Conference heron::euro_swas_ai

	I'm currently involved in a evaluation of Epitool and so far I can
	only agree that Epitool seems to fit nicely into the DEC portfolio
	of tools.

	With Epitool you can compile your knowledge base and use LISP's system
	building facility to produce an executable image, ie you don't need
	LISP on the production machine.

	It is very fast! One performance study showed that it was faster than 
	Nexpert and was as fast as OPS5 (it was actually faster than OPS5 if 
	you count the initialization part). Please no religious discussions 
	about OPS5 being the fastest tool around!

        The performance study is based on: Evaluation of AI languages and 
	Knowledge engineering environments: Phase III:
	Expert System Implementations Dr. William Metterey.

	I'll be glad to post the actual numbers later on when we have 
	put them in a readable form.

	About the Notesfile: One of the things we are proposing (if we ever
	get to a point where Epitool is used and supported throughout DEC) is
	that a notesfile about Epitool is maintained by somebody (perhaps in
	Sweden) and that person regularly (2 times a day ?) takes away any
	sensitive information and sends the questions to Epitec in Link�ping.
	We can use X25 or a modem for that, not Snailmail

	That way we should be able to have fast and accurate responses to
	questions. I have talked to Epitec about this and they think that
	they can have some sort of response back within 24 hours.
	That response can be of several types: The solution to the problem
	or just a notification that they have received it and they are looking 
	into the problem together with an intelligent guess about when they
	can answer the question completely.

	Try it you'll like it!	

	/ Ulf

First of all, does this mean that the agreement with Epitec is in place
(or getting there)?

Then two technical questions:

1) Will Epitool be available both on UIS and DECwindows?
2) Will Epitool run under VAX Lisp/Ultrix (DECwindows)?

Thanks and regards
Arne

    To answer a few questions regarding negociations with Epitool:
    
    1.Epitec has given us a deadline of March 1 to let them know whether
    we want to sign them up for a worldwide DDS agreement. Chuck Kiezulas
    (CAIM) is driving the process and has to get Corporate ARB (Application
    Review Board) to agree on a meeting scheduled for Feb 28 (close
    call!). If the OK is given, we would negociate for an EXCLUSIVE
    worldwide DDS agreement, except for Scandinavia (Sweden, Norway,
    Finland) and Denmark where the agreement would not be exclusive (i.e. 
    Epitec would probably continue selling their PC-386 version in those 
    countries).
    
    2. The support issue will be addressed up-front (there are many
    options at this point and time), to avoid repeating the same situation
    as with Nexpert.
    
    3. The Decwindows (actually MOTIF/OSF) version of EPITOOL will be
    promoted first, but I believe the UIS version will also be sold.
    
    4. There is already a version of EPITOOL in ULTRIX, although not
    fully tested and qualified.

    I just want to emphasize the importance of point 2 in Pascal's answer
    (.3): the support issue must be addressed upfront.
    
    NEXPERT is also a great tool, in the beginning we loved it so much that
    we would have given away our kingdom for it, but the support headaches
    being as they are, it turned out to be more like a Trojan horse for us.
    
    Let's do it right with EPITOOL and maybe we can straighten out things
    for NEXPERT as well - no need to throw away the silver just because we
    think we discovered some gold.
    
    Ren�.
    

    Good news (I guess)!
    Corporate ARB (Application Review Board) approved yesterday the
    proposal to sign an exclusive DDS (Digital Distributed Software)
    agreement with Epitec for EPITOOL.
    
    Themis Papageorge, Dino Lachiusa (who is taking over Chuck Kiezulas's
    job dealing with third-parties) and Jack Rahaim made a successful
    presentation which was well received.
    
    This means that final negociations will now take place around topics
    like pricing, royalties, support, internal versions and all the legal 
    stuff that goes with it. This could take anywhere between 2 and 6 months,
    so we can expect to announce Epitool around the Q1FY90 timeframe.
    
    This time, we will try and have all the structure in place to really
    be successful with a third-party product. Your comments are welcome.
    
    Pascal

    Thanks for the news, Pascal.  What can/should we tell customers
    at this point? 
    
    
    						Mark

    As a general rule, we should NOT hint anything to customers before
    the formal announcement (which date has not been decided). So I
    would recommend that we only share this information internally and
    wait for Corporate directions regarding timing of announcement.
    
    Pascal

    Greetings,
    
                I heard IBM just bought 17% of EPITEC. Anyone know if
    that is true? Does DEC have any investment in the company?
    
    Just wondering,
    
                     Dikk

Epitec is part of a family of Swedish companies. The parent, Pronator,
is a company of 2,500 people and revenues of 1.1B Kronor, $185M in
1987. The Pronator Group's commercial concept is to build, develop,
and operate knowledge-intensive service companies. A wholly owned
company of Pronator is Enator (600 persons, 270M Kronor, $45M).

Epitec is owned by Enator (40%), Pharmacia (10%), Alfa Laval (10%),
Rank Xerox (10%), and a group of the founders.

Enator has an option to buy another 21% of the company and hence
become the majority owner. This option may be exercised in late 
spring 1989.

Enator's business concept is to combine data processing and management
know-how and thereby provide quality services designed to increase the
competitiveness of businesses and organizations. Enator accepts total
responsibility for implementing development projects from concept to
finished system. Epitec contributes to Enator's business concept by
providing a complete range of knowledge engineering products and
services.

IBM Sweden (not IBM Corporate) is buying 17% of Enator (not Epitec).

It is my understanding that Epitec told DEC about the new situation as
soon as possible.

One of the reasons that Enator have been successful in Sweden (and
Europe) is that they are vendor independent and according to the
press release they will continue to be that. Pronator will still be the
biggest single owner in Enator.

P�R EMANUELSON/EPITEC TOLD ME THAT EPITEC, ENATOR, AND PRONATOR STILL 
WANTS THE DDS-AGREEMENT TO COME THROUGH.

I have no idea about the current status of the proposed DDS-agreement
and the official message to DEC about this whole thing but it's clear
to me that Epitec wants the DDS-agreement.

/ Ulf

    
    Thanks for the clarification, Ulf.
    
    This notesfile always attracts the most competent and knowledgeable AI 
    folks. From now on, I bring all my tough and arcane questions here.
    
    Thanks,
    Dikk                                                               
          

                    Thoughts on the selling of Epitool (ET)

			DIGITAL CONFIDENTIAL
    
James Soper @VBO					June 12, 1989
 (HERON::SOPER)


    Note: The author just recently joined digital.  Prior to that, he spent
    3 + 3/4  years working for IntelliCorp (the publisher of KEE) in
    California and Munich.  He worked in presales, training, support,
    consulting, and product development.   This document reports his
    thoughts about how to sell of Epitool (ET) as an  expert systems shell. 
    It is based on his experience working with and selling  KEE, and on 2
    weeks of working with Epitool.


Epitool and the COMPETITION:
    
    Selling ET against the MAJOR TOOLS: ("MTs" = ART, Knowledge Craft, KEE)
    
    The ET developer interface on VAXES compares very poorly with Nexpert
    and the MT's on their machines of choice (MACs, Suns, Lisp Machines). 
    The interface is far too slow and it is an important factor in making
    a sale.  

    ET CANNOT afford get into a features war with the MTs, it will get
    clobbered. You have to sell ET from its strengths.

    ET is NOT the best tool to use when the problem is very complex (lots
    of classes, deep hierarchy, lots of relations between  objects), or
    when the user will need to do a lot of customization (uncertainty
    factors, special interfaces, new ways of controlling rule chaining,
    etc.).        
    
    It should be possible to sell ET well against the MT's to customers
    who want to use only VAXes, who do not have unlimited budgets, who are
    starting in AI, and/or for whom delivery performance is very important.
    

    Selling ET against NEXPERT
    
    This will be the biggest competition for ET.  They are similar in
    price, and more or less in features.  Nexpert has had a long 
    headstart.  You probably could get into a features war with Nexpert -
    but, the  educating the customers will be very difficult.  Neuron Data
    can fill in a lot  of boxes in a features list, just as many as
    Epitool.   It also demos extremely well.   The fast graphics on a
    Macintosh, in the hands of a good demonstrater, can leave  people
    walking away with the impression that this is an advanced and
    easy-to-use  tool [ET's interface does NOT give that impression].  The
    problem with Nexpert,  is that it is limited in what you can really do
    in each of those feature boxes.   ET needs to show the customer that it
    does not have those limitations.   I refer the reader to experienced
    Epitool/Nexpert users to explain the  differences.


HOW TO SELL EPITOOL:

    You must sell a product from its strengths. 

    1) The major strength of ET vis a vis the MT's is as a development/
    DELIVERY  system.  ET was designed from the ground up with delivery in
    mind - you can  compile your KB into LISP, and potentially into ADA,
    Pascal and C.  This could  offer significant size and speed advantages
    at delivery time (it becomes  realistic to think of using 286's as a
    delivery machine, something that the  major tools cannot [yet?] offer). 
    Delivery speed and size on VAXes and PC's  probably should be the major
    message of ET.

    Note: I have NO experience using the compile-to-LISP/ada?/etc. feature
    of  Epitool.  I also have not seen any speed/size comparisons between
    the delivery  kb's of ET and that of the MT's.  Therefore, emphasizing
    ET's delivery  potential on the assumes that Epitool can produce a
    smaller, faster delivery KB  than an MT can, on a VAX or PC.  This
    assumption is based on the design  philosophy of ET, and NOT on real
    experience.  Tests should be run soon to  verify this assumption.  

    Note: I cannot predict how the speed of Epitool on a VAX will compare
    to that of  an MT on a SUN or EXPLORER II.  They will be tough
    competition, simply because  those machines are faster than VAX lisp.

    Note (May 5, 89): I have seen one performance test written up.  I
    showed  significant performance improvements of ET vs KnowledgeCraft. 
    However, the  write-up implies that they tested KC in interpreted mode. 
    All of the MT's now have compilers.  Any fair performance test should
    compare ET versus the  COMPILED version of the MT.

    Note (June 12, 89): At the Expert Systems show in Avignon France, the
    Inference distributor was showing ART/IM, which they claim is an ART
    redisigned and slimmed down (no viewpoints) for delivery - also on
    286's.  If ART/IM starts to sell well, a performance comparison will
    need to take it into account.

    Note (June 12, 89): Any serious attempt to penetrate the market has to
    include a version of ET-delivery running on (DEC) PC's.  A MINIMUM of
    25% of your customers will insist on it.
    
    
    2) A major strength of ET can be SERVICE.  You can differentiate ET from
    the  MT's, and from Nexpert based on service (please see the chapter on
    service in  Tom Peter's Thriving on Chaos on how to sell a product with
    service.).

    - DEC @VBO had very bad experience with Nexpert support - it didn't
    exist.

    - ET is the only medium-to-large tool "Made In Europe". This is a
    significant  advantage for European customers.  Telephone support can
    more effective when  support personnel does not have to wait till late
    in the afternoon to call  California for answers.   IntelliCorp's
    European support staff in Munich is very much overworked.  In the USA,
    this  is less of an advantage.  At least IntelliCorp's service in the
    USA is  reasonably good.  For Nexpert, a cottage industry has built up
    providing the  service and training that Neuron Data does not provide
    (Battelle in San  Francisco, among others).  In the USA, I would
    suggest considering a marketing  message based on "European Quality and
    Craftsmanship" (Volvo, Mercedes Benz,  SAAB, etc., have very good names
    there).

    - Worldwide, DEC/Epitech can offer 24 hr turnaround time for support
    through the  use of E-mail [this is assuming that the customers can
    E-mail into DEC/Epitech].   This should include patches.  Currently in
    Europe, getting a bug fixed by an MT  vendor can take weeks - in part
    because of the standard problems of getting  tapes across continents
    and oceans, and through customs.  It should be  technically possible
    for a customer to send a bug report directly to Sweden  by  E-mail, and
    to  receive a patch in 1 or 2 days, by E-mail.  This level of  service
    would put ET head and shoulders above the rest.

    - Note, IntelliCorp's training and consulting services are very good.
    They are  also very expensive (consulting:  $1,250/day).  I do not know
    what the situation  is with Neuron Data.

    - Please note: It is important to get a reputation for great service
    from the  start.  Thay means that support [communication] services must
    be in place at  product roll-out time.


    3) The DEC connection - DATA BASES.  There is a strong demand for tools
    that can  reason over data stored in data bases.  IntelliCorp built
    KeeConnection in order  to meet this demand (Nexpert can also connect
    into some data bases.).   KeeConnection is a very sophisticated tool
    that allows the customer to do all  kinds of things with the data as
    it's loaded from the data base into the  knowledge base, and sent back
    out to the data base.  While there is a large  market for his type of
    tool, it is combinatorially fragmented: each customer  uses a different
    data base on a different machine, and wanting to run their  shell on
    some other machine.  It quickly proved impossible to meet the 
    fragmentation of demand.  Offering a data base connection should easier
    job for  DEC/ET, because the number of host machines can be  limited to
    1 or 2 (VAXES and  PC's(?)), and it should be possible to target the
    most popular DB's on just  those machines.  A DB tool would also be, by
    definition, simpler to implement in  ET than in KEE; ET can do fewer
    things, so there is less to program in. I do not  know of any tool that
    ET has to connect it to data bases.


    4) The DEC Connection - MULTI USER.  Customers who have VAXes,
    automatically  expect that an expert system running on a VAX would be
    multi-user, just like a  data base.  They want to have several users
    (four, scores, even hundreds), able  to reason over the SAME
    knowledgebase.  To my knowledge, there is NO shell  anywhere, on any
    machine that can do this.  This is both a big market, and  completely
    untapped.  Technically, it is rather difficult to do (which is why 
    nobody has yet done it).  However, DEC is in a advantageous position to
    be able  to do it (because of it's knowledge of VMS and VAXLISP, and
    because ET can  probably deliver in a smaller size), and to reap the
    benefits from it.  I would  like to suggest that DEC and Epitech
    consider making a multi-user (delivery) ET  a 1 year goal of it's
    marketing strategy.

    (Note: June 9, 89: The people selling ART/IM at Avignon, mentioned
    above, claim that ART/IM has this multi-user capability.  I didn't have
    time to check it out.)
    
    
WHEN TO INTRODUCE EPITOOL?

    Please note: Outside of Sweden, ET is starting out with a name
    recognition of  0.0000000001.  I have never seen it listed in any tool
    comparison; nobody in  Intellicorp - USA had ever heard of it; it was
    poorly marketed in Germany, etc.  etc. etc..  This is an unknown
    product.


    I do not believe that we will be ready to sell Epitool for at least 3
    to 6  months.  My reasons are given below:


    o The developer's interface is far too slow (I have the UIS version on
    a 3500).   This will have a major impact on sales.  Nexpert and KEE
    running on their  machines of choice will destroy ET just on this
    point.  I must recommend that  we consider rewritting ET's interface in
    C (a major project) - because VAXLisp  is too slow to handle the job.

    o VAXLISP is too slow. When trying to sell ET on a VAX vs. tool X on
    Machine Y,  ET is likely to loose.  I don't know if this is due to the
    VAX, to the  implementation of LISP, or what. But the reality is that
    all the tools run visibly slower on VAXes than on almost any competing
    machines.  In effect, given that  the major language of AI is LISP, and
    that all the major shells use LISP,   biggest single problem in selling
    AI on VAXes is VAXLISP.

    o The sales staff is not yet ready.  Expert systems shells are highly
    complex,  technical products.  They MUST [I repeat: MUST] be explained
    and demonstrated  by people who know the product very well, and who are
    motivated to sell it. DEC  salesmen will be up against representatives
    from the MT and Nexpert vendors  whose jobs depend on knowing and
    selling that product.  A knowledgable  salepserson should be able to
    sell at least two to four times as much as someone  who just has had a
    1 week course about the product.
 
    o To my knowledge, DEC has no product manager for ET on line, and no
    sales  program: there are no strategies, no demos, no performance
    comparisons, nor  other materials that we can use to show the strengths
    of ET.

    o The support and training staff are not ready.  I don't think
    Epitech/DEC is geared up for handling the quantities of  customers that
    DEC might be able to bring in (if the above problems were taken  care
    of).  Epitech has only 22 people, 8 of them consultants.  IntelliCorp
    has many more people than that just doing support and training. 
    Epitool will need  fewer support people because the product is less
    complex, and they will at the  start make much fewer sales.  But if DEC
    wants to make service a prime reason  for buying Epitool, the support,
    documentation, and training staff must start to  grow soon, and it MUST
    BE WELL TRAINED.  It would take 1 to 3 months to find a good person,
    and 3 more months  of practice with the tool before they can become
    effective.

    
                          EPITOOL Notes File
    
        This is to announce the creation of a specific notes file dedicated
    to EPITOOL. It is called HERON::EPITOOL.
       Such a notes file is now needed since we are going to introduce this
    product on our portfolio. Its birth is connected to the next  AI  FORUM
    which takes place in Sophia-Antipolis the week of the 19th of june, and
    which is focused on EPITOOL.
    
              So please now direct your comments on this new notes file and
    Long Life to EPITOOL !!!
    
    DANIEL