Through my work, I have dealt with thousands of Excel, and for those that remember Lotus, financial models, and I have to say that most of them are wrong. I want to summarize five of my least favorite things about the “bad” models I have had the unfortunate experience to deal with during my career. All of these issues, except the last one, are sufficient for me to walk away from the work or opportunity as they indicate that I am likely to find more problems in the finance area or business.

Excessive precision.

When I see models with numbers in the millions and show accuracy to the dollar, or even worse, the cent, red flags fly. Just because Excel can calculate to ten plus decimal places, you don’t need to use it. That much precision is a distraction and is usually wrong, especially if it is a forecasting model. In the words of Niels Bohr, “Prediction is very difficult, especially about the future.” Thus models that predict amounts to the dollar a year away are wrong!

Ask yourself what level of accuracy is needed. If you are dealing in tens of millions and show numbers to the thousand, the error level is 0.01%, which is more than enough precision. Another test is, would you be willing to bet on the outcome being right to that level of accuracy, and if not, ask why you are showing it. Remember the adage, “I would rather be 90% right and imprecise than precise and 100% wrong.”

Hardcoded numbers.

Models that have hardcoded numbers in them also raise red flags. Users forget they are there, and they remain forever with no rhyme or reason, leading to wrong results. If things change, it is hard to find them and correct them in all the cells that need changing. Another issue with hard-coded numbers is where a correct model produces #N/A or #DIV0 results due to the values entered, and the model cannot adjust. To fix these, instead of using Excel logic statements, users change the formulas to exclude the offensive input. However, when the correct data is obtained and entered, everyone has forgotten about these changes. As a result, I have seen large companies use fatally flawed models in planning, but no one has realized it. Finally, requiring hardcoded numbers to change a model to get it to produce correct results could mean the model is fatally flawed in design and operation – time to start again.

No tracking of results.

Recently I met a firm that had an Excel model it had been using for all its forecasting for years. I inquired about how accurate it was, and the answer was that the last month had been a huge surprise, but no one had ever tracked its results against actual. If the model’s results are not measured, its effectiveness is unknown, and no improvements are made. Thus one could be relying on something wrong for years and not realize it. No model is perfect, and they are like an iterative process, use them, measure the results, and then adjust them to improve the results.

Understand the logic of the model.

Understand what the model is trying to accomplish. It is often good to diagram out how all the parts will fit together and where the different components will reside in the model. Also, think through all the pieces carefully about how they work and what they do. It is my experience that usually 6 – 10 items contribute to the majority of the values, and so they need the most precision; the rest will not change things much, and we don’t need to focus on those as much. Also, beware of some essential items, i.e., exchange rates.

Another Excel model I recently reviewed had no exchange rate assumptions even though the company had significant European operations – this oversite would lead to incorrect forecasting. Finally, remember all the parts! The complex financial model used business forecasting mentioned above showed P&L, Balance Sheet, and Cash Flow statements; however, the model didn’t distinguish between book and tax depreciation, leading to incorrect cash flow statements well as errors elsewhere.

Layout.

A model should be like a book or an essay – an introduction – the characters – the plot – the conclusion. I have seen everything mixed up, and you cannot follow the logic or flow, which makes it hard to read and understand. Make the model easy for the user to follow and read. Thus, I would recommend:

• A tab that explains what the model does and describes what each tab does as well.

• Color Coding. Use color-coding to distinguish between input, calculation, and output areas.

• Indentation. Use the cell indents for indentation, not another column

• Colors. Use colors to format types of cells. It makes it very easy to see what are inputs, assumptions, outputs, etc. If all input cells are yellow, then the user can quickly know what to change. Excel provides this through cell formatting.

• Data Validation. Use lists to control user inputs, which prevents mistakes from accidentally happening.

• Spell check. Excel provides it, so use it.

• Make it clean and easy to read.

• Tabs, Use more tabs with a purpose rather than one colossal tab that is difficult to navigate around. Break tabs down into Assumptions, Working, Results. These tabs make it much easier to follow.

• Assumptions must feed the model. All assumptions should feed into the appropriate parts of the model – just showing them on an Assumptions page is useless. Remove any cells that contain assumptions that don’t feed into anything or do calculations. Users will change them to no avail.

• Check cells. Use check cells to ensure that things are correct. If you are building a forecasting model, put in a line showing that the balance sheet balances. Thus if it doesn’t, you can easily see it rather than suddenly realize it when it is too late.

I hope that you find this useful. Good luck, forecasting.

© 2013 Marc Borrelli All Rights Reserved

Recent Posts