Logo Quantitative consulting

Dr. B oris V aillant - Quantitative Consulting

About us

About us

Quantitative Consulting for Business

In today's business life, quantitative methods are needed everywhere. The ability not only to efficiently organize but also to meaningfully interpret your data, and act upon it, is an important competitive advantage in the marketplace.

The new website is in the making and our new presentation will be available soon: [not yet]

Archive of the old Website (2006-2008)

Technical presentation (2008)

Adaptive discrete choice design

Adaptive discrete choice design - presentation held at UseR! 2008

We present a concept for truly adaptive discrete choice designs. Based on the Bayesian paradigm, our algorithm uses sequential Monte Carlo methods to update the posterior probability after each new answer and generate new product comparisons based on a variety of possible target measures. We provide results comparing different adaptive and fixed-design strategies from a simulation study performed in R. Our algorithm outperforms classical adaptive methods based on utility balance. Our method is consistently based on discrete choice theory and should therefore also lead to more valid results. [pdf presentation - long! (5MB)]

Short PDF articles (2006-2008)

Bundling and Upselling

[Marketing and Pricing] Tools for Bundling and Upselling

Bundles are everywhere. While one is usually inclined to think of the more notorious examples of software bundles or copiers and paper, essentially every product is a bundle of features and services. The art of bundling consists in knowing which parts to sell together and which to sell separately. Be careful! Bundling projects are just as common as unbundling (aka "add-on pricing") projects. [pdf presentation]

Calculating tariff scenarios

[Marketing and Pricing] Calculating tariff scenarios

Many industries (Telcos, Banks, Utilities) work with two-part tariffs, consisting of a fixed access fee and a usage component. While this allows to specifically target different customer segments (think of prepaid tariffs vs. flat fees in mobile telephony), it also makes the assessment of the effects of tariff changes very complex. [pdf presentation]

Internal benchmarking systems

[Business analytics] Internal benchmarking systems

The usefulness of transaction based benchmarking systems is often hampered by a lack of comparability. No two transactions are the same, specific conditions for the product, the client or the market seem to apply everytime you try to make a comparison. Our method consists in constructing fair reference lines taking into account all major influences. While our main examples are pricing-related the same methods can be applied for performance-, quality- or other measures. [pdf presentation]

Business plans with Bayes nets

[Strategy and Planning] Business plans with Bayes nets

Bayes nets give a systematic approach to creating business plans or to sizing markets. The methodology can not only be used when building a tool for your business plan, the graphical representation also simplifies the internal discussion of its structure and the hypotheses involved. [pdf presentation]


[Strategy and planning] Setting up new reports

Good reports are an essential ingredient for any efficient management organization. Many good frontends for datawarehouses exist, permitting flexible setup and distribution of reports. However, the identification of those areas needing special attention, the right choice of indicators to track or the definition of warning thresholds require massive analytical input. Read our [pdf presentation] to learn more.

Customer analytics

[Customer analytics] Customer analytics: Fitness club example

Analyze your customers' behavior to tailor your offer to their needs. In all those subscibtion type business situations, where your customers are known to you, a wealth of data can be used e.g. to predict and avoid churn. Typical examples are telecommunications, insurances or even fitness clubs. Have a look at our example here: [pdf presentation]

Short Online articles (2006-2008)


[Training] Dos and Don'ts in Excel

There's no denying it. As much as quantitative consultants would wish that dedicated statistics tools like R, Splus, SAS or business intelligence tools like BO and Cognos played a more prominent role, the program that is virtually ubiquitous when it comes to analyzing business data is Excel. Its large distribution base as part of the Office package, its ease of use and its flexibility make it the most popular analysis tool by far.

The advantages of Excel constitute at the same time its largest drawback. Excel does not prescribe how to store or format data, it does not distinguish between data management and data presentation and in general invites the user to be repetitive and uneconomical with his or her data.

If in a project data sources are given to me in Excel, I usually count in several extra days of work to bring the data into a usable format. Even if you are not working with a consultant: The necessity to share data and to regularly update it, the fact that the data may be useful in answering new questions not thought of before, and not least the fact that we all belong to a rather forgetful species, place a number of restrictions on the way how to Excel should be used.

The following list of recommendations is mainly concerned with how you as the user should use the program: be specific, be precise, be encompassing and be systematic, if you know BO or Access: use Excel like you use BO or Access. Some of the recommendations are concerned with Excel's built-in "intelligence". Here, the message is: Do not trust it.

One remark before we start: Many people use Excel to "build models" by chaining formulas, putting different functions of their models on different pages etc.. While I do believe that most of this modelling should better be done in a statistical programming language like R, the following "commandments" can only in part be applied in that case.

Dos and Don'ts in Excel I

Do not Do
Do not mix data and analysis. Do keep your data in one table and as close to "normal form" as possible. Use the pivot table function for analysis.
If you keep your data in a standardized form, you don't have to reinvent the wheel, every time you make an analysis. You can use standard tools like Pivot-tables, and are able to describe your data concisely.
Do not keep separate sheets of data (e.g. January, February, etc.). Do create a single sheet for your data.
Just imagine you've stored every month in another sheet and somebody asks you to compare Q3 to Q2! You will have to start linking formulas across sheets, but of course the data positions in the sheets will have moved etc. etc.
Do not keep any kind of formatting in your data. Do ensure that your data is completely unformatted by copying the whole data sheet and then using "Paste special: Values" into a completely new sheet
Formatting has a way to lead a life of its own. What was supposed to be a number becomes a text or a date etc. If you want to make sure there is no formatting left, use a completely new sheet.
Do not use linked cells, colors, decimal points or the like. Especially do not use formatting like colors, or bold type to code information Do make your color coded information explicit in a separate column. If you must by using values like "green", "blue" etc.
Have you ever tried to navigate through a sheet with linked cells with your cursor? Or to link a formula to a linked cell? Or to just copy data from an area with linked cells?
Never use the date format for dates. Do use a safe date format, e.g. 2006, 12, 24 in separate columns or 20061224 instead.
No matter how good it looks right now, it will not survive any move to a machine with another localization than yours.
Never use the currency format. Do keep track of currencies as pure text in a separate column.
Again, you won't be able control changes of localization.
Do not use tags or abbreviations that could be construed to be dates. Do make sure there are no ambiguities.
If, for instance, you should decide to label lines by "3.1", "3.2" etc., rest assured that Excel will start interpreting this as the 3rd of January etc.. Usually this happens when you least expect it and you may only find out much later that you have worked with corrupted data.
Do not keep formulas in your data sheet. If you need formulas to calculate some columns (e.g. revenue from pieces sold and price/piece): Do use formulas for the calculation, but paste back the values immediately afterwards. Keep a copy of the first few lines of your data sheet with formulas in a separate sheet, so that you can repeat the procedure, later.
In large data sets, formulas will simply kill your performance. They will also bug you, whenever you interchange columns or sort your data.In the worst case, your results may be wrong after sorting without your noticing it.
Do not hide information in formulas (e.g. by using constants: A1*2,33) Do make all information used in formulas explicit in separate columns
This is just one more instance of the principle that everything in your data should be explicit.
Do not use intensive (not summable) quantities in your data sheet, except for the calculation of extensive quantities Do use extensive (summable) quantities in your data sheet
I'll be somewhat less assertive on this requirement. The point is that you cannot sum quantities in EUR/Kg or the like. You might want to look at minimums or quantiles in a given segment, though. So, be sure what you do.
Do not use different spellings for names (e.g. Region, Product names etc.) Do make sure that names are always written exactly the same way.
It is a real pain when you have to search for all the sales with "Meier&Co", "Meier &Co." and "Meier & Co". A common cause for nonunique names are double spaces. Get rid of double spaces by using Replace: "  " (2 spaces) by " " (1 space). For "Meier&Co", along with all their possible spellings, try replacing ".", "&", " " by "" (nothing).
Do not rely on the "numeric" format. Do make sure that numbers are really numbers by multiplying them with 1 in a separate column and using this result instead of the original data.
What looks like a number in Excel may still not be recognized as a number. The formatting options usually only influence the display but not the data type itself.
Do not hide columns or lines in your data sheet Do introduce a separate column with 1 or 0 indicating, whether a specific line is appropriate for your purpose or not.
Hiding elements may make sense in presentation sheets, not in your data sheet.
Do not allow empty cells as values. Do make your intention explicit.
Did you mean 0, or "to be discussed", "does not apply" etc.?
If you are German and are keen to belong: do not pronounce the name of the program "Ecksl" Do try to pronounce it more like "Exelle"

So, now that you have put your data in a nice unformatted table, you can start making analyses using the pivot table function. Have a look at the [xls-example].

Contact Details

Dr. B oris V aillant - Quantitative consulting

B lumenhof   23, 53 119 B onn, G ermany
+49 176 64 11 65 94
i n f o @ quantitative-consulting.eu
USt-ID / VAT-ID: DE247366130

© 2006-2008   www.quantitative-consulting.eu