A few days ago I recalled briefly some analysis undertaken by Bath Uni I think in the mid '90s about employers' accuracy in calculating SMP (and SSP). The error rate was high - 30% or so of cases had an error IIRC.
AI - the computer - will require programming, even if it is to evaluate information/data and make a decision.
The range of responses/action the AI could take may have to be determined/restricted (eg insufficient data to identify a course of action other than refer the decision/action to a human).
AI has to 'learn', which may be achieved by trial and error - but that requires huge amounts of experiences / data and 'feedback' (ie was the response /action correct - indeed, what is 'correct'?).
Trial and error. Did the AI operating the Tesla car that caused a collision and the death of the human 'driver' last year 'learn' from its mistake? If so, how? Has Tesla re-programmed the AI?
AI offers significant opportunities to improve so many things, but we should be aware of its limitations.
Matt, hi.
All this sounds good/bad depending on one's viewpoint.
The AI 'learning' aspect is fascinating. Machines neither 'think' nor 'learn' (not yet, anyway) so instead currently rely on their programming to decide the action to take. They will need lots of information and experiences and indeed coding to enable a choice of actions. There are huge practical problems that will have to be overcome. And we shouldn't overlook that an accountant or other service provider may have a liability to their clients if they give incorrect advice.
The robot you've outligned is surely no more than a bot that will sift queries / messages and deal with those which are 'simple' (eg ask for more information) or refer the matter for resolution by a higher intellect (the accountant?).
Thank you for this Kate. I have tended to 'delete' all HMRC emails about webinars - so missed this one which sounds like it was really informative.
Certainly my own move onto receipt of the state pension has revealed how HMRC allocates my personal allowance.
HMRC's tax coding systems were never perfect. And I suspect that there were many errors leading to under and overpayments of tax.
I remember the days when tax relief on mortgage interest was provided via tax codes...with the huge numbers of coding changes occurring several times a year.
And with the baby boomers increasingly moving into receipt of pension payments from two or more sources this is inevitably a potential source of coding errors for HMRC.
And of course many individuals have two or more sources of earned income nowadays.
I wonder whether overall HMRC is better today with issuing coding notices than it was say thirty / forty years ago?
The plan for RTI was imposed on HMRC but in my opinion it was too ambitious and unrealistic. HMRC had an implementation deadline to work to in order that earnings info would be instantly available to the DWP for purposes of UC payments to claimants. Well, UC also had an utterly unrealistic overly ambitious plan - and remains it would seem many years from roll out (if ever) across the UK.
In addition, HMRC has been subject to headcount reductions - again imposed by the Treasury.
It seems to me highly probable that HMRC senior staff warned the Treasury against unrealistic plans for RTI and headcount reductions.
The blame for this shambles rests in my opinion with Treasury ministers.
You're right to draw attention to the problem with how this filing expectation works.
I imagine that many employers (and their payroll systems) are unaware a FPS is required each pay period for each employee/pension recipient and think that all they need to do is send a FPS file containing FPS records for those actually paid in the period. This misunderstanding may in part derive from HMRC persistently referring to the FPS as though it is a file rather than a record for each person.
Perhaps employers should consider requesting a new PAYE scheme to distinguish between those with different payment frequencies?
I also have been told by an HMRC call centre operator to ignore the entries on the dashboard. To answer my query about a specified charge the operator looked at the actual entries held by HMRC which were different to those shown in the dashboard.
Specified charges arise where no RTI return (FPS or EPS) is submitted for the PAYE scheme for a tax month. Once on the system these SCs have the potential of causing mayhem in the dashboard, and HMRC's debt management department will pursue payment for it even though a payment made for the scheme will partly or fully extinguish the SC entry. Avoid SCs!
As regards UC, RTI and BACS, I'm in the HMRC camp with this. If it's ok for non-BACS payment to be passed to DWP and accepted for purposes of UC what makes a BACS-hashed supported RTI return better for UC purposes? Surely the RTI data in the FPS returns is what is important?
My answers
Francois - this is a great article.
Lol.
A few days ago I recalled briefly some analysis undertaken by Bath Uni I think in the mid '90s about employers' accuracy in calculating SMP (and SSP). The error rate was high - 30% or so of cases had an error IIRC.
AI - the computer - will require programming, even if it is to evaluate information/data and make a decision.
The range of responses/action the AI could take may have to be determined/restricted (eg insufficient data to identify a course of action other than refer the decision/action to a human).
AI has to 'learn', which may be achieved by trial and error - but that requires huge amounts of experiences / data and 'feedback' (ie was the response /action correct - indeed, what is 'correct'?).
Trial and error. Did the AI operating the Tesla car that caused a collision and the death of the human 'driver' last year 'learn' from its mistake? If so, how? Has Tesla re-programmed the AI?
AI offers significant opportunities to improve so many things, but we should be aware of its limitations.
Matt, hi.
All this sounds good/bad depending on one's viewpoint.
The AI 'learning' aspect is fascinating. Machines neither 'think' nor 'learn' (not yet, anyway) so instead currently rely on their programming to decide the action to take. They will need lots of information and experiences and indeed coding to enable a choice of actions. There are huge practical problems that will have to be overcome. And we shouldn't overlook that an accountant or other service provider may have a liability to their clients if they give incorrect advice.
The robot you've outligned is surely no more than a bot that will sift queries / messages and deal with those which are 'simple' (eg ask for more information) or refer the matter for resolution by a higher intellect (the accountant?).
Thank you for this Kate. I have tended to 'delete' all HMRC emails about webinars - so missed this one which sounds like it was really informative.
Certainly my own move onto receipt of the state pension has revealed how HMRC allocates my personal allowance.
HMRC's tax coding systems were never perfect. And I suspect that there were many errors leading to under and overpayments of tax.
I remember the days when tax relief on mortgage interest was provided via tax codes...with the huge numbers of coding changes occurring several times a year.
And with the baby boomers increasingly moving into receipt of pension payments from two or more sources this is inevitably a potential source of coding errors for HMRC.
And of course many individuals have two or more sources of earned income nowadays.
I wonder whether overall HMRC is better today with issuing coding notices than it was say thirty / forty years ago?
"HMRC's quite complex" national minimum wage regulations?
Hmm, the regs are not HMRC's but are those of the former trade and industry department (or whatever its name is today).
HMRC has the task of NMW enforcement.
"Flawed real time information (RTI) from HMRC’s payroll systems"
That text may be correct...but I'm presuming you mean 'from HMRC's PAYE systems'?
.......................................
PAYE in meltdown
The plan for RTI was imposed on HMRC but in my opinion it was too ambitious and unrealistic. HMRC had an implementation deadline to work to in order that earnings info would be instantly available to the DWP for purposes of UC payments to claimants. Well, UC also had an utterly unrealistic overly ambitious plan - and remains it would seem many years from roll out (if ever) across the UK.
In addition, HMRC has been subject to headcount reductions - again imposed by the Treasury.
It seems to me highly probable that HMRC senior staff warned the Treasury against unrealistic plans for RTI and headcount reductions.
The blame for this shambles rests in my opinion with Treasury ministers.
HMRC's RTI system - filing expectation
Kate
You're right to draw attention to the problem with how this filing expectation works.
I imagine that many employers (and their payroll systems) are unaware a FPS is required each pay period for each employee/pension recipient and think that all they need to do is send a FPS file containing FPS records for those actually paid in the period. This misunderstanding may in part derive from HMRC persistently referring to the FPS as though it is a file rather than a record for each person.
Perhaps employers should consider requesting a new PAYE scheme to distinguish between those with different payment frequencies?
What a mess!
I also have been told by an HMRC call centre operator to ignore the entries on the dashboard. To answer my query about a specified charge the operator looked at the actual entries held by HMRC which were different to those shown in the dashboard.
Specified charges arise where no RTI return (FPS or EPS) is submitted for the PAYE scheme for a tax month. Once on the system these SCs have the potential of causing mayhem in the dashboard, and HMRC's debt management department will pursue payment for it even though a payment made for the scheme will partly or fully extinguish the SC entry. Avoid SCs!
As regards UC, RTI and BACS, I'm in the HMRC camp with this. If it's ok for non-BACS payment to be passed to DWP and accepted for purposes of UC what makes a BACS-hashed supported RTI return better for UC purposes? Surely the RTI data in the FPS returns is what is important?