00100	                     COMPUTER SCIENCE DEPARTMENT
00200	
00300	                         STANFORD UNIVERSITY
00400	
00500	                     Stanford, California 94305
00600	
00700	

00800	                                                       August 16,1971
00900	
01000	Dear President Lyman:
01100	
01200		Early  in  July,  I  read  in  the  Wall  Street Journal that
01300	Stanford had received the very first IBM 370/145 computer and that it
01400	was  to be used for "streamlining administrative computing".    I had
01500	known that the Administrative Computing Facility wanted a 370 but had
01600	imagined   that   in   the   current  budgetary  stringency,  it  was
01700	inconceivable that they would get it.  Reading this  and  remembering
01800	that  I  had  read in Dave Phillips's report that Stanford spent more
01900	than $1,000,000 out of general funds on administrative  computing  in
02000	1969-70  triggered  off a sense of obligation to look into the matter
02100	and to protest if my feeling that an administrative monster had  been
02200	created was justified.
02300	
02400		I  did  look  into  the  matter,  and I came to the following
02500	conclusions:
02600	
02700		1.    Stanford spends more  by  about  a  factor  of  two  on
02800	administrative  computing  than any other university I could find out
02900	about, computing the expense either on a per student or on a  percent
03000	of operating budget basis.
03100	
03200		From  one  source  I obtained a figure of $50 per student per
03300	year, reducible by economy but not halvable.  Another source gave  me
03400	$6  to  $20  per student per year.  Still another source said between
03500	.25% and 1.25% of the operating budget with no  clear  definition  of
03600	what expenditures are included in the operating budget.
03700	
03800		2.  I  don't  think  the expense is justified by the level of
03900	service obtained or even by the level of service  the  administrative
04000	computing  group  is  trying  to  achieve.     The examples they cite
04100	concerning the sought  for  ability  to  give  immediate  statistical
04200	reports  in  answers  to  queries about the large data base they have
04300	just doesn't seem worth its enormous cost with the  present  hardware
04400	and  software.  A large part of the questioning here must be directed
04500	at the administrative offices that consume these  reports  and  which
04600	will  get  the on-line system. Perhaps their budgets are too large or
04700	perhaps the money they spend on all this computation  is  not  really
04800	spendable on other things.
04900	
05000		3.   The  cost-effectiveness  of  this expense has apparently
05100	never been certified by any body capable both of judging  what  these
05200	services  should  cost  and  what is cost-effective from the point of
05300	view of the university as a whole.   In  particular,  the  university
05400	Computer  Facilities  Committee,  although  it  has  heard reports on
05500	administrative computing, did not attempt to arrive at an independent
05600	opinion.  This is true even though consultants have been employed and
05700	there has been much discussion of  details  such  as  which  computer
05800	configuration  is  best  or  what  applications  can  be done on what
05900	machines.
06000	
06100		4.   Project INFO, aimed at providing on-line access  to  the
06200	administrative data base to all levels of administration, is a  cross
06300	between a sponsored research project and an internal development.  To
06400	the extent that it is  the  former,  it  should  be  subject  to  the
06500	conventional peer group evaluation of its scientific or technological
06600	merit and fully paid for by the sponsor.  Project INFO has  not  been
06700	evaluated   for  technical  merit  by  either  the  Computer  Science
06800	Department or the  Business  Schools.      If  strictly  an  internal
06900	development, then it needs to be evaluated for its cost-effectiveness
07000	in terms of the needs of Stanford as a whole.
07100	
07200		5.    I  have heard an allegation that the amount of computer
07300	time spent on Project INFO exceeds the costs allocated to it.  It  is
07400	also  said  that  overhead  costs are not allocated to the project so
07500	that the cost to the university is further understated. This may also
07600	be contrary to government auditing rules.
07700	
07800		6.  In  my  opinion, the goal of the Administrative Computing
07900	Group,  namely  to   make   Stanford's   administrative   information
08000	accessible  in  an on-line way, is worthwhile.  However, it cannot be
08100	achieved at reasonable costs by the methods they are  using,  and  it
08200	may not be achievable at reasonable costs today by any methods. There
08300	are two problems:  the hardware and the software.  Their equipment is
08400	barely  capable  of  doing  the on-line work in addition to the batch
08500	processing, they may have to add more especially  memory  to  realize
08600	the  planned  applications, and the consoles presently being used are
08700	very expensive.  Their software costs are also very high, $50,000  to
08800	$100,000  per  major application, and this is partially a consequence
08900	of the IBM standard method of debugging.  It  may  be  quite  a  time
09000	before  IBM  gets  a  proper  system  for  on-line  applications, and
09100	Stanford cannot afford to develop one for administrative use. Sharing
09200	some  costs  is  possible  with  academic  computing at Stanford, and
09300	sharing development costs with  other  universities  is  conceivable.
09400	This  opinion  on  what  the technical problems are is tentative; the
09500	opinion that the present plan is not cost-effective is firm.
09600	
09700		7.   The opinion that the Administrative Computation Facility
09800	is  not cost-effective is general among faculty members, researchers,
09900	and administrators who have had dealings with  it.   However,  no-one
10000	has  felt  it  his  business to protest.  It is also not precisely my
10100	business to protest; I am not even a user of its potential rival  the
10200	Computation  Center since the Artificial Intelligence Project has its
10300	own (government furnished) facility.  However,  there  are  potential
10400	savings here that may be as large as $500,000.
10500	
10600		Possibly  relevant  in  estimating  the cost-effectiveness of
10700	Stanford's  administrative  computing  is  the  experience   of   the
10800	psychiatric  clinic.   They  received  an  outside bid of $15,000 for
10900	services for which the Clinic Billing Office proposed to charge  them
11000	$78,000.   They  also expected to get better service from the outside
11100	company on the basis of the  experience  of  the  Palo  Alto  Medical
11200	Clinic,  the  Menlo Medical Clinic and various private practitioners.
11300	Stanford met the outside bid  and  retained  the  business.   Medical
11400	billing  is  a  well  defined  operation  and  so a yardstick bid was
11500	readily obtainable; the registrar is perhaps not  so  fortunate.   On
11600	the  other  hand, the low charges of outside medical billing services
11700	are based on a large volume and considerable competition.
11800	
11900		8.     This  situation  apparently  came  about  because  the
12000	administrative  computing  facility  is under the jurisdiction of the
12100	part of Stanford that handles the money.  Therefore, empire  building
12200	with  Stanford  money  is  easier  here  than in the academic part of
12300	Stanford.   Stanford has other little monopolies that are often  more
12400	expensive  to  use  than  outside  services.     I observed a similar
12500	situation in the Palo  Alto  School  District  when  I  served  on  a
12600	committee to look into their computing set-up.
12700	
12800		9.  Stanford badly needs on-line computer facilities, and the
12900	expenditure of general funds for this is justified, but research  and
13000	teaching  need  it  much  more  than administration.   The Artificial
13100	Intelligence Project has good on-line facilities (a visit would  show
13200	you what I mean), and the rest of Stanford should have them too.
13300	
13400		My  sources of information were the following:  conversations
13500	Einar Stefferud, James  Farmer,  and  Benjamin  ...     (all  outside
13600	Stanford)  about  expenditures  by  other  schools  and conversations
13700	within  Stanford  with  Robert  Augsberger,  Hank   Epstein,   Edward
13800	Feigenbaum,  George  Forsythe,  Bill  Miller, Dave Phillips,and Peter
13900	Rosenbaum;  I  shall  not  quote  them  since  they  can  speak   for
14000	themselves.
14100	
14200		I also had a conversation with Mike Roberts and John Gwynn in
14300	which they answered fully all the questions I was able  to  formulate
14400	at  the  time.    I  came away with the impression that they would be
14500	successful in implementing their  system  but  that  there  would  be
14600	substantial  costs  in  addition  to  those  planned for.   I was not
14700	satisfied with their answer to the cost-effectiveness questions.
14800	
14900		It is not reasonable to take action on  the  basis  of  these
15000	conclusions  unsupported  by  further  information.       In order to
15100	establish a basis for action, I suggest that you form an ad hoc group
15200	that   will   determine  the  answers  to  the  following  questions.
15300	Alternatively, the Computer Facilities Committee could be asked to do
15400	the job, but it might be unpleasant for them.
15500	
15600		1. Which if any of the above conclusions are correct?
15700	
15800		2.    What  level  of administrative computing expenditure is
15900	justified at Stanford?
16000	
16100		3.  How can these services be obtained most economically? The
16200	possibility  of  converting  the  370/145  to  a  370/135  should  be
16300	considered and so should  the  possibility  of  doing  administrative
16400	computing  on  the  campus  machine.  However, the main way of saving
16500	money is to decide that some of the proposed new applications  cannot
16600	be afforded.
16700	
16800		4.     The scientific and technological merit of Project INFO
16900	should be evaluated, its true cost  to  Stanford  determined,  and  a
17000	decision made as to whether and how it should be continued.
17100	
17200		5.     Are personnel changes required to implement the policy
17300	recommended?
17400	
17500		The  group  needs  competence  in  university  administrative
17600	computing,  accounting,  the technology of on-line computing, and the
17700	ability to weigh what is best for the University as a  whole.     All
17800	these qualities can be found among the Stanford faculty.
17900	
18000		I would welcome the opportunity to discuss this question with
18100	you in person.
18200	
18300	
18400					Sincerely yours,
18500	
18600	
18700	
18800					John McCarthy
18900	
19000					Professor of Computer Science
19100	
19200	
19300	cc. R. Augsberger, K. Creighton, G. Forsythe, W. Miller, M. Roberts