A manual task was initiated to compile a cross-reference table to do just that.

The task involved using whatever clues could be gleaned from the supplier's free-text description of the delivery site, to infer the client organisation that it belonged to, and then to look up the Client ID in a reference table.

Sometimes the client organisation was immediately apparent, for example if it said  "The University of Manchester - Rendezvous Bar" that would be clear. But more often than not, descriptions were terse or ambiguous, like "The Atrium Restaurant", "Jester's Bar" or "5 Michael Street" and would need further detective work, phone calls and time spent on Google.

It was taking a substantial amound of time to work through 10,000+ site descriptions like these and to match them to perhaps 150+ client IDs. The work was tedious and proceeding slowly.

I devised and developed an Excel 'power tool' to assist the user with the matching process.

It was designed to learn from the user's matching decisions, and to apply this knowledge to automatically recognize and match other similar records in future.

For example, once it had seen examples where (say) Jester's bar was paired with (say) Aberdeen university, or learned that the cafeteria for Wimbledon 6th form college was on St Michael's Street, it would be able to make these matches automatically in future.

It would also quickly learn that "The Student Union Bar" or "School Road" were unreliable pieces of evidence for matching, because without additional clarifying information, they're ambiguous.

The tool scored and sorted its matches based upon the strength of evidence it found in making the match, allowing the user to work through the list, skipping over the high-confidence matches and working on the harder, unresolved cases.

The tool made the matching task much faster, tractable, and possibly even a little more interesting.

Having been shown one or two example matches, it could match the next 50 similar cases on its own, so the matching process was greatly sped up.

It also built up a superhuman memory for the myriad fragments of information needed to identify site locations from ambiguous text.

The user didn't need to keep notes of the names and addresses of particular bars and cafes, nor remember which client organisations had which client codes. The tool took care of this automatically.

And if a mis-identification had been made, then the tool was perfectly able to quickly re-appraise all previous matches and update its recommendations.

The more the tool was used, the smarter it became. It only took a few days to develop, and a few hours to train to useful level, but it saved lots of time and doubtless avoided many manual errors in completing this task.