Excerpts from:


James Love, "ISDN Pricing, What Went Wrong," paper presented at
the Harvard Information Infrastructure Project (HIIP), Policy
Roundtable on Next-Generation Communications Technologies:
Lessons from ISDN, June 24, 1998, NIST, Gaithersburg, MD


      [Section on measured of usage costs]



The Model

     The public switched network (PSN) contains a number of
elements, and various accounting models have been used to
allocate costs.  These accounting models were developed at a time
when the PSN was used primarily for voice services.  Some of the
recent problems with pricing of ISDN services are errors
associated with the use of cost estimates based upon voice usage
for data services.

     To illustrate the most basic problems, consider the
following stylized model of PSN costs.


(1)  C    =    L    +  O +  V

     Where

     C         =    Cost of service
     L         =    Loop costs
     O         =    Other (non-loop) fixed costs
     V         =    usage sensitive costs


Non-tariff sensitive costs

     Although there are lots of different ways to allocate the
fixed cost components L and O to PSN services, one might start
with the assumption that these are within a few dollars per month
of the costs of POTS service.

     The ILECs have filed public data on the NTS costs of BRI
ISDN in several regulatory proceedings.   In an October 18, 1995
filing with the FCC, Bell Atlantic reported that its monthly loop
costs were $14.56 for POTS and $16.77 for ISDN, a difference of
$2.21 per month, or 15 percent. (Exhibit 3, Bell Atlantic data
request, CC Docket No. 95-72, October 18, 1995). In addition,
Bell Atlantic reported NTS non-loop costs for ISDN of $4.18 per
month. On page 4 of that pleading, Bell Atlantic
said:

     Shown in the answer to question 1 and as discussed in
     Bell Atlantic's Comments and Reply Comments in this
     proceeding (copies attached), the difference in the
     costs between dial tone lines and Basic Rate ISDN loops
     is minimal. The small difference that does exist is due
     to an additional ISDN plug-in card which is required to
     transmit ISDN signals.

     Bell Atlantic's 1995 estimate of $6.31 for the incremental
NTS cost of ISDN is actually higher than for the other major
LECs. In the same proceeding, US West estimated that its NTS cost
of ISDN was only $1.18 higher than POTS. (Exhibit 4, June 29,
1995 filing by US West in CC Docket No. 95-72). Data on the NTS
cost of ISDN were reported somewhat differently in the
FCC's December 24, 1996 Notice of Proposed Rulemaking, Third
Report and Order, and Notice of Inquiry, on page 36,
Table 2 (Exhibit 5, FCC 96- 488). The following data were
reported:



      Table 1
Ratio of Costs of
Standard Analog
Service to BRI ISDN
Service
LEC                  Outside Plant (loop   All NTS costs
                    only) costs
Ameritech            1:1.07                1:1.45
Bell Atlantic        1:1.01                1:1.36
NYNEX                1:0.85                1:1.23
Pacific Bell         1:1.05                1:1.13
US West              1:0.80                1:1.07
Average Ratio of     1:0.96                1:1.124
Costs

Source: FCC 96-488,
December 24, 1996
NPRM, Table 2, page
36


     For the five ILCEs that responded to the FCC inquiry, the
average incremental cost for all NTS BRI ISDN costs was 12
percent of the POTS cost.  (Both NYNEX and US West reported BRI
loop costs lower than POTS loop costs.)  Not only are these cost
differentials are relatively minor items, relative to current BRI
ISDN tariffs, but they could only be expected to fall, as the
costs of ISDN equipment fall, as other computer and
telecommunications hardware prices have fallen.


Traffic Sensitive Costs

     The larger and most important disputes in the US ISDN cases
concern the estimates of traffic sensitive costs.  The ILECs and
many PUC staffs have used older average cost pricing models to
calculate these costs, often with disastrous results.  In
general, the traffic sensitive costs are supposed to be estimates
of the resources that are required to actually use voice or data
services.  In one sense, making a telephone call or sending a
packet of information over a network doesn't cost anything,
assuming that the network is already built, and it is not
congested.  However, it is essential that one consider the costs
to "build out" the network to accommodate a new user, or more
specially, additional use.

     The PSN is now built to provide circuits to consumers, and
it is built on the assumption that only a minority of consumers
use the network at any one time.  When a user makes a connection
and sends or receives information, there is no extra depreciation
of the facilities, only a demand to use network capacity.
"Traffic sensitive" costs are the costs of providing circuits
during the peak period.  These are shared resources.

     In the U.S., voice services are currently designed so that
there is one circuit for seven or eight residential lines.  For
business users, it is roughly one circuit for four lines.  These
are based upon averages.  Not everyone places the same demands on
the network.

     In regulatory proceedings, the costs of providing circuits
are sometimes allocated to users on the basis of time of usage,
according to the following.

(2)       c    =    k / M

     Where

     c    =    cost per unit of time for a circuit
     k    =    the cost of a circuit
     M    =    the load on the circuit in units of time.

     This is mostly an accounting convention.  It is a way of
allocating cost to different users, on the basis of a measure of
the use of the network.  It does not measure the costs imposed on
the network, because it does not distinguish between peak and off
peak usage.  If all users have approximately the same
distribution of calls in peak hours, this is not a major problem.
However, if the distribution of peak usage is different for
different network uses, it will be a problem, and can lead to
misallocation of network resources.

     This is clearly the case when looking at Internet usage.  To
appreciate this, let us consider some stylized (and actual) usage
data.

-    According to PUC staff in one state, studies have found that
residential voice callers use the network an average of 5 minutes
in the peak hour, and that 16 to 17 percent of residential usage
occurs in the peak hour.  Business usage was said to be about
twice as much as residential users, with about the same percent
in the peak hour.  Based upon this, and assuming even
distribution throughout the month, residential usage would be
about 30 minutes per day, or about 15 hours for the month.

-    According to a LEC in another state, a study of voice
traffic indicated residential usage of 640 minutes per month, or
10.67 hours per month, and 21.3 minutes per day.

-    A network engineer in anther state told me that in designing
PBXs installations, the assumption was the 14 to 18 percent of
calls occur in a peak hour, and that 16 or 17 percent peak usage
sounds about right.

     As indicated above, most ILECs provision residential lines
at one circuit for every seven or eight customers, so a consumer
requires .125 or .14 of a circuit.  Business lines require .25 of
a circuit.

     Looking at the residential data only, and assuming that the
cost of the circuit (excluding the loop cost) is $10 per month,
we can estimate the per minute cost of calls using equation (2).
And, we can also consider the cost of a nailed up line, where 1.0
of a circuit is used 24 hours per day, seven days a week, 4.3
weeks per month, or 43,344 minutes.


                             Table 2
                 Residential Calls, k = $10/mo.

          .125 * $10 / 640    =    $.0020 minute
          .125 * $10 / 900    =    $.0014 minute
          1.0  * $10 / 43,344 =    $.00023071


     It is obvious that the load on the circuit is an important
variable.  If the nailed up user is assigned costs at $.002 per
minute, the use of a $10 circuit is assigned an $87 cost.

     What do we know about Internet calls?  Various studies of
Internet usage by ILECS suggest the average holding times are
significantly longer than voice calls, but this is hardly
surprising and not very important.  More interesting is the
amount of time online, and the distribution of that time over the
day.  One interesting source of data is the ISP ratio of
customers to modems. When America Online got in trouble after it
introduced flat rate pricing, it only had 200,000 modems for 8
million customers, a ratio of 40 to 1.  Today America Online says
it has 700,000 modems for 12 million customers, which about 17 to
1.  One ISP I talked with said they provision residential
customers at 12 to 1.

     Most ISPs provide one modem for each 10 to 20 residential
dial-in customers.   ISPs which cater to businesses sometimes
provision at 8:1 or better.

     One interesting result from this very basic analysis is that
the ISPs provide less peak capacity than is already available on
the PSN for voice usage.  Consider the following:

                             Table 3
                          Peak Capacity

     AOL when flat rate announced       .025
     AOL today                          .059
     Independent ISP                    .083
     Residential Voice network          .124 to .14
     Business Voice network             .25


     We also looked a proprietary data from a number of ISP hunt
groups for a large ISP.  In general, ISPs want to have enough
lines to permit customers to connect to the service, and they
also want to economize on expenditures on lines.  For a large
hunt group, the "sweet spot" for an ISP is when the peak hour is
used at 70 to 93 percent of capacity.  Anything above 93 percent
and service is unacceptable for consumers (who have trouble
connecting).  We obtained the average number of hours for the
load on circuits from hunt groups.  The data were aggregated by
load.  The first is hunt groups which used 70 to 80 percent of
capacity in the peak hour, and the second is for hunt groups that
used 80 to 93 percent of capacity.  Using 75 and 86.5 percent
(the two midpoints for each group), we can calculate the percent
of the load in the peak hour.

                         Table 4
          Load on IPS Hunt Group Circuit

          Percent
          Peak                          Percent
          Capacity       Daily          of load
          Used           Load           in peak hour

          80 - 93        10.53 hrs      8.2%
          70 - 80        8.87 hrs       8.5%


     Here it is useful to recall that for a given circuit, the
ISP itself has a daily load of 8.87 to 10.53 hours.  A perfectly
even load would be 24 hours, and 4.167 percent of the load in the
peak hour.  Compared to a nailed up line, the ISP callers are
using the circuit 37 to 44 percent of the time.  In contrast, a
circuit that serves voice callers is used 12.8 to 16.7 percent of
the time.  This is a case of more is better, because the more
intensively the circuit is used, the less the average usage cost
is.


                             Table 5
                     Per Minute Usage costs

Residential @ 640                  $.0020
Residential @ 900                  $.0014
ISP @ 8.5% peak load               $.00062
ISP @ 8.2% peak load               $.00053
Nailed up (4.167 % peak load)      $.00023