Your Financial Analysts and FP&A are anything but Analysts and That's a Problem

Let's just get right to the point.  The vast majority of your "analysts" are not anything more than process maintainers.  This is the reason you can't get any actual models constructed from scratch or are stuck with models nicknamed "the beast" comprising sheets and sheets of formulas that are no longer used, but no one understands how to clean it out.  This is also the reason all your budgets come from Three Statement financials and all your variance "analysis" is little more than the difference between budget and actuals.  As we move forward this will only get worse given the usage of AI as the crutch for lack of aptitude, knowledge and general ability.  As I predicted, there will be a reliance on professional certifications and proctored exams proving the passage and competence of knowledge for hiring in the very very near future.

       Data Analytics

• Descriptive
• Diagnostic
• Prescriptive
• Predictive

Analysis is the ability to reason and make conclusions from events and data.  This analysis can be descriptive, the most basic of data analysis which is the best most companies can get out of their "analysts".  Descriptive analysis is simply telling you what a set of data indicates about a given topic.  In the case of a business, the data is most often financial.  You may get a set of P&Ls or Balance Sheets, from which the basics are told to you of rising or tightening margins, increased expenses, or changing asset and liability positions.  You have sat through these meetings covering financial accounting ratios going up or down.  Well ... You could read this yourself.  If you know a ratio is supposed to go up to be considered positive and its not (or vice versa) then something needs to be done. ​

This type of analysis can be done by anyone with a basic understanding of financial statements.  It doesn't take further abilities in data science, economics, mathematics, statistics or operational research.  In fact, it really doesn't even take a great depth of knowledge in accounting or finance which is why you find yourself lacking in the area of analytics and strategy, but we will get into this further on.  What you will see is that each level takes an added skillset that is lacking in most and is not found in "years of experience" because it is only found in further study.

Since this is all most companies see, most execs probably are unaware of the levels of analytics and how those you have as process maintainers will never get you deeper than Diagnostic.  Don't believe me, look and see how many are crying out for Prescriptive (strategic insights) and Predictive (forecasting).  Fact is that since prescriptive and predictive require a depth of statistics and calculus understanding and ability, something that only a handful of candidates  actually possess regardless of the "years of experience". 

Unfortunately, this also means that most in an organization, also lacking math skills,  do not have the ability to tell if the "analysts" are wrong in their assumptions or usage of various techniques ... and yes, this is important.  So ... lets look at analytics and what you are lacking.

A step deeper than descriptive analysis is diagnostic analysis.  This might be called a root cause evaluation of the descriptive data.  It is the why something changed for the positive or negative.  When you call for a root cause or variance analysis, you are asking for a diagnostic evaluation.  The problem often comes from the fact that root causes are the drivers that affect change in those financial accounting statements.  These are the operational activities and accounting measurements, whether labor cost, labor time, material cost, material amount, freight costs, product purchase and inventory pricing, internal and external fail costs, costing methods such as standard costing vs normal costing and so on. 

Operational activity is generally what would be deemed, Big Data.  The operational data is then used, translated, to an accounting number.  Just consider the usage of standard costing for the pricing of inventory which reflects a "standard cost" to cost labor, total time of labor per product (or project), cost material, and total material usage per product (or project).  How far do you think an analyst would get in the data without understanding accounting and the granular level of operational activity that makes up those accounting entries?

Since these operational data frames can consist of hundreds of thousands of rows, most "analysts" avoid it.  It takes the additional skillset of data science familiarity in both the tools and languages, such as working with databases, SQL language and possibly more advanced tools, although much of it could be accomplished with SQL, a spreadsheet program (such as Excel) and a BI tool (such as Power BI or Tableau).  This takes little more than an understanding in data tables, views, and the creating of sets for extraction-transformation-loading into the BI tool. If your "analysts" are simple looking at the difference between two numbers such as budget vs actual and can't dig deeper, then you are in trouble.

(e.g. Let's say you know labor costs increased.  Do you know if this was due to increase in average hourly rate, average time to completion, contracted allowable billable hours and rate that were lower than OT hours you were paying, etc... This is all granular. Operational. And true root cause. 

I am often asked to make deep dives into labor; in one situation of commissions I was able to use regression analysis to prove the executive assumption of a standard base rate of sales agents was not only not the minimum wage, but it wasn't even consistent across brick and mortar locations or agents.  It was stepped every $12,000 up to $48,000.  This caused considerable influence to the fixed costs of the business, none of which was known by executive management post-acquisition.

In a second instance, I was examining squeezed margins in a heavily labor-service business.  The most likely culprit is labor time or labor hourly.  In this instance it was a bit of both.  On a granular billable and nonbillable time evaluation, I discovered increasing OT payments to labor but standard AR billing to customers for the same contracts.  The employees were into their OT hours in the third day of the pay period  and were being assigned to contracts with restricted billable hours and rates, common for large  clients whose billings dominated revenue for that period and squeezed margins.)
​ 

Prescriptive analysis is the ability to draw strategic insights from the data to offer suggestions as to operational direction to correct a negative or to strengthen a positive.  Obviously, this requires a skillset to read and manipulate large swatch of data as to root causes so that proper insights can be drawn from instigating a change in any of these drivers, but further understanding into economics, operations, and strategy would prove to be more than a luxury skillset in your analysts if desiring this level of analysis.  Could you really expect an individual to make useful suggestions without understanding operational processes, applications, or the overall value chain for the company?

A prescriptive assessment of any strategic value would need to not only consider drivers, but also consider current throughput and operational capacity, efficient/effective capital investments and capital budget modeling for evaluation.  This often involves resource and capability analysis, constraint analysis, and an understanding of the competitive position of the company within the industry, not to mention pricing strategy and some basic laws of economics. 

​Again, this isn't found in "years of experience", although some could argue that a longer time in an industry allows someone the time to witness actual industry cycles.  I would argue that that could be seen in data when searching for cyclic or seasonal activity, but the point is that this analysis build upon the basic understanding of financial accounting and operational data, data science, and basic modeling, as it dives into strategic management, DCF models and risk evaluation, macroeconomics, and microeconomics.

(e.g. This area often involves risk evaluation because decisions require the directing of finite resources for use in a specific area, foregoing their use in another area.  Capital is most common, but all decisions and activities in a company represent capital in some way.   

In one company utilizing floor financing, there was a push to increase inventory holdings.  The general and often parroted idea was that "it can't be sold if we don't have it".  Of course, if you are missing sales relative to stockouts, then this would be valid, but when the concept is simply the holding of more inventory, inventory you are financing in a time of rising rates, this is may not be prudent.  I offered an adjustment to the idea to state that inventory that is desired should be held relative to its movement in the cash conversion cycle.

I created a model reflecting 30-day average inventory held relative to its value (the basis for floor financing costs), average DSI (days sales of inventory) for categories of product units, and rising prime rates as a basis for variable financing costs.  I also offered a customer analysis by brick and mortar location for inquiries into category of product and closings by product category to get a feel for which products moved at different locations.  Between these two models, I was able to display scenario changes in rising rates, sensitivity of sales relative to rising rates and product, and finally the financing cost so that invested capital into inventory could be done effectively.  The analysis saved hundreds of thousands in financing costs that might have been.)  

 

Predictive analysis is the soothsayer aspect of analytics, the forecasting of possible future outcomes.  This is most commonly used for the forecasting of revenues relative to a current product distribution, new product line, new product segment, incremental revenue from a marketing campaign, penetration rates into new markets, and so on.  It is an estimation of the relationships of historic data and how that data, if holding true into new territory/new markets/future time periods, would look given historic patterns were maintained.  We can future tinker with this future outlook through changing scenarios (scenario analysis) of the historic relationship affected by internal or external business changes and through sensitivity analysis.

Perhaps the biggest misconception about predictive analysis is that it can be found without understanding statistical methods and it really can't be done without it.  Every variable tracked over time is by definition a statistical random variable with a distribution, mean, median, mode, variance, and so on.  If this wasn't the case, insurance companies and financial institutions wouldn't be using actuaries, statisticians, and engineers to predict value at risk and expected shortfall in defaults, deaths, failure rates, bond valuations,  financial investments, portfolios of investments, and so on. 

​The future is risk. You can't be certain your prediction of revenues will prevail and not flop. You can't be certain market adoption will take place in a given time frame if at all.  You can't be certain of anything in the future, hence the future is risk. Risk can't be known exactly, can't be measured exactly, can't be timed exactly, so this risk is modeled with the most unfortunate, but necessary assumption that historic relationships will continue into the future. 

(There are some methods that attempt to model changes in the relationships by modeling volatility, such as GARCH models (General Auto-Regressive Conditional Heteroskedastic), but these are primarily used in financial investments as opposed to corporate financial data.)  

Most commonly, predictive models use regression analysis to estimate coefficients for formulas of an dependent to independent variable(s) relationship.  These formulas predict the outcome of the dependent variable relative to the input of values for the independent variables.  The model uses a "least squares" error method to find the best fit of a linear approximation for the dependent variable.  However, not all relationships are linear.

ARIMA models (Auto-Regressive Integrated Moving Average) use values from previous time intervals to predict future values.  The data must be 'stationary data', a statistical measure and hence the use of the integrated or differenced part of the name, and the auto-regressive introduces more variability into the future values, while the moving average smooths out this variability.

Either case can prove to be useful relative to needs and, yes, these methods are often used in ALL ERP forecasting capabilities since there are only a handful of mathematical/statistical methods used in application algorithms.  And we haven't even gotten into the usage of statistical distributions. 

You might wonder why understanding statistics is essential.  It is due to the fact that the standard deviation is the only way to arrive at a confidence interval and the only useful way to assess weather the forecasted values can be represented by historic data.  The most naive and dangerous thing a company can do is to believe forecasted values can be evaluated based on a basic differencing of the forecast vs actuals because that is not how the forecasting of a random variable or its variance is used.  Random variables offer a confidence interval for which the reported forecast value falls in the middle of this interval.  Say that you have a 90% confidence interval for your forecast.  This implies that 9 times out of 10 the actual should fall within that range ... not be the exact forecasted value.  If you find this to be true, then you can be relatively certain your historic data and model and working properly.  If not, you have some work to do.  Perhaps your variables are nonlinear, not linear (or vice versa).  Perhaps you are choosing the wrong independent variables that are driving the output for the independent variable ... but something is off.

I will ask again ... do you think you can gain an understanding of econometrics and statistics through "years of experience"? 

(e.g. I have used many types of statistical methods to forecast activity levels relative to historic data relationships.  Logistic curves to represent marketing penetration into new geographic markets, ARIMA models to forecast revenues, expectation models to assess demand during strategic pricing analysis, and Weibull distributions to model warranty claims.)       

If you haven't read how "HR is F-in up Your Company", it does directly correspond to the largely poor quality of hires that are filling ranks of analysts within companies.  This is further exacerbated by another topic I have covered, "The Process Trap" in which "years of experience" only exemplifies ones ability to maintain a model or process that has existed for some time.  This could be internal reports, external reports, or dashboards.  There is very limited digging into any area unless a financial number is off and there is very limited creation of new methods and models for evaluation and analysis of operations or finances.  

In part, roles are being filled to meet a quota with little regard to anything other than meeting category requirements.  Another part is a "years of experience" component that supposedly acts as a basis for knowledge and ability.  This is entirely ludicrous when you consider that the first thing done after an interview is to conduct a test on a tool like Excel, one of the most widely used tools in finance and accounting.  If a "5 to 7 years of experience" candidate needs to be tested on their abilities to use a basic tool for analytics, what then makes you believe their knowledge basis and competency of the subject matter is up to par?  If their experience is as valuable as stated, then the use of the tool over years and years should correspond.  After all, how are you going to conduct your analytics over years of experience without using the tool?  Even ERP systems largely require extraction of data to CSV files for analysis on Excel. 

Perhaps more importantly is that the basics of skills and knowledge, as pointed out above, is non-existent in probably 75% - 90% of FP&A departments to conduct analytics or construct models.  Figure out how many can do basic basic algebra and statistics within your departments and you will find my numbers fit.  This is a problem when you begin to realize that all models are equations and all equations are developed from data dives and data analytics. And this is only the basis for modeling and analysis.  Equations and data analysis only create the structure.  The art and "muscle" fit to the structure come from the assumptions which comes from a deeper understanding of operations, finance, accounting, and economics. These positions are treated far too lightly. I, myself, spent two semester long graduate courses developing models from scratch for business valuations and capital budgeting, and investment portfolios and fixed income debt because the curriculum was developed with a STEM skillset in mind.

This problem won't solve itself through the use of the current recruiting and HR system.  Again, see my article linked above.  AI is only going to make this more difficult.  I have now seen many comments saying AI is being used in interviews, listening to the questions offered by HR (who know nothing of the subject matter except for the keywords) and then giving a response for the interviewee to read.  A HR rep that doesn't know an Income Statement from a Balance Sheet is not going to be able discern whether an individual is reading an answer.  Someone with a background in finance might ... because they will offer better questions, less generic, with more discussion on specifics.

Just to give an example of "The Process Trap" in hiring and internal promotions and push the notion that analysts are anything, but I will cite a favorite discussion. At one point in my career, I was discussing the development of a model with a CFO for a contract.  The need was first a cash flow forecasting model and second a budgeting model.  The interesting thing is that they specifically said they have never had either for the company ... only the CFO started as the FP&A Manager, then promoted to Director and finally to CFO.  Over that time they hired another FP&A Manager that had been with them for a few years.  How, I had to wonder, have you been promoted through the ranks in the area of analysis and modeling and you never were able to develop a model?  Not only that, but your current individual in the same area also can not develop these models!!! 

Well, it came to be that they stole my example model and believed they could use it to get a rough template for their model I guess.  After all, basic P&L, BS and CSF pro-forma models are pretty easy to follow.  I could only shrug my shoulders.  I was more amused by the lack of competency and how someone managed to find their way up the chain in a not so small organization.  Over the years I have found that these individuals are not uncommon.

If I was to build out an analytics team, I would find it much easier to find candidates with degrees in math, statistics, engineering, data science, physics, and so on, that then passed professional certifications in the CMA or CFA.  It is much easier to teach accounting and finance to math and stats people than to teach math to ... well... anyone that didn't do math or stats.  I think you would find that these people could not only develop better models, but conduct better analysis with better insights, and have already used databases and SQL, or languages and tools similar, in their endeavors so they could act as associates to IT and data development.  Many of you are missing out and if you want to create an environment of success for the future, you need to figure out how to fix it.