analytics on prepaid balances and cancellations helps estimate refunds.

Inventory Valuation

Forecasting also supports the inventory reserve. Instead of writing down stale coffee when it is found, the controller can estimate the reserve at each month end from the age profile of roasted stock and the forecast, giving a more accurate cost of goods sold and fewer surprises at year end. Auditors can test the reserve by comparing past forecasts with actual sell-through.

Analysis: Building the Capability Through the SDLC

Systems Analysis

The project begins with an initial investigation and feasibility study. The analysis defines the decisions to be supported, churn offers, roasting schedules and order holds, and the data each needs. Economic feasibility compares the estimated $250,000 a year of benefits with costs; technical feasibility considers whether the e-commerce, subscription and accounting systems can export data; operational feasibility asks whether staff will act on the results.

Conceptual and Physical Design

Conceptual design chooses between buying analytics features in existing platforms, building on a cloud data warehouse with a visualization tool or hiring an outside firm. For a company of this size, the recommended design is a cloud data warehouse fed nightly from the four systems, with a visualization tool for dashboards and a part-time analyst or consultant to build the churn and forecasting models. Physical design specifies the data model, keyed on customer and product identifiers shared across systems, the refresh schedule, access controls and the reports each manager receives.

Implementation and Conversion

Implementation proceeds in phases: first the data warehouse and descriptive dashboards, then the forecast for roasting, then the churn model, then order fraud scoring. A pilot with one product line or one month of subscribers tests each model before it drives decisions. Staff are trained on what scores mean and when to override them.

Operations and Maintenance

Models degrade as customer behavior changes, so each model's accuracy is reviewed quarterly against actual results and retrained when it falls below a set threshold. Data definitions are documented, and a manager owns each dashboard.

Governance and Ownership

The project needs an owner outside IT. The chief financial officer should sponsor it, because the benefits fall in margin, inventory and controls, and a steering group of the heads of e-commerce, roasting and accounting should approve priorities. Data definitions, such as what counts as an active subscriber, should be agreed once and documented, so that every dashboard reports the same number.

Analysis: Risks and Limits

Analytics raises its own risks. Privacy obligations limit how customer data can be used, especially for advertising, and the company's privacy notice must describe its use of analytics. Poor data quality, such as duplicate customer records, produces wrong predictions with false confidence. Models can overfit past patterns, as forecasts built on pandemic-era coffee demand did for many roasters. Models can also treat customers unfairly, for example by flagging orders from certain areas as risky, which calls for review of the factors a model uses. Finally, dependence on platforms and marketplaces means the company does not control all of its data, which argues for growing the web store and subscription channels the company owns.

Security is part of the risk. A data warehouse that brings together customer, payment and sales data becomes an attractive target, so access should be limited by role, payment card data should stay tokenized and never enter the warehouse, and the controls described in the course's privacy module should apply to the analytics environment as fully as to the store.

Conclusion

E-commerce gives Montclair Roasting detailed data on every sale, and analytics turns that data into decisions across its key processes: keeping subscribers, roasting the right coffee, holding risky orders, monitoring transactions and accounting for prepayments. The estimated benefits, around $250,000 a year, justify a modest investment if it is built in phases through the systems development life cycle and governed with attention to privacy, data quality and model risk. For accountants, the same capability that informs managers strengthens control and assurance.

The first step is modest: a data warehouse and three dashboards, built in one quarter, that let managers see subscriptions, roasting and orders in one place before any model is trusted to make a decision.

References

Brynjolfsson, E., & McElheran, K. (2016). The rapid adoption of data-driven decision-making. American Economic Review, 106(5), 133-139. https://doi.org/10.1257/aer.p20161016

Chen, H., Chiang, R. H. L., & Storey, V. C. (2012). Business intelligence and analytics: From big data to big impact. MIS Quarterly, 36(4), 1165-1188. https://doi.org/10.2307/41703503

Romney, M. B., & Steinbart, P. J. (2018). Accounting information systems (14th ed.). Pearson.

Vasarhelyi, M. A., Kogan, A., & Tuttle, B. M. (2015). Big data in accounting: An overview. Accounting Horizons, 29(2), 381-396. https://doi.org/10.2308/acch-51071