U.S. patent application number 10/139503 was filed with the patent office on 2003-01-16 for analytically determining revenue of internet companies using internet metrics.
Invention is credited to Dao, Fu-Tak, Martija, Ricardo, Spacek, Thomas, Weerahandi, Samaradasa.
Application Number | 20030014336 10/139503 |
Document ID | / |
Family ID | 26837288 |
Filed Date | 2003-01-16 |
United States Patent
Application |
20030014336 |
Kind Code |
A1 |
Dao, Fu-Tak ; et
al. |
January 16, 2003 |
Analytically determining revenue of internet companies using
internet metrics
Abstract
With respect to a current quarter of unreported revenue for
certain Internet companies, by processes performed by a computer
revenue to date is analytically determined and future revenue for
the remaining quarter is statistically projected by modeling
revenue based on "Internet metrics". Actual revenue performance is
obtained and one or more "Internet metrics" are measured for a
given Internet company. Using the revenue and measured Internet
metric data from prior quarters, a regression analysis is performed
in order to generate multiple models that reflect the relationship
between the Internet metrics and revenue. From these models, one is
selected that will most likely yield the best revenue estimates.
This resultant model and current Internet metric data are
subsequently used to estimate the company's revenue for the current
day, week, month, or quarter. These estimates are also used to
project the company's revenue for future days, weeks, months, and
quarters.
Inventors: |
Dao, Fu-Tak; (Bridgewater,
NJ) ; Martija, Ricardo; (East Brunswick, NJ) ;
Spacek, Thomas; (Albuquerque, NM) ; Weerahandi,
Samaradasa; (Pittstown, NJ) |
Correspondence
Address: |
Joseph Giordano, Esq.
Telcordia Technologies, Inc.
445 South Street 1G112R
Morristown
NJ
07960
US
|
Family ID: |
26837288 |
Appl. No.: |
10/139503 |
Filed: |
May 3, 2002 |
Related U.S. Patent Documents
|
|
|
|
|
|
Application
Number |
Filing Date |
Patent Number |
|
|
60288769 |
May 4, 2001 |
|
|
|
Current U.S.
Class: |
705/30 ;
705/347 |
Current CPC
Class: |
G06Q 30/0282 20130101;
G06Q 30/0201 20130101; G06Q 10/06375 20130101; G06Q 40/00 20130101;
G06Q 40/12 20131203; G06Q 40/02 20130101 |
Class at
Publication: |
705/30 ;
705/7 |
International
Class: |
G06F 017/60 |
Claims
We claim:
1. A method for estimating current revenue of a company, said
method comprising the steps performed by a computer of: obtaining
the company's actual quarterly revenue performance for a plurality
of prior quarters, obtaining data points for an Internet metric
over the plurality of prior quarters, wherein the Internet metric
is related to the company, generating a revenue model for the
company using the obtained actual quarterly revenue performance and
the obtained Internet metric data points, obtaining one or more
current data points for the Internet metric, and estimating the
company's current revenue by applying the obtained current Internet
metric data points to the generated revenue model.
2. The method of claim 1 wherein the obtained actual quarterly
revenue performance and the obtained Internet metric data points
for the plurality of past quarters are obtained for no more than
six prior quarters.
3. The method of claim 1 wherein the generated revenue model is a
quarterly revenue model.
4. The method of claim 3 further comprising the step of scaling the
generated quarterly revenue model to a weekly revenue model, and
wherein the obtained current Internet metric data points comprise
one week of data points, and wherein the estimated company's
current revenue is revenue for this one week.
5. The method of claim 1 wherein the generated revenue model is a
weekly revenue model, the obtained current Internet metric data
points comprise one week of data points, and wherein the estimated
company's current revenue is revenue for this one week, said method
further comprising the steps of: obtaining current data points for
the Internet metric over a plurality of weeks, estimating the
company's weekly revenue for each of the plurality of weeks, and
summing the plurality of estimated weekly revenues to obtain the
company's revenue for the current quarter.
6. The method of claim 1 wherein the generated revenue model is a
weekly revenue model and wherein the estimated company's current
revenue is for the current week, said method further comprising the
step of projecting revenue for a following week.
7. The method of claim 6 wherein the revenue projection is made
through a running average technique.
8. The method of claim 7 further comprising the steps of:
projecting revenue for all remaining weeks in a quarter beyond the
current week, and summing all revenue estimates and revenue
projections for a quarter to obtain a quarterly revenue
estimate.
9. The method of claim 1 wherein the Internet metric is a page-hits
metric, a visitors metric, a transactions metric, or a hosts
metric.
10. A method for estimating current revenue of a company, said
method comprising the steps performed by a computer of: obtaining
data points for at least two Internet metrics over a plurality of
past quarters, wherein said at least two Internet metrics are
related to the company, creating new Internet metric data points
over the plurality of past quarters by combining the data points
from the at least two Internet metrics, generating a revenue model
for the company by using the created new Internet metric data
points, creating one or more new Internet metric data points for
the current quarter by combining one or more current data points
obtained for each of the at least two Internet metrics, and
estimating the company's current revenue by applying the created
new Internet metric data points to the generated revenue model.
11. The method of claim 10 wherein the new Internet metric data
points for the past and current quarters are created through
standard sums.
12. The method of claim 11 wherein standard deviations are used as
a weighting factor in the standard sums.
13. The method of claim 10 wherein the generated revenue model is a
quarterly revenue model, the method further comprising the step of
scaling the generated quarterly revenue model to a weekly revenue
model, and wherein the created new Internet metric data points
comprise one week of data points, and wherein the estimated
company's current revenue is revenue for this one week.
14. A method for estimating current revenue of a company, said
method comprising the steps performed by a computer of: obtaining
data points for each of a plurality of Internet metrics over a
plurality of past quarters, wherein said Internet metrics are
related to the company, generating a plurality of revenue models
for the company by using the obtained data points for each of the
plurality of Internet metrics, choosing from the plurality of
revenue models a revenue model for estimating current revenue, and
estimating the company's current revenue by applying currently
obtained Internet metric data points to the chosen revenue
model.
15. The method of claim 14 wherein each of the plurality of
generated revenue models corresponds to one of the plurality of
Internet metrics.
16. The method of claim 15 further comprising the step of:
combining the past quarter of data points for each of the plurality
of Internet metrics to create one or more new Internet metrics and
corresponding data points, and wherein the plurality of generated
revenue models further includes a revenue model corresponding to
each of the new Internet metrics.
17. The method of claim 16 wherein the choosing step comprises the
steps of: computing r.sup.2 for each of the plurality of generated
models, and choosing the revenue model for estimating current
revenue based on the largest r.sup.2 value, where r.sup.2 is the
coefficient of determination.
18. The method of claim 17 wherein the choosing step further
comprises selecting the Internet metric or combined Internet metric
used to make a prior revenue estimation.
19. The method of claim 14 wherein each of the plurality of
Internet metrics is a page-hits metric, a visitors metric, a
transactions metric, or a hosts metric.
Description
RELATED APPLICATION
[0001] The present application claims the benefit of U.S.
Provisional Application No. 60/288,769 filed on May 4, 2001,
entitled "Methods for Analytically Determining Revenue of Internet
Companies Using Internet Metrics."
BACKGROUND OF OUR INVENTION
[0002] 1. Field of the Invention
[0003] Our invention relates to methods for analytically
determining the revenue of certain types of Internet companies.
More particularly, our invention relates to methods for using Web
based and equipment based metrics related to Internet companies for
analytically determining the current revenue and statistically
projecting the future revenue of these companies.
[0004] 2. Description of the Background
[0005] People are continuing to use the Internet as a medium for
communication, education, entertainment, information exchange,
electronic commerce (E-commerce), etc. Accordingly, new businesses
are emerging and businesses in virtually every sector of the
economy are using the Internet to provide new services and reach
new and existing customers more effectively and cheaply. In
particular, this invention relates to firms including pure
E-commerce companies, "click and mortar" companies, portals, and
Internet Service Providers (ISPs). Hereinafter, these types of
companies will be collectively referred to as "Internet" companies.
Although there are many other types of companies whose business
relates to the Internet, our focus is on the types of Internet
companies just listed.
[0006] The financial community typically does not become aware of
the revenues generated by "traditional" companies until several
weeks after the company quarters end, when revenue data is
announced. The same holds true for the above "Internet" companies.
Although past quarterly data is useful, the financial community
needs daily, weekly, and monthly information, as well as
projections to the end of the quarter, to aid in their everyday
decision-making. As such, there is a need by the financial
community to estimate and forecast the revenue performance of the
Internet sector. In addition to using information directly provided
by companies, financial institutions currently use fundamental and
technical analysis, such as revenue estimates based on number of
employees, past sales analysis, and trend analysis of past
revenues, to estimate and forecast revenue. However, given both the
rate at which the Internet in general is growing and the volatility
within the Internet sector, these estimation and forecast
techniques are proving to be inadequate. In addition, there is
always a need to make more accurate estimates on a more timely
basis.
SUMMARY OF OUR INVENTION
[0007] It is desirable to provide methods that overcome the
shortcomings of the prior art and more accurately estimate and
project, on a more timely basis, the economic performance of an
Internet company. Our invention satisfies these and other desires
by providing a method performed by a computer for estimating
current revenue and projecting future revenue of an Internet
company through Web based and equipment based metrics related to
that company.
[0008] Through experimentation and research, we have discovered
that certain physical events that occur at an Internet company's
Web environment and the amount of certain types of physical
equipment used by an Internet company are strongly correlated to
and predictive of the revenue generated by that company. We refer
to measures of these physical events and physical equipment as
"Internet metrics". Based on our discovery, we have invented
methods for estimating current revenue and projecting future
revenue of an Internet company, thereby overcoming the issues of
the prior art. Specifically, we have discovered that at least four
Internet metrics are highly correlated to the revenue generated by
Internet companies and when properly modeled, these metrics can be
used to estimate company revenue for the current day, week, month,
and quarter, and to project company revenue for future days, weeks,
months, and quarters.
[0009] The Internet metrics we determined to be predictive of
revenue include: the number of page hits at a company's Web site
("page-hits metric"), the number of visitors to a company's Web
site ("visitors metric"), the number of transactions conducted at a
company's Web site ("transactions metric"), and the number of
Internet hosts (i.e., IP addresses) supported by an Internet
Service Provider ("hosts metric"). A fifth metric, currently under
study to verify its correlative nature, is the "delay" within an
Internet company's web environment ("delay metric"), which is a
measure of how busy the servers, routers, and other equipment are.
Each of these metrics represents a numerical count relative to the
duration of time over which the metric is measured. As such,
"page-hits" represents the sum, over all visitors, of the number of
pages browsed by each visitor at an Internet company's Web site
over each measurement period.
[0010] "Visitors" represents the number of "unique" visitors to
visit an Internet company's Web site over each measurement period.
"Transactions" represents the number of physical transactions to
occur at an Internet company's Web site over each measurement
period. "Transactions" is based on "https requests" and is
currently measured by counting all https-requests that begin with
"https://". However, transaction counts can also be determined by
counting sub-fields of the "https://" requests. "Hosts" represents
the number of IP addresses supported by an ISP over each
measurement period. Although we have discovered that these metrics
are strong indicators of revenue, nothing in our invention
precludes the use of other Internet metrics to estimate revenue as
these metrics may arise as the Internet industry continues to
develop.
BRIEF DESCRIPTION OF THE DRAWINGS
[0011] FIG. 1 is a flow chart of a method for determining a revenue
model for an Internet company in accordance with the present
invention.
[0012] FIG. 2 is a high level block diagram of a computer, Internet
company Web site, and processors for collecting Web based and
equipment based metrics, on which computer can be implemented
methods for estimating current revenue and for projecting future
revenue of the Internet company based on the collected Web and
equipment based metrics in accordance with the present
invention.
[0013] FIG. 3 is a chart illustrating weekly revenue estimates for
an Internet company made using methods in accordance with the
present invention.
DETAILED DESCRIPTION
[0014] Our inventive method for estimating and projecting revenue
comprises three general steps, as shown by FIG. 1. In the first
step, 102, actual quarterly revenue performance is obtained and one
or more of the "Internet metrics" are measured for a given Internet
company and are maintained within an on-going computer database. In
some instances, only one metric is most relevant to a given company
and, in other instances, multiple metrics are most relevant. In the
second step, 104, the revenue and measured Internet metric data
from prior quarters are used to perform regression analyses in
order to generate multiple models that reflect the relationship
between the Internet metrics and revenue. From these models, one is
selected that will most likely yield the best revenue estimates. In
the last step, 106, the resultant model and current Internet metric
data are used to estimate the revenue for the current day, week,
month, and quarter, and to project the revenue for future days,
weeks, months, and quarters. Each of these steps is further
described below. The methods in accordance with embodiments of our
invention are executed by a computer. For example, as illustrated
in FIG. 2, software control for the method steps in accordance with
the present invention may be stored as software in memory 204 and
executed on processor 205 within computer 202.
[0015] The first general step of our invention, step 102, requires
the on-going collection of data points, these data points
constituting actual quarterly revenue performance and the Internet
metrics. The revenue data is readily available, as it is publicly
released following the end of a quarter. The Internet metrics are
not as easily obtained although various methods exist; however, no
one method is critical to our invention. In general, data points
relative to the page-hits, transactions, and visitors metrics are
obtained by examining the Web activity related to consumers
browsing an Internet company's Web site. Data points relative to
the hosts metric and delay metric are more difficult to obtain.
Several Internet collection methods are briefly described below,
which methods can be categorized as direct and indirect. Regardless
of the method of collection used, the results are ultimately stored
in a database, such as a database 206 in computer 202, and are
required for steps 104 and 106 of our invention, as described
below.
[0016] A first collection method is to obtain the data directly
from an Internet company. For example, Web servers typically log
access activity. These logs can be used to determine data points
for the transactions, visitors, and page-hits metrics. Hosts counts
can be obtained directly from an ISP's management systems. This
data can subsequently be uploaded to computer 202.
[0017] A second collection method is to indirectly obtain the
metrics, without the help of an Internet company, as seen in FIG.
2. One such method is to collect the page-hits, transactions, and
visitors metrics through random sampling. Under this method, a
population set is chosen and each member of this set agrees to have
his/her personal computer log all Internet activity relative to
particular Internet companies. These logs are then analyzed and the
results statistically adjusted to represent the public in general.
Several companies currently provide such services. A second method
to gather the page-hits, transactions, and visitors metrics is to
physically monitor a network and gather the data, discarding user
specific data. With respect to the delay metric, one method is to
transmit test packets to a companies web server(s) and measure the
response time. Sophisticated algorithms are applied to this
response time to eliminate time spent within the public Internet
network and to estimate how "busy" the equipment (routers, servers,
etc.) within the Web environment is. With respect to the hosts
metric, U.S. Pat. No. 6,178,451 B 1, "Computer Network Size Growth
Forecasting Method and System", by C. Huitema and S. Weerahandi,
describes a method for obtaining an ISP's hosts counts, the
teachings of which are incorporated herein by reference. Nothing in
our invention precludes using other methods for collecting
data.
[0018] Once collected, the revenue and Internet metric data points
are categorized into three general categories: (1) past actual
revenue performance, (2) past Internet metrics, and (3) current
Internet metrics. With respect to the terms "past" and "current",
"past" data is all data collected up through the most recently
reported quarterly revenue and "current" data is all data collected
since the most recently reported quarterly revenue. In accordance
with the methods of our invention, the past revenue and past
Internet metric data are used to generate a revenue model (step
104) that is subsequently applied to the current Internet metrics
to estimate and project revenue (step 106).
[0019] Because data collection is on going, at the close of a
quarter, the current data becomes past data and is subsequently
used to generate a model for the next quarter. However, our
research has shown that due to the volatility and rate at which the
Internet industry is changing, "past" data becomes less predictive
of revenue the "older" the past data becomes. As such, in
accordance with the methods of our invention, no more than six
quarters of past revenue and Internet metric data are used to
generate the next quarter's model. From a pictorial standpoint, a
"moving-window" is placed over the data and advanced by one quarter
at the end of each quarter. However, as the Internet industry
stabilizes, nothing in our invention precludes the widening or
narrowing of this window to include more or less past data in
generating models
[0020] As indicated, past revenue is collected on a quarterly basis
due to the methods of reporting. For the purposes of discussion,
these quarterly data points can be expressed as a data set as shown
in equation (1), where "m" represents the most recently reported
quarterly revenue and "n" represents the number of quarters over
which the regression analysis will be performed (as indicated, n is
currently set to 6).
{R}={ . . . , R.sub.(m-n), . . . , R.sub.(m-2), R.sub.(m-1),
R.sub.m} (1)
[0021] With respect to the Internet metrics, each metric represents
a count of a physical event or physical device. Currently, the
collection methods used by our invention measure these events and
devices on a weekly basis, although daily, monthly, or quarterly
counts can also be made depending on the method of collection. Not
all metrics apply to all companies and therefore not all counts are
performed for all companies. Assuming, for discussion purposes,
that all five metrics described above are collected for a given
company on a weekly basis, the set of weekly metric data points for
each metric can be expressed as equations (2-6)
{D.sub.trans}={ . . . , D.sub.(trans)(k-j), . . . ,
D.sub.(trans)(k-2), D.sub.(trans)(k-1), D.sub.(trans)k,
D.sub.(trans)(k+1), D.sub.(trans)(k+2), . . . } (2)
{D.sub.page-hits}={ . . . , D.sub.(page-hits)(k-j), . . . ,
D.sub.(page-hits)(k-2), D.sub.(page-hits)(k-1), D.sub.(page-hits)k,
D.sub.(page-hits)(k+1), D.sub.(page-hits)(k+2), . . . } (3)
{D.sub.visitors}={D.sub.(visitors)(k-j), . . . ,
D.sub.(visitors)(k-2), D.sub.(visitors)(k-1), D.sub.(visitors)k,
D.sub.(visitors)(k+1), D(visitors)(k+2), . . . } (4)
{D.sub.host}={ . . . , D.sub.(hosts)(k-j), . . . ,
D.sub.(hosts)(k-2), D.sub.(hosts)(k-1), D.sub.(hosts)k,
D.sub.(hosts)(k+1), D.sub.(hosts)(k+2), . . . } (5)
{D.sub.delay}={ . . . , D.sub.(delay)(k-j), . . . ,
D.sub.(delay)(k-2), D.sub.(delay)(k-1), D.sub.(delay)k,
D.sub.(delay)(k+1), D.sub.(delay)(k+2), . . . } (6)
[0022] where the k.sup.th data point is the last weekly measurement
made for the last quarter, the (k-j).sup.th data point is the
oldest past data point that will be used to determine the current
model, and the (k+1).sup.th, (k+2).sup.th, etc. data points are
weekly measurements for the current quarter.
[0023] In accordance with the methods of our invention,
determination of the revenue models in step 104 below requires that
the data points comprising the past Internet metrics be expressed
on the same scale as the revenue data. As a result, assuming again
that all five metrics are collected for a given company on a weekly
basis, the (k-j).sup.th to k.sup.th data points in equations
(2)-(6) must be combined and scaled to "quarterly" counts prior to
beginning step 104. The result is a new set of "Past" "quarterly"
metric data points and can be expressed as shown in equations
(7)-(11), where "m" represents the quarterly data point
corresponding to the most recently reported quarterly revenue and
"n" represents the number of quarters over which the regression
analysis will be performed.
{P.sub.trans}={ . . . , P.sub.(trans)(m-n), . . . ,
P.sub.(trans)(m-2), P.sub.(trans)(m-1), P.sub.(trans)m} (7)
{P.sub.(page-hits)}={ . . . , P.sub.(page-hits)(m-n), . . . ,
P.sub.(page-hits)(m-2), P.sub.(page-hits)(m-1), P.sub.(page-hits)m}
(8)
{P.sub.visitors}={ . . . , P.sub.(visitors)(m-n), . . . ,
P.sub.(visitors)(m-2), P.sub.(visitors)(m-1), P.sub.(visitors)m}
(9)
{P.sub.hosts}={ . . . , P.sub.(hosts)(m-n), . . . ,
P.sub.(hosts)(m-2), P.sub.(hosts)(m-1), P.sub.(hosts)m} (10)
{P.sub.delay}={ . . . , P.sub.(delay)(m-n), . . . ,
P.sub.(delay)(m-2), P.sub.(delay)(m-1), P.sub.(delay)m} (11)
[0024] With respect to estimating revenue using the current
Internet metrics, the revenue model resulting from the regression
analysis in step 104 is a quarterly model because the regression
analysis is performed on quarterly representations of the past data
points. As such, if a full quarter of current metric data has been
collected, this data can be combined and scaled to a quarterly
count to estimate the current quarterly revenue. However, in
accordance with the methods of our invention, the revenue model can
also be scaled to daily, weekly, and monthly revenue models and can
be used to estimate revenue for the current day, week, or month by
applying corresponding expressions of the current data. The use of
the revenue model is further described below in step 106.
[0025] Turning to the second general step of our invention, step
104, the revenue data set and past Internet data sets obtained for
a given company from the data collection step above are next
statistically analyzed to generate revenue models of this company.
Specifically, steps 104-A through 104-D illustrate the steps a
computer, for example computer 202 in FIG. 2, would perform to
generate revenue models of a given company and to select a given
model to ultimately estimate current revenue and project future
revenue. Methods in accordance with the present invention use
regression analysis techniques to generate and select this
model.
[0026] As indicated above, depending on the type of Internet
company, more than one type of Internet metric may apply. However,
it is not readily apparent which metric or whether a combination of
metrics will provide the "best" prediction of revenue. As such,
under methods consistent with our invention, a plurality of revenue
models using different combinations of the metric variables are
first generated and from these models the model most likely to
yield the best revenue estimate is determined based on statistical
characteristics, such as the coefficient of determination
("r.sup.2"). Specifically, revenue is first modeled with respect to
each metric independently and then modeled with respect to
combinations of metrics, resulting in a plurality of revenue
models. The model most likely to yield the "best" revenue estimate
is then determined and used to estimate current revenue and to
project future revenue.
[0027] For the purpose of discussion, the following discussion
assumes that transactions, page-hits, visitors, and hosts Internet
metrics apply to a given company to be analyzed. However, as
indicated above, only one or two metrics may be applicable to a
given company, in which case fewer models are generated. In
addition, the methods of our invention do not preclude the use of
additional metrics, as these metrics may evolve as the Internet
industry continues to mature. As such, additional models may be
generated.
[0028] Beginning with step 104-A, a plurality of revenue models is
first generated wherein each model uses either a single metric
variable or multiple metric variables, the latter models being
generated to determine if multiple metrics will have statistical
characteristics that will most likely yield a better estimate of
revenue than any one metric taken individually. Starting with step
104-A1, the individual metric models are first generated, where
each model has the form of the linear equation:
R=aM+b (12)
[0029] where "R" is the estimated revenue, "M" is the quarterly
Internet metric, and "a" and "b" are unknown coefficients. While
this model can change in the future as the nature of the E-Commerce
industry changes, the model in equation (12) has been shown to
provide an accurate fit between revenue and the Internet metrics.
The result of this first step is four models of the form:
R.sub.(trans)=(a.sub.(trans))(M.sub.(trans))+b.sub.(trans) (13)
R.sub.(page-hits)=(a.sub.(page-hits))(M.sub.(page-hits))+b.sub.(page-hits)
(14)
R.sub.(visitors)=(a.sub.(visitors))(M.sub.(visitors))+b.sub.(visitors)
(15)
R.sub.(hosts)=(a.sub.(hosts))(M.sub.(hosts))+b.sub.(host) (16)
[0030] In steps 104-A2 and 104-A3, the "multiple" Internet metric
revenue models are generated. (Note, as indicated above, this
discussion assumes that more than one Internet metric applies to a
given Internet company. If only one metric applies, steps 104-A2
and 104-A3 are never executed, step 104-A1 results in a single
model, and this model is subsequently used in step 106 below to
estimate and forecast revenue.) Our research has shown that the
Internet metrics may have a collinear relationship and as such, the
variables must be "combined" to address this issue. We chose to
combine the metrics using "standard sums", whereby the Internet
metrics are standardized using relative unit weights and then added
to create a new set of Internet metrics. Each new metric represents
a unique combination of the original Internet metrics. Note that
nothing in our invention precludes the use of other methods, such
as principal component analysis, to combine two or more metrics.
Using each new metric, revenue is again modeled multiple times
wherein each model has the form of the linear equation:
R=aM'+b (17)
[0031] where "R" is the estimated revenue, "M'" is the new Internet
metric, and "a" and "b" are unknown coefficients.
[0032] Beginning with step 104-A2, the new set of Internet metrics
is created through the "standard sums" technique using combinations
of two or more of the quarterly representations of the existing
Internet metrics. The result is a new set of metrics, each with a
corresponding set of past quarterly data points. Assuming the
presence of four metrics as above, eleven new metrics are created
as shown by Table 1, the first column showing the new Internet
metrics and the second column showing the constituent Internet
metrics that comprise each new metric.
1TABLE 1 Combined Internet Metrics New Internet Metric Component
Metrics 1 P.sub.(trans,page-hits) P.sub.(trans), P.sub.(page-hits)
2 P.sub.(trans,visitors) P.sub.(trans), P.sub.(visitors) 3
P.sub.(trans,hosts) P.sub.(trans), P.sub.(hosts) 4
P.sub.(page-hits,visitors) P.sub.(page-hits), P.sub.(visitors) 5
P.sub.(page-hits,hosts) P.sub.(page-hits), P.sub.(hosts) 6
P.sub.(visitors,hosts) P.sub.(visitors), P.sub.(hosts) 7
P.sub.(trans,page-hits,visitors) P.sub.(trans), P.sub.(page-hits),
P.sub.(visitors) 8 P.sub.(trans,page-hits,hosts) P.sub.(trans),
P.sub.(page-hits), P.sub.(hosts) 9 P.sub.(trans,visitors,hosts)
P.sub.(trans), P.sub.(visitors), P.sub.(hosts) 10
P.sub.(page-hits,visitors,hosts) P.sub.(page-hits),
P.sub.(visitors), P.sub.(hosts) 11 P.sub.(trans,page-hits,hosts)
P.sub.(trans), P.sub.(page-hits), P.sub.(visitors),
P.sub.(hosts)
[0033] Specifically, the "combining" of the metrics by use of
"standard sums" is performed by dividing each data point of the
constituent past quarterly Internet metric data sets by a weighting
factor and then "summing" corresponding data points (actually, only
the most recent "n" elements need be summed). The result is a new
metric and corresponding set of "n" past quarterly data points.
This procedure is shown below in equations (18), (24), and (28) for
the "P.sub.(trans,page-hits)", "P.sub.(trans,page-hits,visitors)",
and "P.sub.(trans,page-hits,visitors,- hosts)" metrics
respectively. The other eight metrics are similarly defined by
equations (19) to (23) and (25) to (27), not shown. 1 { P ( trans ,
page - hits ) } = { { P ( trans ) } W ( trans ) } + { { P ( page -
hits ) } W ( trans ) } = { ( P ( trans ) ( m - n ) W ( trans ) + P
( page - hits ) ( m - n ) W ( page - hits ) ) , , ( P ( trans ) m W
( trans ) + P ( page - hits ) m W ( page - hits ) ) } ( 18 ) { P (
trans , page - hits , visitors ) } = { { P ( trans ) } W ( trans )
} + { { P ( page - hits ) } W ( page - hits ) } + { { P ( visitors
) } W ( visitors ) } = { ( P ( trans ) ( m - n ) W ( trans ) + P (
page - hits ) ( m - n ) W ( page - hits ) + P ( visitors ) ( m - n
) W ( visitors ) ) , , ( P ( trans ) m W ( trans ) + P ( page -
hits ) m W ( page - hits ) + P ( visitors ) m W ( visitors ) ) } (
24 ) { P ( trans , page - hits , visitors , hosts ) } = { { P (
trans ) } W ( trans ) } + { { P ( page - hits ) } W ( page - hits )
} + { { P ( visitors ) } W ( visitors ) + { { P ( hosts ) } W (
hosts ) } } = { ( P ( trans ) ( m - n ) W ( trans ) + P ( page -
hits ) ( m - n ) W ( page - hits ) + P ( visitors ) ( m - n ) W (
visitors ) + P ( hosts ) ( m - n ) W ( hosts ) ) , , P ( trans ) m
W ( trans ) + P ( page - hits ) m W ( page - hits ) + P ( visitors
) m W ( visitors ) + P ( hosts ) m W ( hosts ) ) } ( 28 )
[0034] where "W.sub.(trans)", "W.sub.(visitors)", and
"W.sub.(page-hits)" are the weighting factors. Our invention
currently defines the weighting factor as the standard deviation of
each Internet metric data set , equations (7)-(10), over the "n"
most recent values. Our research has shown that the collinearity
between the metrics is adequately accounted for by using standard
deviation as the weighting factor. However, our invention does not
preclude the use of other weighting factors. The "W.sub.(trans)",
"W.sub.(visitors)" and "W.sub.(page-hits)" "W.sub.(hosts)"
weighting factors are shown in equations (29)-(32) below. 2 W (
trans ) = ( trans ) = i = m - n m ( P ( trans ) i ) 2 - ( n ) ( P (
trans ) _ ) 2 ( n - 1 ) ( 29 ) W ( page - hits ) = ( page - hits )
= i = m - n m ( P ( page - hits ) i ) 2 - ( n ) ( P ( page - hits )
_ ) 2 ( n - 1 ) ( 30 ) W ( visitors ) = ( visitors ) = i = m - n m
( P ( visitors ) i ) 2 - ( n ) ( P ( visitors ) _ ) 2 ( n - 1 ) (
31 ) W ( hosts ) = ( hosts ) = i = m - n m ( P ( hosts ) i ) 2 - (
n ) ( P ( hosts ) _ ) 2 ( n - 1 ) ( 32 )
[0035] where "{overscore (P.sub.(trans))}", {overscore
(P.sub.(page-hits))}, {overscore (P.sub.(visitors))}, and
{overscore (P.sub.(hosts))} are the average transactions,
page-hits, visitors, and hosts metric values as computed over the
"n" most recent data elements in equations (7)-(10),
respectively.
[0036] In step 104-A3, the eleven new metrics are each modeled as a
linear equation resulting in eleven additional models, three of
which, "R.sub.(trans,page-hits)",
"R.sub.(trans,page-hits,visitors)", and
"R.sub.(trans,page-hits,visitors, hosts)" are shown below in
equations (33), (39), and (43). The remaining eight equations are
similarly defined by equations (34) to (38) and (40) to (42), not
shown.
R.sub.(trans,page-hits)=(a.sub.(trans,page-hits))(M.sub.(trans,page-hits))-
+b.sub.(trans,page-hits) (33)
R.sub.(trans,page-hits, visitors)=(a.sub.(trans,page-hits,
visitors))(M.sub.(trans,page-hits,
visitors))+b.sub.(trans,page-hits, visitors) (39)
R.sub.(trans,page-hits, visitors,hosts)=(a.sub.(trans,page-hits,
visitors,hosts))(M.sub.(trans,page-hits,
visitors,hosts))+b.sub.(trans,pa- ge-hits, visitors,hosts) (43)
[0037] In step 104-B, the "least squares line" or "regression line"
is determined for each individual and multiple metric model,
(13)-(16) and (33)-(43), by determining the least squares estimate
for each of the model coefficients: "a.sub.(trans)",
"b.sub.(trans)", "a.sub.(trans,page-hits)",
"b.sub.(trans,page-hits)","a.sub.(trans,page-h- its, visitors)",
etc. Using the revenue data set equation (1), the past individual
metric data sets equations (7)-(10), and the new combined metric
data sets equations (18)-(28), the least squares estimate of each
coefficient is determined, as shown in equations (44)-(75) for the
"a.sub.(trans)", "b.sub.(trans)", "a.sub.(trans, page-hits)",
"b.sub.(trans, page-hits)", "a.sub.(trans, page-hits, visitors,
hosts)","b.sub.(trans, page-hits, visitors, hosts)" coefficients.
The least squares estimate equations for the remaining twenty-four
coefficients are similarly defined by equations (46) to (51) and
(54) to (73), not shown 3 a ^ ( trans ) = i = m - n m ( P ( trans )
i ) ( R i ) - ( n ) ( P trans _ ) ( R _ ) i = m - n m ( P ( trans )
i ) 2 - ( n ) ( P ( trans ) _ ) 2 ( 44 )
{circumflex over (b)}.sub.(trans)=({overscore
(R)})-(.sub.(trans))({oversc- ore (P.sub.(trans))}) (45)
[0038] 4 a ^ ( trans , page - hits ) = i = m - n m ( P ( trans ,
page - hits ) i ) ( R i ) - ( n ) ( P ( trans , page - hits ) _ ) (
R _ ) i = m - n m ( P ( trans , page - hits ) i ) 2 - ( n ) ( P (
trans , page - hits ) ) 2 ( 52 )
{circumflex over (b)}.sub.(trans, page-hits)=({overscore
(R)})-(.sub.(trans, page-hits))({overscore (P.sub.(trans,
page-hits))}) (53)
[0039] 5 a ^ ( trans , page - hits , visitors , hosts ) = i = m - n
m ( P ( trans , page - hits , visitors , hosts ) i ) ( R i ) - ( n
) ( P ( trans , page - hits , visitors , hosts ) _ ) ( R _ ) i = m
- n m ( P ( trans , page - hits , visitors , hosts ) i ) 2 - ( n )
( P ( trans , page - hits , visitors , hosts ) _ ) 2 ( 74 )
{circumflex over (b)}.sub.(trans, page-hits, visitors,
hosts)=({overscore (R)})-(.sub.(trans, page-hits, visitors,
hosts))({overscore (P.sub.(trans, page-hits, visitors, hosts))})
(75)
[0040] where "n" is the number of past data points in the revenue
and metric data sets deemed to be predictive of the current revenue
(as indicated above, n=6 quarters is currently used),
"P.sub.(trans),""P.sub.- (trans, page-hits)", "P.sub.(trans,
page-hits, visitors, hosts)", etc. are data points from the
original and new metric data sets equations (7)-(10) and (18)-(28),
"{overscore (R)}" is data points from the revenue data set equation
(1), "{overscore (P.sub.(trans))}", "{overscore (P.sub.(trans,
page-hits))}", "{overscore (P.sub.(trans, page-hits, visitors,
hosts))}", etc. are the "average-metric-value" of each original/new
metric data set as computed over the "n" most recent data elements
in equations (7)-(10) and (18)-(28), and "{overscore (R)}" is the
"average revenue value" as computed over the "n" most recent data
elements in equation (1). The result of step 104-B is fifteen
revenue estimation equations as shown in equations (76)-(90) ((81)
to (85) and (87) to (89) not being shown) and represented by Models
150 in FIG. 1.
{circumflex over
(R)}.sub.(trans)=(.sub.(trans))(M.sub.(trans))+{circumfle- x over
(b)}.sub.(trans) (76)
{circumflex over
(R)}.sub.(page-hits)=(.sub.(page-hits))(M.sub.(page-hits)-
)+{circumflex over (b)}.sub.(page-hits) (77)
{circumflex over
(R)}.sub.(vistors)=(.sub.(visitors))(M.sub.(visitors))+{c-
ircumflex over (b)}.sub.(visitors) (78)
{circumflex over
(R)}.sub.(hosts)=(.sub.(hosts))(M.sub.(hosts))+{circumfle- x over
(b)}.sub.(hosts) (79)
{circumflex over (R)}.sub.(trans, page-hits)=(.sub.(trans,
page-hits))(M.sub.(trans, page-hits))+{circumflex over
(b)}.sub.(trans, page-hits) (80)
{circumflex over (R)}.sub.(trans, page-hits,
visitors)=(.sub.(trans, page-hits, visitors))(M.sub.(trans,
page-hits, visitors))+{circumflex over (b)}.sub.(trans.page-hits,
visitors) (86)
{circumflex over (R)}.sub.(trans, page-hits, visitors,
hosts)=(.sub.(trans, page-hits, visitors, hosts))(M.sub.(trans,
page-hits, visitors, hosts))+{circumflex over
(b)}.sub.(trans.page-hits, visitors, hosts) (90)
[0041] Each of these equations can be used to estimate the current
quarter's revenue if quarterly representations of the combined
metrics are available (i.e., a full quarter of data points have
been collected).
[0042] As indicated, the completion of step 104-B results in a
plurality of individual metric and multiple metric revenue models,
as shown by the equations above. The next step is to determine
which of these models has the statistical properties to likely be
the "best" estimator of current revenue. Different methods exist in
the art for determining how well a "least squares equation"
performs. One method used by our invention is to compute the
"coefficient of determination", also called "r.sup.2", for the
equation, although nothing precludes the use of other methods. The
coefficient of determination for any least squares equation ranges
in value between "0" and "1" with "0" indicating a weak model fit
and "1" indicating a strong model fit.
[0043] Beginning with step 104-C, the coefficient of determination
is computed for each of the determined metric models, equations
(76)-(90). The model with the largest resultant "value" is then
chosen, in step 104-D, as the model to estimate current revenue.
The equations to compute the coefficient of determination for
"{circumflex over (R)}.sub.(trans)", "{circumflex over
(R)}.sub.(trans, page-hits)", "{circumflex over (R)}.sub.(trans,
page-hits, visitors)", and "{circumflex over (R)}.sub.(trans,
page-hits, visitors, hosts)" are shown below in equations (91),
(95), (101), and (105). The remaining eleven equations, (92) to
(94), (96) to (100), and (102) to (104), are similarly defined. 6 r
( trans ) 2 = 1 - i = m - n m ( R i - R ^ ( trans ) i ) 2 i = m - n
m ( R i - R _ ) 2 ( 91 ) r ( trans , page - hits ) 2 = 1 - i = m -
n m ( R i - R ^ ( trans , page - hits ) i ) 2 i = m - n m ( R i - R
_ ) 2 ( 95 ) r ( trans , page - hits , visitors ) 2 = 1 - i = m - n
m ( R i - R ^ ( trans , page - hits , visitors ) i ) 2 i = m - n m
( R i - R _ ) 2 ( 101 ) r ( trans , page - hits , visitors , hosts
) 2 = 1 - i = m - n m ( R i - R ^ ( trans , page - hits , visitors
, hosts ) i ) 2 i = m - n m ( R i - R _ ) 2 ( 105 )
[0044] where "{circumflex over (R)}.sub.(trans)," is the estimated
revenue using the "n" most recent data points from the transactions
metric data set equation (7), "{circumflex over (R)}.sub.(trans,
page-hits)" is the estimated revenue using the "n" most recent data
points from the combined transaction/page-hits metric data set
equation (18), etc.
[0045] In step 104-D, the model with largest coefficient of
determination is chosen as the model that will most likely provide
the best estimate of current revenue. This model, for discussion
purposes, will be referred to as:
{circumflex over (R)}=M'+{circumflex over (b)} (106)
[0046] where "" and "{circumflex over (b)}" are the "a" and "b"
least squares estimate coefficients from the chosen model, and "M'"
is the metric (either single or multiple) of the chosen model. In
another embodiment of our invention, the individual or multiple
metric used to make prior revenue estimates is also considered in
step 104-D when choosing the present model.
[0047] Turning to the third general step of our invention, step
106, equation (106) can now be used to estimate current revenue and
to statistically project future revenue of the modeled Internet
company. Revenue estimation will first be described followed by
revenue projection.
[0048] Equation (106) can be used to estimate a company's current
revenue over a given period of time based on current measurements
of the M' metric. M' is either an individual or multiple metric.
Assume first that M' is an individual metric. As indicated above,
the collection methods currently used by our invention measure the
metrics on a weekly basis, although daily, monthly, and quarterly
measurements can also be made. Assuming weekly measurements are
made, the (k+1).sup.th, (k+2).sup.th, etc. data points from the
metric data sets, equations (2)-(6), can now be used to estimate
revenue. Specifically, if a full quarter of weekly measurements
have been made (e.g., thirteen measurements), the resultant data
points can be combined and scaled to a quarterly count and
substituted for M' in equation (106) to estimate revenue for the
current quarter. However, a more useful application of our
invention is to estimate revenue as soon as possible. As such,
under methods consistent with our invention, equation (106) can be
scaled to estimate revenue for the current week and month, as shown
by equations (107) and (108), respectively, were "x.sub.week" is
the number of weeks in the quarter and "x.sub.month" is the number
of months in the quarter. 7 R ^ week = a ^ x week M ' + b ^ x week
( 107 ) R ^ month = a ^ x month M ' + b ^ x month . ( 108 )
[0049] Hence, using equation (107), the weekly metric data points
can be used to estimate revenue on a week-by-week basis. By
combining and scaling the weekly data points to monthly counts,
equation (108) can be used to estimate revenue on a monthly basis.
Similarly, if the metric is measured on a daily basis, equation
(106) can be scaled to estimate daily revenue.
[0050] Assume next that M' is a multiple metric and, for discussion
purposes, is a combination of the "transactions" and "page-hits"
metrics. Similar to the individual metric, the combined metric can
be used to estimate revenue for the current week, month, and
quarter through equations (106), (107), and (108). However, similar
to step 104-A2 above, the (k+1).sup.th, (k+2).sup.th, etc. data
points of the "{D.sub.trans}" and "{D.sub.page-hits}" data sets,
equations (2) and (3), cannot be applied to the revenue equations
until these data points are combined using principals similar to
equation (18) (i.e., standard sums).
[0051] As such, under methods consistent with our invention, the
"{D.sub.trans}" and "{D.sub.page-hits}" data sets are first
individually expressed as quarterly data points, monthly data
points, or maintained as weekly data points, depending on the
desired estimate, using all data points from the (k-j).sup.th
through the current measurement. Next, using the "standard sum"
principals set forth in step 104-A2, a weighting factor is
determined for each of the resultant data sets using the standard
deviation of these data sets. Finally, the data points of the
resultant data sets are weighted and corresponding points are
summed, resulting in a new combined data set that can be applied to
equations (106), (107), and (108) to estimate current revenue.
[0052] For example, using equation (107) to estimate revenue for
the current week, equations (109) and (110) show the weekly
weighting factors for the "{D.sub.trans}" and "{D.sub.page-hits}"
data sets 8 W ( trans ) = ( trans ) = i = k - j k + g ( D ( trans )
i ) 2 - ( j + g ) ( D ( trans ) _ ) 2 ( j + g - 1 ) ( 109 ) W (
page - hits ) = ( page - hits ) = i = k - j k + g ( D ( page - hits
) i ) 2 - ( j + g ) ( D ( page - hits ) _ ) 2 ( j + g - 1 ) ( 110
)
[0053] where (k+g) is the most current weekly data point, and
"{overscore (D.sub.(trans))}" and "{overscore (D.sub.(page-hits))}"
are average weekly metric values over the (k-j).sup.th to
(k+g).sup.th data elements. Using these weighting factors, the
resultant combined data set is shown in equation (111), the data
points of which can be used with equation (107) to estimate the
revenue for each week. 9 { C ( trans , page - hits ) } = { { D (
trans ) } W ( trans ) } + { { D ( page - hits ) } W ( trans ) } = {
( D ( trans ) ( k - j ) W ( trans ) + D ( page - hits ) ( k - j ) W
( page - hits ) ) , , ( D ( trans ) ( k + g ) W ( trans ) + D (
page - hits ) ( k + g ) W ( page - hits ) ) } ( 111 )
[0054] In accordance with the methods of our invention, FIG. 3
shows weekly revenue estimates and actual quarterly revenue results
for four quarters for an Internet company. The thirteen points
comprising quarters 302, 304, 306, and 308 represent weekly revenue
estimates using the methods of our invention. Once a full quarter
of revenue estimates are made, the resultant values can be summed
to estimate the quarterly revenue, prior to the actual revenue
being reported. (As reference, bars 320, 322, 324, and 326
represent actual reported quarterly revenue.)
[0055] In addition to estimating revenue for the current day, week,
or month, it is also useful, in advance of the availability of
metric data for future time periods, to project revenue for the
remaining days, weeks, or months of the quarter and to subsequently
use the estimated and projected values to project the quarterly
revenue as a whole. For example, the five points comprising quarter
310 represent weekly revenue estimates for the current quarter.
These points can be used to statistically project the revenue for
each of the remaining eight weeks comprising the quarter.
Subsequently, the five revenue estimates and the eight revenue
projections can be summed to project the quarterly revenue, as
shown by bar 328. Under methods consistent with our invention, a
"running average technique" is used to make these forecasts thereby
capturing the trend of the revenue estimates, however nothing
precludes the use of other projection techniques.
[0056] More explicitly, the running average technique is used to
statistically project the revenue for the remaining eight weeks of
quarter 310 as follows. The sixth weekly revenue point is projected
by averaging the prior N weekly revenue points beginning with the
fifth week. The seventh weekly revenue point is projected by
averaging the prior N weekly revenue points beginning with the
sixth week (i.e., the projected sixth week is used to project the
seventh week). This method is continued until the thirteenth weekly
revenue point is projected by averaging the prior N weekly revenue
points beginning with the twelfth week. The quarterly revenue is
then projected by summing all thirteen weeks, this projection being
represented by bar 328 in FIG. 3. This embodiment of our invention
currently uses N=6 although other values can be used.
[0057] The above-described embodiment of our invention is intended
to be illustrative only. Numerous other embodiments may be devised
by those skilled in the art without departing from the spirit and
scope of our invention.
* * * * *