Reconciliation of data entered

[MikeWright2]

Our Enterprise Architect is asking a lot of questions about how we (the RPA Team) are ensuring the data our Bots have entered into SAP is reconciled against the process input data.  He wants to know how we (which also includes the business) are checking to make sure sensitive data is 100% correct i.e. we've automated the input of Pricing data from spreadsheets, nobody wants to get Pricing data wrong. I've explained the Bots can check what they've done but will only be 'marking their own homework', and that the Bots will always enter data 100% accurately (no transposing characters/numbers).  I have also stated that the business still own the process so it is up to them to determine the level of risk they're comfortable with.   I'm running out of ways to explain this.  Has anyone had similar conversations or push back from their IT dept?  If so, how was it addressed?  What was the 'coup de grâce' that satisfied the concern?   Is there any material that covers this topic specifically? Appreciate any help or insights others can offer.  

[AndreyKudinov]

You might question how it is ensured that human enter pricing data 100% correct. I've seen cases where like 3 people miss basic error in data entry. Then explain that robots likely do it much more accurately (whether you are comfortable claiming 100% accuracy or not is up to you). Explain that you can add extra checks, but that requires additional execution time, which costs money and you should only do that if risks are high enough to cover the cost.   

[John__Carter]

Short of having a human manager check what the robot has entered before committing changes, the best realistic option is to have the robot read back it's own input and 'double check' the entries are as expected. You may also consider introducing a limit to the $ amount a robot can enter, and any case over that maximum must be worked by a human. But as aikudino says, maybe arrange an accuracy contest between robots and humans. I know who will win...

[ScottFasser]

Seems like you could create a separate robot process to pull samples from each data set after the initial process for checking and then generate a report or alerts if there are mis-matches? Would that work?

------------------------------
Scott Fasser
------------------------------