Lloyds Banking Group Claims Microsoft Copilot Saves Staff 46 Minutes a Day (theregister.com)
- Reference: 0179838262
- News link: https://slashdot.org/story/25/10/20/223252/lloyds-banking-group-claims-microsoft-copilot-saves-staff-46-minutes-a-day
- Source link: https://www.theregister.com/2025/10/20/lloyds_banking_copilot/?td=rt-3a
> Lloyds Banking Group [1]claims employees save 46 minutes daily using Microsoft 365 Copilot , based on a survey of 1,000 users among nearly 30,000 deployed licenses. According to Lloyds Banking Group (LBG), the rollout is "helping teams summarize documents, prepare for meetings, and reduce administrative tasks." Almost 5,000 engineers are also using GitHub Copilot. Vic Weigler, chief technology officer at the finance corp, said in a statement: "We converted 11,000 lines of code across 83 files in half the expected time."
>
> An insider at the bank, a self-professed fan of the technology, listed some of the ways it was being used in their business area. These ranged from the mundane -- drafting and summarizing emails, transcribing meetings, and comparing documents to group standards -- to the eyebrow-raising, such as drafting legal clauses, undertaking due diligence, and creating complex Excel formulas. They told us the next step is creating bots and agents to perform repetitive data-based tasks and rolling out the technology to customer-facing processes. That said, they also noted the AI tools occasionally make mistakes. The "golden rule," is to "never use the output without checking it."
[1] https://www.theregister.com/2025/10/20/lloyds_banking_copilot/?td=rt-3a
Added expense of copilot (Score:1)
Does the cost of copilot add up to more than paying those employees for 46 minutes of work? Is it worth the risk of confidential data being exposed?
Shooting themselves in the foot (Score:2)
so the outcome will be:
- reduced pay for those 46 minutes
- increased workload to fill those 46 minutes
- both ...therefore the real data is probably much higher savings
Re: (Score:1)
Reality take: 46 minutes more actual productive work can be gained from this. Invest in the system and increase the workload accordingly.
All others are "I have no idea how reality works/I am a horse pulled cart affectionado and I hate new technology that improves efficiency so I pretend I don't understand how efficiency improvements work".
The best part is that at this point, AI isn't limited by model quality. It's limited to people knowing how to integrate it into their workflows. And this is going to get b
Re: (Score:2)
> so the outcome will be: - reduced pay for those 46 minutes - increased workload to fill those 46 minutes - both ...therefore the real data is probably much higher savings
And here I thought you might have been talking about an AI report highlighting how the average executive golfer only works about 46 minutes per week, at an unjustified cost of WTF per hour..
eyebrow-raising (Score:4, Interesting)
...and creating complex Excel formulas
By christ, they should put 'guard rails' in that stop you ever doing anything complex in Excel. I dread to think of how much of that bank relies on cronky spreadsheets instead of proper, auditable, tested systems.
You: Create an excel formula to take X and Y and Z, and amortise over 6 months, then leverage A, B and C and exponentially decay adding those month on month to what you first came up with.
Copilot: Certainly! But first, please give me the magic number from your manager authorising this insanely stupid idea
Re: (Score:2)
You are enamored of " proper, auditable, tested systems." Such a mindset is appropriate for supporting "cash cows" with decades of history. Write the C-code/COBOL and get on with it. For clever new ideas chalk+slate or BOTE sketches prove more valuable. I prefer an artists 3'x2' artists sketchpad for my "hobby " calculations. For others an Excel spreadsheet replaces that slate+chalk in modern business hypotheticals and toy models. Toys are just bigger now than when Archimedes used sand-tab
Re: (Score:2)
Of all of the things I've heard complaints about misuse of Excel for over around three decades now, literal financial spreadsheets is a first for me.
How big of a Windows discount (Score:2)
...did MS give them to say that?
If that were true (Score:4, Insightful)
Then the staff can leave 46 minutes early.
Gotta love it. (Score:4, Insightful)
Business processes have become bloated because we slowly automate away the actual work. So, instead of actual work, we create unneeded and unnecessary streams of communication, now mostly emails and giant meetings filled with banal nonsense rather than meaningful discussions. And now we're at the point where we're automating away the banal, nonsensical streams of bullshit we've tacked onto work in order to keep people involved in processes as the actual work is automated away. A bullshit spewing, hallucinating LLM is perfect to automating away endless streams of communication that will then need to be summarized by another LLM to be readable by humans. We're at the point where the single sentence communication, "Hey, Bob, I need the price of $widget in March of last year for a report," will now turn into a 10k word essay produced by an LLM, at the other end another process will attempt to summarize that essay to its essence before a human reads it, and Bob will send Joe the projected price of $widget's_other_cousin ten years from now. But, maybe he'll luck out and the LLM he uses to respond will hallucinate the correct answer as it writes up his response, and the LLM at Joe's end won't mangle it completely as it tries to summarize the essay it receives before spitting out the summarized answer from the document produced by Joe's LLM.
This is business efficiency at its absolute peak! No wonder everyone's excited by AI!
Self-professed fan? (Score:3)
"An insider at the bank, a self-professed fan of the technology"
Lets see how much of a fanboy he is when HE gets replaced by it. And yes, it probably is a man as psychopathy tends to be more common in males.
Complex Excel scripts with Copilot? (Score:2)
... and that's surprising? It's a case where people are constantly having to deal with syntax, which we process poorly. It has immediate, stateable goals, and can be directly tested and checked and basically involves getting a machine to help you communicate with a machine. Pretty clear use case. I use Copilot for just that, all the time. It's not like "Hey, make this 15 step process with scripts", but when you want to offset a row based on an entry elsewhere and process it a certain way, it sure beats di
Shitty longwinded reports now easier to make (Score:1)
So mindless middle managers can churn out shitty longwinded reports that no-one reads 10% faster.
Woo. Pee.
Well, it's Lloyd's (Score:2)
I figure since they [1]also insure [lloydsbank.com], they probably have actuaries that can estimate the savings in a fairly rigorous way.
[1] https://www.lloydsbank.com/insurance.html
But will cost customers ... (Score:2)
... uncountable time and money!
Day 14,739: Management style now "nano". (Score:4, Insightful)
The "golden rule," is to "never use the output without checking it."
Just gonna state, plainly and flatly, that they are already not following their golden rule, and it will only get worse until disaster occurs.
also (Score:2, Interesting)
What constitutes "checking"? If the AI does something different from how you would have done it, do you rewrite it or say "eh, looks good to me"? What happens when the way the AI did it turns out to have weird corner cases?
Example - I asked ChatGPT for code to do some holiday calculations specifiying that it should only use code from the C standard library. To compare whether two dates were two days apart, it came up with something using mktime that more or less worked. Problem a) it assumed that time_t was
Re:also (Score:4, Insightful)
Summarizing a report for a prospective reader of said report is pretty much by-definition not checking it, because checking it would require reading both the generated summary and the actual report.
Summarizing reports seems like a great way for someone writing the report who has to say something about an unfortunate development managing to downplay it where the LLM fails to include it in the summary. It allows obfuscation of negative news or status potentially, and the consequence would likely fall on the person who should have read the report but instead got the not-even-cliffs-notes version.