Voice assistant for sales data with automated bank-deposit reconciliation

Case study overview

I built a voice assistant for a multi-location retailer to answer sales questions and reconcile deposits. Leaders wanted quick answers during morning check-ins, not another dashboard. Accounting needed a daily list of mismatches to fix before opening.

Constraints

Everything had to run on their network with strict access control. I had read only access to sales and accounting exports, and no card data could be stored or spoken. Confirm your approach with your attorney or compliance officer (PCI, banking, and privacy rules vary by state and industry) before you deploy.

Bank data arrived as nightly files. Existing systems could not change, so adapters sat beside them and any corrections required a person.

My approach

I added a small API that pulled sales by location and tender type from a read-only replica. The voice layer used a fixed set of intents and asked for a policy-approved PIN (or MFA) before sensitive answers.

Every exchange wrote to an audit log with timestamps and caller ID. Rate limits and timeouts kept calls short, and a chat option handled noisy spaces.

Reconciliation logic

A nightly job grouped point-of-sale batches by settlement date and location, then matched deposits by amount within a date window. Differences and missing items were flagged for review with clear reason codes.

The assistant answered sales by day and location, expected deposits, and open exceptions. It never read card numbers, and people approved any changes.

What I learned

Short prompts and firm confirmations reduce mistakes. Business calendars matter, especially weekends and holidays. Logs and real phrasing from callers are the fastest way to tune intents.

Try this this week

• Map sales, deposit, and fee sources, and name data owners.
• Write your top ten spoken questions and expected answer formats.
• Create a read-only reporting replica with a least privilege service account.
• Schedule a nightly bank export and store it in a secure folder.
• Build a basic matcher for amount and date, and read one report by phone using a PIN.
• Confirm data handling and authentication choices with your compliance advisor or attorney.

How I can help

If you want help, I build assistants and automations like this. My AI Agents and Personas, AI Chatbots, Business Automation, and AI Consulting and Strategy services apply, and I also provide AI phone and chat agents and can connect to AppWT CRM.