Building models to aid decision-making

Written by: Jonathan Briggs

February 9, 2005 [3943 views]

A session as a guest lecture for Kingston University’s Decision Making module at Level 3 in Science and Level 2 in Computing and Information Systems. It will introduce some of the concepts of using quantitative and qualitative models to help in decision-making. Because of my background in Electronic Commerce, many of the examples will be drawn from retail.

Consider data produced from a two aspects of an online shop

1. Sales data broken down by time, date, search engine referrer, order value
2. Comments and reviews posted to via feedback form or on a forum

The first provides quantitative data, the second qualitative. Both can for the basis for models that may help us take decisions about how the store, its products, its prices or its responses to customers might be improved.

Why create models?

• Allow us to create a “mental image” of what is going on
• Allow us to communicate core ideas; “the essentials of how a problem or system might operate”
• Allow us to predict (forecast) what might happen in the future
• Allow us to try and isolate key levers that we could try and influence or change?
• Allow us to see the interconnections in a system
• Allow us to play around and experiment with ideas,“what if we did this…?”
• Help us to solve problems that occur (a drop in sales, increase in unhappy customers)

What can we use to create models?

1. Formula: x+3y = z, financial models, statistical models
2. Pictures: rich pictures
3. Diagrams: charts, causal loop models good introduction to Causal Loop Models
4. Rules: if this then that
5. Simulations: Sim City, queuing simulations

Features of models

1. They approximate reality through representations of essential components
2. They can produce false results
3. They cannot fully represent the complexity of most processes
4. They leave out external variables that cannot be controlled
5. Different people will produce different models

Consider the following problems?

1. Do news reports of custom’s drug seizures mean that more or less drugs are being imported?
2. Does increasing interest rates reduce rises in house prices?
3. Is global warming real?
4. How many people will buy 3G phones in the next 12 months?

Modelling in e-commerce
Try and isolate the key ideas in order to understand what sorts of things could be changed in order to try and improve revenue.

1. Prices – how does price effect the number of sales
2. Shipping costs – would it be better to include shipping in the item cost?
3. Multiple purchase discounts – are people affected by Buy 2 Get 1 Free?

Some things are hard to include in the model and should be left outside or fixed as constants. We might want to check the validity (sensitivity) of the model by altering these variables. Of course in other models these could be the things we try and change keeping our e-commerce store fixed.

• Competitors
• Effects of government regulation or tax changes

It is vital that we select the right sorts of indicators that will help us spot trends in our data:

• Order value and trends in orders
• How the customer found the store
• Sales per day/week/month
• Top selling items
• Poorly selling items
• Most effective advertising
• Most effective internal promotions

Patterns in data
Structuring data and spotting patterns in data helps, and forms the basis for future developments in Information Systems.

• Companies with internal data standards can maximise the value of information (if they know what they are looking for)
• Companies can inter-operate with their suppliers, industry organisations and markets if they use standard data interchange formats (models)
• XML is increasingly being used to build communication languages between companies XML Tutorial
• Companies are attempting to mine their legacy data to spot patterns and find prospects and repeat business A commercial introduction to data mining
• Researchers are working on improvements to the model behind the web to help make mining of web data easier. The ‘semantic web’ would allow more meaning to be added to web data. useful starting point for exploring research into the semantic web

Recent comments:

What do you think?

Add your comments