Marketing Gate
Professor John M. McCannFuqua School of Business
Duke University
The initial version of Marketing Gate is built in Goldworks, an AI programming environment from Gold Hill, Inc. To review, Marketing Gate contains ANALYZERS which know how to write paragraphs. The manager interacts with Marketing Gate by selecting paragraphs to be written. But, what is a paragraph in this context? What does it mean for a manager to "select the appropriate paragraph"?
To answer these questions, let us return to our hypothetical marketing manager. He knows that his brand, Munchkins, is not performing as expected and he must do an analysis and write a memo to his management about the problem. He turns to Marketing Gate and tells it that he needs to do an analysis, as opposed to performing a planning, execution, or control task. Since he does analysis in Marketing Gate by writing a report containing paragraphs, Marketing Gate presents him with the paragraphs it knows how to write.
This version of Marketing Gate knows how to write three paragraphs: a BDI-vs-CDI paragraph, a Promotional-Activity paragraph, and a Sales-Development paragraph. Marketing Gate presents these as paragraph "templates" which contain the structure of the paragraph in skeleton form. A template will be "filled in" after the manager has indicated the appropriate context.
The above screen shows the top part of the BDI-vs-CDI paragraph template, along with a small part of each of the other two paragraph templates. The manager can browse through the paragraph templates to learn the contents of each one. For example, he can examine the Promotional-Activity paragraph:
The manager can see from this paragraph template that it makes a statement about the amount of promotional activity in a market or a chain in terms of price cuts, feature activity, and display activity.
The manager believes that he should write the BDI-vs-CDI paragraph in order to learn about the regional nature of the business. The next step is to indicate the appropriate context by filling in the Brand, Location, Chain, and Weeks at the top of the paragraph template. This results in a "partially filed" paragraph template, as shown below.
The manager has indicated that he wants to have the paragraph written for Munchkins in all markets and for all chains for week 53. The next logical step would be to instruct Marketing Gate to write the paragraph, which would result in Marketing Gate completing (1) the graph in the middle of the paragraph and (2) the observations and conclusions section at the bottom of the paragraph. Since these sections contain incomplete sentences, Marketing Gate will complete each sentence by filling in the blanks.
Before instructing Marketing Gate to write the paragraph, the manager decides to explore the paragraph template to learn what values Marketing Gate can use to complete the observations and conclusions section. By putting the cursor on a blank and clicking the mouse, the manager obtains a list of the values which Marketing Gate can use to fill that blank.
Our manager sees that the first blank in the sentence
MUNCHKINS as well developed in as in all other markets
can be filled with "is" or "is not" to indicate that the brand is or is not well developed in that market. The manager could use the same procedure for exploring the template's other blanks.
Before writing this paragraph, the manager must fill in one other aspect of the context; he must fill in the markets of interest. Our manager fills in Washington, New York, and Atlanta and instructs Marketing Gate to write the paragraph.
At this point, the manager is interacting with the MMKS aspect of Marketing Gate. Now it is time for the MMKS to interact with the MMIS aspect of Marketing Gate and for the manager to wait for the paragraph to be written. The ANALYZER associated with the BDI-vs-CDI paragraph is brought into play because it knows how to get the MMIS to provide the appropriate data, how to reason about that data, and how to fill in the remaining blanks in the BDI-vs-CDI paragraph template. The ANALYZER first determines what information it needs by examining the context part of the paragraph template. Then it prepares the appropriate commands for the MMIS (these commands usually are nothing more than the name of the DATA VIEWER along with the appropriate parameters determined from the context section of the paragraph template). After the MMIS has executed the DATA VIEWER, the results are sent to Marketing Gate's ANALYZER, which makes the appropriate interpretations and writes the final paragraph, which is shown below.
From this paragraph, our manager learns that Munchkins is not well developed in Washington or in New York. And Marketing Gate has concluded that "the situation in Washington requires much attention."
Our manager wants to see a better graph of the supporting evidence than is available in the paragraph, and instructs Marketing Gate to produce a Lotus 1-2-3 graph of the data in the paragraph.
This expert system's ability to start a popular PC DOS program such as Lotus 1-2-3 and to send it data is an important feature of Marketing Gate. It allows marketing managers to employ the software they are accustomed to using.
The manager has written a paragraph and now he must decide what to do with it. He instructs Marketing Gate to "set disposition" of the report and obtains the following options:
Marketing Gate needs to know the importance of the observations and conclusions. If they are "Very important" then Marketing Gate knows that (1) it should remember the results for later use and (2) the paragraph should become part of the report. If they are "Not important yet" then Marketing Gate knows that the observations and conclusions should be remembered but that the paragraph should not be inserted into the report. If they are "Irrelevant" then Marketing Gate should remove these observations and conclusions from its memory. The latter would be the case if the manager had reason to know that the analysis or data were flawed and thus the results were incorrect.
Our manager selects Very important from the menu, which causes the paragraph to be placed in the word processor, which is Microsoft Word. At any time, the manager can go out of Marketing Gate and into Lotus or Word.
He can use Word to edit the paragraph or he can simply view the current status of the report.
Marketing Gate is an implementation of the Report Writing Assistant concept described in the last chapter. The manager tells it what paragraph to write for what brand/market/time situation, and Marketing Gate writes the paragraph. Further, it can place the paragraph into the report by inserting it into a word processor or desktop publishing system.
To accomplish its tasks, Marketing Gate interacts with the MMIS to obtain the data for the paragraph, with Lotus 1-2-3 to graph the data, and with the word processor. As a knowledge-based system, it must contain sufficient knowledge of each of these systems to produce the desired results. Finally, it must know how to reason with the results obtained from the MMIS.
In our BDI-vs-CDI paragraph, this reasoning knowledge is not very hard to imagine nor very complex. Basically the ANALYZER knows that a market is well developed if its Brand Development Index (BDI) in the market is greater than the Category Development Index (CDI). It knows how to calculate the ratio of BDI to CDI in each market and how to compare them with the same ratio for all markets taken together. If a market has a very low ratio, and it is lower than any other market, then it requires attention. Whether it requires "no," "some" or "much" attention depends upon the value of the ratio. In our case, the ratio was 0.33 (10/30) in Washington, thus causing Marketing Gate to conclude: "For MUNCHKINS, the situation in WASHINGTON requires much attention."
Marketing Gate might be very useful to a manager who writes many similar reports or who naturally thinks in terms of words rather than in terms of numbers and analysis. Marketing Gate lets that manager do analysis by writing rather than doing analysis by pushing around a lot of numbers. Marketing Gate allows managers to concentrate on what they want to say rather than on (1) how to say it, (2) how to do the analysis to support it, (3) how to get the data into the appropriate format, (4) how to use the MMIS, and (5) how to get the results into a report. This in itself might be enough to justify the development and implementation of systems like Marketing Gate. But, if it can do these things, it can do much more for one simple reason: Marketing Gate now knows what the manager has learned during the interaction with the system. It has stored in its memory the observations and conclusions in the BDI-vs-CDI report. It now knows that
The next sections will explore the implications of Marketing Gate knowing what the manager knows.
At this point, our hypothetical marketing manager has learned that he has a problem in the Washington market. The next logical step would be a further exploration to learn the reasons for the problem or to pinpoint further the source of the problem. This leads to the idea of an advisor which provides guidance or advice about the next analysis step.
Once I get going on a problem, I have to think through all of the analysis steps I need to take. I am pretty good at this, but sometimes I forget to do something, or I have trouble remembering how to get the computer to perform the necessary work. Sometimes I get so deep into a series of analyses that I loose track of all the things I have learned: I then have trouble identifying the next steps I should be taking. It would be nice if there were a friendly advisor looking over my shoulder as I do an analysis. This advisor could make suggestions about the next steps and even tell me when it thinks I have gathered enough evidence.
Existing marketing management information systems cannot perform an advisory function because they do not know what the manager has learned in the early stages of a session with the computer. They provide the information for making observations and drawing conclusions, but they do not know the observations and conclusions that resulted from the information. Since Marketing Gate does know these things, it is relatively easy to let it give advice about the next paragraph which the manager should write. It does this via its Suggestions menu.
In order to provide these suggestions, Marketing Gate must know about the linkages among paragraphs. For instance, knowing that a problem exists in a market could lead to one of three tasks. A manager could
This is, in fact, the type of advice provided by Marketing Gate:
Marketing Gate suggested five next steps, along with its degree of confidence in each of the suggested steps. At the top of the list, it has suggested that the next step is the Sales-Development paragraph, that it be written for Munchkins in the Washington market, and that all chains be examined. Further, it is 80% confident that this is the proper step, as evidenced by the "(0.8)" at the end of the first line. This suggestion could be based upon the following knowledge:
If a problem exists in a market, then there is 0.8 evidence that you should examine the chains within the market with the Sales-Development paragraph.
Marketing Gate's certainty that this is the next step is based upon the certainty that a problem exists in the market. Recall that Marketing Gate has concluded that Munchkins is not well developed in Washington and in New York. But it recommends examining Washington with 0.8 certainty and New York with 0.6 certainty. It does this because it is more certain that Washington requires attention than it is that New York requires attention. Therefore, Marketing Gate is basing its suggestions on knowledge of the following form:
If there is C evidence that a problem exists in a market, then there is C evidence that you should examine the chains within the market with the Sales-Development paragraph.
Marketing Gate also suggests that the manager examine the reasons for the problem by writing the Promotional-Activity paragraph, or that the manager end the analysis phase of the work and enter the planning phase. But, it is less certain of these activities because its advice is based upon the belief that one should pinpoint a problem before trying to understand its causes or fix it.
If the manager accepts Marketing Gate's suggestion about writing the Sales-Development paragraph for Washington, then the appropriate paragraph template appears on the screen.
Since Marketing Gate knows that the brand under study is Munchkins, that the problem market is Washington, and that the prior analysis has been performed for week 53, it goes ahead and customizes the new paragraph with these parameter values. It uses its chain database to fill-in the appropriate chains in the Washington market. Again, this is an example of applying the knowledge gained during a prior task in the current task.
The manager can see that this paragraph will tell him about Munchkin's performance in each chain in terms of the chain's market share (%ACV) and the percentage of the brand's Washington volume which moves through the chain (%CASES). At this point, he writes the report and learns that the problem is unique to Giant Food, which has 77% of the market for this category although only 50% of Munchins' volume moves through Giant stores.
This paragraph can be sent to the word processor where it will be merged at the bottom of the report.
The manager can ask for further suggestions, in which case he is advised that the source of the problem should be determined by running the Promotional-Activity report for Munchkins in Washington for Giant.
By accepting this advice, the manager gets an analysis of Giant's promotional activities for Munchins during the past year in terms of the level of price cuts, displays, and retail ad features.
Marketing Gate has concluded that promotional activity for Munchkins was low in Giant in Washington. It has observed that there was "no" price cut activity, and "some" feature and display activity during the past year. The manager can examine the basis for these observations and conclusions by examining the data in Lotus 1-2-3, where it is placed by Marketing Gate when the manager indicates that he wants to go to Lotus. For instance, the manager would obtain the following graph of display levels upon moving from Goldworks to Lotus:
Although Giant did provide two periods of display activity, Marketing Gate used its knowledge of the average display activity for this category to conclude that there was only "some" activity. It could have concluded that there was "no", "some", or "much" activity for each of the three causal factors of price cut, display, and feature.
This Promotional-Activity paragraph can be merged into the report, which will then contain three paragraphs. The first paragraph pinpoints Washington as the problem market, the second one pinpoints Giant as the problem chain, and the third paragraph concludes that the reason for the problem is the low level of promotional activity for Munchkins by Giant.
At this point, it would be logical for the manager to try to fix the problem by designing a marketing program which induces Giant to provide more promotional activity for Munchkins. When asked for its suggestions, Marketing Gate does in fact suggest that it is time to enter the planning phase. Marketing Gate is certain that this is the proper step because (1) the problem has been pinpointed at the chain level (the lowest level for which Marketing Gate has data) and (2) the cause of the problem has been determined.
Marketing Gate has also suggested other analysis steps, with lower levels of certainty, if the manager wants to pursue Munchkins' problem in the New York market.
In this section, the power of the "computer knowing what the manager knows" came into play. Because Marketing Gate has made observations and reached conclusions, it can provide advice to the manager about the appropriate next analysis. This leads to a system which not only knows how to perform different types of analyses, but also knows how to structure these analyses into a story which pinpoints a problem and determines the cause.
Since it knows about the problem, it can offer advice about the type of approach the manager should take in fixing the problem. In the example problem pursued in this chapter, Marketing Gate could provide advice about the type of marketing program which should be designed to increase the level of trade support by Giant for Munchkins. It could also assist the manager in designing this marketing program. The next section describes these features of Marketing Gate.
Previous sections have discussed the three main component's of Marketing Gate's architecture:
The MMKS is composed of three components, with the first being the ANALYZERS which analyze the information contained in the MMIS' DATA VIEWERS. These facilities are designed to improve the manager's ability to do analysis and build a story of the brand's performance. This is the detective part of marketing management. After this work is completed, the manager becomes a designer -- he designs marketing events (promotions, advertisements, etc. )and thus builds a marketing program for the brand. In this section, we describe the second MMKS component, the DESIGNER, a knowledge-based system which assists in the design of a marketing program.
The purpose of the Marketing Gate's DESIGNER component is evident in our continuing dialogue with the hypothetical marketing manager:
Now I have to plan, to design a marketing program. It would be nice if the system knew what I have just learned so that it could help me develop a plan to solve the problem. The system could recommend the type of marketing event I should be considering, and it could provide me with a computer environment for designing the marketing program. After I have designed an event, it could help me assess the market's reaction.
The initial version of Marketing Gate has one DESIGNER which assists a manager in the design of a trade promotion. In the Munchkins scenario, when the manager asks for suggestions, Marketing Gate recommends that a trade promotion be designed for the Washington market, with particular attention paid to the reaction of Giant Food to the trade deal. If the manager agrees with this advice, Marketing Gate brings up a trade promotion planning paragraph.
The context for this plan has been filled in by Marketing Gate according to the problem being solved. The context section has been completed to contain Munchkins in Washington for all chains in Week 53, and the graph includes the BDI and CDI figures which led to the initial recognition of the problem. This Marketing Gate screen was developed on the premise that the trade promotion event would be designed to correct the BDI/CDI problem. When the trade deal has been designed, the remaining sections of the graph will be updated to show the desired results of the deal, and the observations and conclusions sections will be filled in to indicate the dates and expected results of the deal.
Before these figures can be inserted into the planning screen, the deal must be designed. When the manager indicates that he wants to write this report, Marketing Gate takes him into DealMaker, a knowledge-based system for designing deals.
(See McCann and Gallagher (1990) for a complete description of DealMaker) By filling in this menu, the manager designs a trade deal; the system provides estimates of the performance of the deal in the various chains and markets. When the manager is satisfied with the deal, the system "remembers" the pertinent information about the deal so that it can be of further assistance during the execution and control phases of marketing management.
The result is a screen which indicates the expectations for the event, along with other pertinent information.
This screen shows that the trade deal's goal is to raise the BDI from its current value of 10 to 20, without impacting the CDI. The observations section shows that it is scheduled for week 57 and will run for two weeks. Further, it has been assumed that the promotion will be run without interference from competitive promotions. The conclusions section of the screen indicates that the promotion has been scheduled and, as far as the system knows at that point in time, the event is a complete success. (The system could, and perhaps should, have been developed to indicate that the success of the event is unknown until it has actually occured in the market.)
The last conclusion points to an additional feature of Marketing Gate: assistance in the control function of marketing management. Plans, programs, and events are monitored and evaluated in terms of meeting goals. The next section explores this feature.
The fun part of my job is doing marketing, designing marketing programs and events ... this is how I drive my brand. The worst part is monitoring the events ... determining how well things worked. The marketing research department specifies the types of analyses we can use to evaluate our programs and events. If the system knew what events I was planning and why I designed them, couldn't it monitor the market for me and write a report about the events' performance? I bet it could even track the events as they are happening.
Our hypothetical marketing manager is calling for yet another feature for Marketing Gate: the monitoring of events. This feature can be called a MONITOR. With its addition, we complete the specification of Marketing Gate's components.
A MONITOR is a knowledge-based system which (1) determines the appropriate method for monitoring an event, (2) monitors the event, and (3) reports the results. This section describes the MONITOR's operation by simulating the process a manager would use to monitor the trade deal. A manager would use the system intermittently to determine the status of the deal and get Marketing Gate's evaluation.
The MONITOR's screen looks identical to the DESIGNER's final screen, except that the title in the upper left corner indicates that it is a "control" screen.
At week 53, Marketing Gate knows that the event has been scheduled for week 57, and that it is still a complete success. If the manager were to use the MONITOR again during weeks 54-56, identical results would be obtained because the event has yet to be executed.
When the manager revisits the MONITOR during week 57, he learns that the event is occurring, and that the BDI has indeed jumped to its desired level of 20 and that Marketing Gate is calling the event a "complete success."
When the manager returns to the MONITOR the week after the promotion, he learns that the promotion is concluded. The BDI, however, has returned to its pre-promotion level of 10 and Marketing Gate has termed the promotion a complete failure because it did not solve the BDI problem.
This section has illustrated Marketing Gate's ability to monitor an event. Notice that it chose to evaluate the event in terms of its impact on BDI; it made this choice because of the sequence of analyses and processes which led to the event. Marketing Gate knew that the event was designed to correct the BDI problem, and it knew this because of the results obtained in the initial BDI-vs-CDI analysis which started the analysis process.
These MONITORS are developed to monitor and assess events, e.g., to monitor a trade deal. These are called EVENT MONITORS to indicate that they monitor events. Another type of MONITOR, an OBSERVATION MONITOR, will be included in future versions of Marketing Gate. An OBSERVATION MONITOR focuses on observations which Marketing Gate's ANALYZERS and DESIGNERS have made; it monitors the observations. For instance, the ANALYZER determined that Munchkins was well developed in Atlanta. Now that Marketing Gate has this belief, it could continue to monitor the Atlanta market and report any changes to the manager. Another option: since the design of the trade deal assumed that there would be no change in promotions by competitors during weeks 53 and 58, given appropriate data, this assumption could be monitored and the results reported to the manager, who could then take the appropriate action.
Marketing Gate has been designed to augment rather than replace an existing MMIS. An MMIS computerizes the process of viewing data; all the other aspects of marketing management are left to the manager. The manager must analyze the data, design marketing events, and monitor the markets. Without Marketing Gate, he must accomplish these tasks using his own intuition and cognitive skills; only the data viewing part of marketing is supported by the system. Although the MMIS does play a role in marketing decision making, it is so limited that it cannot truly be termed a decision support system. The phrase "information system" more accurately describes the role of today's marketing systems.
Marketing Gate provides a new architecture for supporting all aspects of marketing decision making. It contains DATA VIEWERS for extracting and viewing data, ANALYZERS for analyzing data, DESIGNERS for designing marketing events and programs, and MONITORS for monitoring events and programs. Finally, it includes an interface to a word processor for reporting the results of each of these activities.
Marketing Gate provides a new and different way to view data and to make statements about the data. It is different because the manager focuses on the statements rather than the data. Existing marketing management information systems provide a data-oriented interface which forces the manager to think about his problems and situations in terms of the data and its structure. The manager must utilize data-focused thought processes to view the appropriate data. After the appropriate data view has been obtained, the manager must use additional cognitive processes to make statements about the situation -- to use the data to make observations and draw conclusions. Marketing Gate is new in that it reverses these processes by providing a statement-focused environment. The manager concentrates on statements about the brands and markets, and the system focuses on the data. The system uses the data to write the statements in the manager's chosen context.
The focus on statements plays two major roles. First, it provides a different way to analyze data. The focus on statements allows the manager to concentrate on understanding the market rather than understanding the data and analysis methods. Therefore, Marketing Gate allows the "marketer to do marketing." The manager can concentrate on understanding and influencing markets rather than on the computer steps required to view and interpret the data.
Second, since the computer makes the statements, it knows the same things that the manager has learned from the data. The computer knew the facts first and presented them to the manager. Hence, the "computer knows what the manager knows." More accurately, the computer knows what it has told the manager. This allows the system to guide the manager through the possible analysis steps by suggesting the next ANALYZER to be run. This focus is also important when the manager moves from analyzing data to designing and monitoring marketing programs. Marketing Gate can use this knowledge to assist in the design of marketing programs. It can suggest types of events, simulate the impact of the events, display its knowledge about the markets, and present the impact of the event in the same terms as those used to identify the problem the event is supposed to solve.
In summary, Marketing Gate
With Marketing Gate, the manager can focus on results and not data, on marketing and not on analysis, because the system assists in all phases of the management process.
The document Marketing Gate's Knowledge discusses and illustrates processes for building a system like Marketing Gate. It introduces the notion of a thought template, maps a thought template into a sentence, and describes how these templates may be obtained by reverse engineering existing documents.