4 ms·

For $55M a year, this is cheap. The mistake was to repeatedly spend 100's of millions to try to automate it. Go take a 35 yr old SF171 or such and scan, OCR an

For $55M a year, this is cheap. The mistake was to repeatedly spend 100's of millions to try to automate it. Go take a 35 yr old SF171 or such and scan, OCR and verify it. Get back to me when you can do that cheaply 10M times.

acchow 13y ago
A 61-day processing period is not cheap for the person waiting for money. You want to digitize old forms for retired people? This is entirely the wrong approach. New data should be in digital form. Not even all new data, just one vertical at a time - new payroll data, new business expense data, new insurance claims data, etc. This shards the work into tractable, testable, measurable units. There will be a (long) period when processing retirement forms will need to pull data from both analog and digital sources. That time will eventually pass, and the analog gradually phased out.
- hindsightbias 13y ago
  Investing 100's of millions into a parallel system that won't yield much until a good portion of your future retirees are in it (and all 50 diferrent federal agencies records are integrated in) is a hard sell in any universe. In the meantime, that system will obsolesce in 5 years and you'll have to redo it all again.
- watt 13y ago
  Are you aware of time periods involved, though? When you say "new data" do you understand that this new data will only become pertinent in about 50 years, when the employee that just came in to system, retires. 50 years later. And meanwhile, you are processing paperwork for people that entered the system 50 years ago: on today, 1964. (Ok, maybe closer 40 years, but the point is the very long storage time.) Even if all new data is digital today, what about 50 years later, how easy will it be to use digital files that were generated 50 years earlier. What if file formats change every 10 years. Then 50 years on, you got 5 different file formats you must be able to work with.
jpatokal 13y ago
I doubt this is insurmountable. For comparison, as of last year, Google Books had scanned 30 million books, not pages, all of wildly varying formats.
- acdha 13y ago
  What was their accuracy rate? What's your process for not being sued when someone's retirement is calculated assuming they made $10,000 when they retired instead of $70,000 because your OCR glitched? Google Books is a wonderful service but few services meet the level of importance and liability which a retirement system will have.
  - hindsightbias 13y ago
    Wait until those glitches show up on 60 Minutes and Congressmen are ranting about another digital disaster... There's no win here when you're dealing with ancient records and dozens of agencies with different standards. It makes me laugh when people see this as an easy project and don't grok the politics and all. Decades back, I saw a "by hand" project that generated >$100M in tax revenues from corporate non-compliance in one test region, and the next year, Congress cut/delayed an automated expansion because the biz folks whined about getting caught.

acchow 13y ago

A 61-day processing period is not cheap for the person waiting for money. You want to digitize old forms for retired people? This is entirely the wrong approach. New data should be in digital form. Not even all new data, just one vertical at a time - new payroll data, new business expense data, new insurance claims data, etc. This shards the work into tractable, testable, measurable units. There will be a (long) period when processing retirement forms will need to pull data from both analog and digital sources. That time will eventually pass, and the analog gradually phased out.

hindsightbias 13y ago
Investing 100's of millions into a parallel system that won't yield much until a good portion of your future retirees are in it (and all 50 diferrent federal agencies records are integrated in) is a hard sell in any universe. In the meantime, that system will obsolesce in 5 years and you'll have to redo it all again.
watt 13y ago
Are you aware of time periods involved, though? When you say "new data" do you understand that this new data will only become pertinent in about 50 years, when the employee that just came in to system, retires. 50 years later. And meanwhile, you are processing paperwork for people that entered the system 50 years ago: on today, 1964. (Ok, maybe closer 40 years, but the point is the very long storage time.) Even if all new data is digital today, what about 50 years later, how easy will it be to use digital files that were generated 50 years earlier. What if file formats change every 10 years. Then 50 years on, you got 5 different file formats you must be able to work with.

jpatokal 13y ago

I doubt this is insurmountable. For comparison, as of last year, Google Books had scanned 30 million books, not pages, all of wildly varying formats.

acdha 13y ago
What was their accuracy rate? What's your process for not being sued when someone's retirement is calculated assuming they made $10,000 when they retired instead of $70,000 because your OCR glitched? Google Books is a wonderful service but few services meet the level of importance and liability which a retirement system will have.
- hindsightbias 13y ago
  Wait until those glitches show up on 60 Minutes and Congressmen are ranting about another digital disaster... There's no win here when you're dealing with ancient records and dozens of agencies with different standards. It makes me laugh when people see this as an easy project and don't grok the politics and all. Decades back, I saw a "by hand" project that generated >$100M in tax revenues from corporate non-compliance in one test region, and the next year, Congress cut/delayed an automated expansion because the biz folks whined about getting caught.