The present stage you can characterise as initial stage of research work. In our opinion it will be necessary to use stochastic processes in the analysis of brand-switching.
Fundamentally, consumer typologies are procedures for reducing data, Rather than work with a multiplicity of individual reactions it is preferable to use this matrix as a basis for seeking essential correlations both between variables and between individuals. Our intention here is to deal exclusively with the individuals aspect and to discuss a type construction proposal which uses a new similarity index.
The degree to which operational research will be accepted as a management tool depends very largely upon whether or not operational researchers can sell their discipline and skill to management. So far all signs are that OR men have as yet failed to do so. In common with continental countries operational research in Britain has penetrated few organisations; the nationalised industries and a handful of multi-national companies employ a substantial number of operational researchers. The main conclusions from the Seminar are: 1. Operational research equates with a way of thinking; 2. Many people are in fact using OR techniques without realising it; 3. Management needs to be converted, not to accept the definition of OR but accept OR as an everyday tool particularly to evaluate alternative plans.
In Group 1 the discussion of brand switching was centred first on whether or not Markov chains were a useful concept in this area.
Group 1 spent sometime discussing the LP Model of Profit in the paper by Randall and Farmer, and eventually decided it was precisely what it said it was.
We understood this work essentially as an automatisation of the actual way of making budgets in a great international company. But the huge size of the model prevents a real back and forth movement of information between central management and decentralised units (feedback effects).
The usefulness of the Markov chains was of course discussed with the two speakers; they had actually adopted very different positions on that point.
The panel pointed out that there were programs available, at least in England, that produced clusters on a more satisfactory basis. The TYPOL method is an "average linkage" procedure based on an effectively random order of presentation rather than, even, the grouping of those maximum linkages first. Phase 3, however, permitted some tidying up. The main usefulness was seen as a relatively cheap method of producing descriptive clusters.
The difference between typology and segmentation was then discussed. The most popular opinion was that: 1. Segmentation implies the isolation of one of the variables as being the one to be predicted; 2. Such restraints are not needed when applying a method of typological analysis; this kind of descriptive analysis therefore permits a total flexibility without any a priori judgment. It remains however that a typology can lead in some cases to directly interpretable segmentations .
We do not envisage the use of linear programming as the speaker himself says that the model is not made to find an optimum solution, but rather the exploration of the consequences of different decisions, assumptions or events. Linear programming looks to us a heavy technique which needs some risky assumptions of linearity.
This paper was presented and seen as two parts: a broad and intriguing first part followed by a highly specific sub-model. The group was not concerned to make spontaneous suggestions as to the treatment of the open ended topics in the first part but devoted its attention to the 'Linear Programming Model of Profit'.
Several points were stressed in our discussion on the performance given by Mr. Randall and Mr. Farmer. The most essential ones are stated hereafter. In order to put the discussion on a clear base it was first explained that here it is a question of a mere OR model whose objective function, which is to maximise, comprises a range of relevant variables. In the following we discussed the problem whether or not it would be useful to reduce to a linear function although the methodical knowing and the knowledge of the market-interrelations is sufficient to work with a non-linear function.