6 hours, down to 2 minutes

The context

Charles is a valued member of our client’s accounting team. He was brought in to help with credit card reconciliation, where the team's daily credit card reconciliation was taking up to six hours daily. The person responsible for that task started first thing and by about 3pm was normally able to work on resolving the problems: missing transactions or late paid items from the payment service provider.

In this historic city hotel, total guest transactions are, on average, 700 daily, meaning 1,400 different records need to be matched from six different systems, representing front-desk, lounge, bar, spa and restaurant via the Property Management System or other Point of Sale terminals. These are matched with two separate files from the hotel's two acquirers.

While the file formats are stable, the team needed to deal with occasional variations in structure and, of course, random missing data. The majority of transactions are matched by science, but some require some form of artistry, including detective work. For example, matching the $300 Spa bill with six different payments of $50 from different guests over a three hour period as their treatments finished.

When engaging with Hotelworld the CFO already knew that traditional RPA (Robotic Process Automation) wouldn’t work as it left too many transactions unreconciled when one or more of the rules was breached.

Charles offered an intelligent solution, utilising a mix of rules-based and AI-based matching.

The change

Charles now works the early shift. The hotel closes off the business day at 4am and by 7am Charles has used his automated secure connections with the hotel’s various systems to pick up the data needed.

He validates the inputs, corrects occasional small mis-matches in structure and runs the reconciliation. He doesn’t complain, or roll his eyes - he doesn’t even need that second coffee to get going. By 7:02 am most days, the reconciliation report is available on his dashboard, along with a notification to key staff to state the results. This gives results by card issuer, department and outlet.

The Match Process

1. Once Charles has checked the validity of the input files, he then takes a step-by-step approach to matching.
   
2. Ingests the records to create his match store. In this case it will generally be 1,400 records daily, but this will also include previously unmatched records awaiting a pair.
3. In this hotel, the average time for acquirer payment is 4 days, so we do not expect a match for those records until then. These are flagged Blue so the users can filter them out. If they remain unmatched on Day 5, they change status and are reported as unreconciled.
4. Using a set of rules, Charles looks for markers in both sides and reports:
   Matches with 100% confidence (Green).
   Lower confidence, ‘likely matches are stored as Amber for a human decision.
   No match or low confidence of match are highlighted as Red.
5. Voids are noted and matched as grey. He’s essentially using his normal matching algorithm but looking for matching records on the transaction side (PMS or POS).
   
   

Critically, he presents the results in an intuitive way that is easy to access and understand. He can be overruled with a few clicks to either pick up an extra transaction or to reject his matching suggestion. Each time either of those happens, he understands what was wrong and self-learns to avoid making the same decision again.

The colleague who didn’t get to start resolving issues until 3pm in her working day, now takes less than 10 minutes to run cross checks on Charles’ output, tweaking where necessary, but then is straight on to problem resolution. When she’s happy with the final result, she instructs Charles to create the ledger file with a simple click and this is done automatically via his connector to their accounting system.

Her day now involves a range of additional tasks that are developing her career and adding greater value to the whole team.

At month end, Charles creates a range of audit and ledger reports, showing the accrual position taking into account the 4 day payment delay. He also ingests the bank statement to validate that stated payments have been received into the bank, immediately escalating any variances.