Retail
SOLUTION
AI in document and enquiry work.
Reading, retyping, classifying, answering. The more a task mixes manual work with judgement, the more the split decides the outcome.
THE PROBLEM
The time does not go only into the task itself.
It goes into checking, into rework, and into the difference between one person and the next.
- Every counterparty sends a different format, so retyping and checking take time.
- Enquiries land on the same few people, and finding the basis for an answer adds up.
- Accuracy and the level of detail in an answer vary by whoever handled it.
- Exceptions and incomplete cases are decided ad hoc, every time.
HOW WE WORK
Automate less than everything, and keep the judgement with people.
The steps either side are joined, and only what needs checking comes back to a person. Deciding what people keep matters more to the result than model accuracy does.
- 01Map the work and its exceptions
- 02Decide what AI takes on
- 03Design where a person checks
- 04Write into existing systems
- 05Adjust while it runs
WHAT WE DO
What ELYSAI takes on
- Mapping how the work runs today, and its exceptions
- Building the extraction, classification and draft-answer steps
- Designing the review screen and the route back when something is sent back
- Writing into existing systems, and the work of getting it into daily use
WHAT IT TAKES
What has to be true to proceed
Widening the scope too early makes the effect impossible to measure. We start with one piece of work, establish it there, and expand from that.
- Which documents or enquiries are in scope, and roughly how many a month
- The current procedure, and who does the checking today
- The system the result is written to, and whether it can be connected
- How much error is tolerable when something goes wrong
FAQ
What people ask about putting AI into everyday work.
Work that happens often, where the basis for a decision can be put into words, and where the cost of an error is understood. Those three together mean the effect shows up early and can be checked in an agreed way. The first conversation goes through your current work and narrows the candidates.
Often, yes. How much building it takes depends on how much they vary, so we look at a number of real documents first and establish the range of formats and the kinds of exception in them. From there we propose what can be handled automatically and where a person takes over.
Errors are assumed, so noticing them and recovering from them are part of the design. What counts as an exception, who reviews it, and how the outcome of that review is recorded are all decided before go-live, not worked out once it is running.
Some of it changes. Changing your procedures is not the goal in itself, though. We start from where the effort and the judgement calls sit in the way you work now, and look first at options that keep the change as small as possible. Where something does have to change, we explain why.
Contact
Let us work out which task to start with.
We go through the current procedure and propose where AI takes over and where people stay. The first conversation is online, with nothing to prepare.
Discuss using AI in your operations
60 minutes, online, at no cost