The framework of analysis execution based on the business financial structure
The concept of this article is that I believe most of the analyses are using to solve a business problem related to the final goals-increasing the revenue or reduce the cost. Target audiences of this article would be (1) entry-level analysts/scientists who want to understand the full picture of the business and which part you are working for. (2) Mid or Mid senior level Analytics manager or strategist who wants to clarify the direction of leading your team. So I would like to introduce a framework that provides different perspectives to cut through the value and solve business problems.
A framework from the perspectives of finance analysis
I have listed sub-categories of each main dimension (Revenue, margin, operating revenue, operating margin), highlighting in yellow the example of daily analysis and related algorithm/indicators(of course just limited examples here, the analysis world is tremendous) to help everybody understand the structure of this framework. You will understand the analysis/model that we build and deliver is to help daily business. Hence as a mid-level leader, you may set your goal and develop your strategy based on this framework and your company situation to win your credit.
I. Vertical angle
We can start our analysis with three sub-dimensions: revenue, margin, and operating revenue, from the vertical angle. (1) Revenue. Think about those AI, ML algorithms. We adapt those algorithms to solve specific problems and tune our model through a better data set or adjusting parameters. For example, we use the recommended system and retargeting profile selection, trying to increase frequency/per customer transaction/ order volume. All of the above are correspond to revenue. Another example is that we do competitor analysis, compare our product with external products, and do A/B tests on APP functions internally, stand out our products or improve customer retention. A growth model is, no doubt, a good methodology to keep the activation, maintain the retention, and contribute social recommendation. The model is also linked to the revenue part.
So, how are we gonna deliver the analysis from a revenue perspective? We can start with the product, analyze the product portfolio, the revenue % of each portion to see which product line may be a risk. We can also start with the region, comparing the revenue in between regions/countries/cities. Remember to break down the data and cascade it. We can also analyze revenue by time frame. For example, we can dig the data and find the best time to put the advertisement, or we can try to find the seasonal trend through time-series methodologies. Or we can start with the channel to see which channel contributes the most, online or offline.
Once you have a cut in dimension, you can leverage your talent pool and do experiments with different ML models to run tests/analyses to solve the problem you see. Then what are the measurements? In many articles or expert sharing, they have mentioned many indicators setting and monitoring a process. I want to highlight here the measurements are different from industry to industry, so leverage the ones your company reviews the most, such as ROI, GMV, ROAS, each rate in the AARRR model for business, and accuracy, precision, AUR for models. The measurements are put on the top of the framework as ratio analysis. It would be best if you had the ratio analysis in every part of your analysis.
(2) Margin, we consider product cost, especially raw material, interns of margin. We are not only trying to increase the revenue but seek a way to control cost. We can try a raw material supplier analysis, find out a substitute for raw material, or find out the economy of scale. Under this dimension, some companies using AI to find better material with lower cost or using AI to forecast the sales and plan its production plan. So, we can deliver our analysis through a margin angle to solve the business problems.
(3) Operating revenue: In this dimension, we focus on SG&A. SG&A, including salaries, marketing expenses, R&D expenses, and freight rate. An example is that we use classified algorithms such as xgboost, random forest, trying to find those customers who may leave, reducing the churn rate, which correlated to the marketing cost. We use deep learning to find the optimal logistics route, which is also an example of reducing freight rates. So Once you defined a business problem, you can start with operating revenue, looking around each item to see if you can correlate your business problem to the above SG&A items.
Normally we used horizontal analysis to find the trend and to address the problem. The next step is to adopt an AI/ML algorithm or some analytics methodology to provide solution/business suggestions on business, such as a customer list for advertisement. R&D example is that proper medical solution that doesn’t need a bunch of time to find out to patients
II. Horizontal angle
From the horizontal perspective, we make a same-period comparison with the same indicators/ratios and eliminate possible effective factors. Here is an example. You can find that a category we mentioned in vertical analysis (here is revenue, through product line and drill down to region ) is the analysis indicator. We can make different kinds of horizontal analyses. It is also important for you or your team to emphasize how better you bring to the company through analysis / ML modeling/algorithm. Remember that the quantity numbers or quality improvement are only meaningful while comparing with the historical-self/industry benchmarks or study suggestions.
(1) MoM, QoQ to check the improvement
Compared with the status of the same period last time, it is the best way to call out your performance, proving your efforts are worthy or suggesting a proper design (such as web page, APP UI, workflow) based on the data support. The time frame of this method is different from industry to industry. It depends on your industry pace, product life cycle, and special-purpose (e.g., new hire observation, new member activation). An essential thing here is that remember to find a proper indicator to prove your efforts and eliminate those factors(e.g., the impact of a device type for recommended customer marketing email list vs. the traditional list) that may affect your experiment.
(2) User journey: the first N days, N days after a special momentum
In some industries and for some products(especially APP), we will put our focus on the user behavior in some specific period, for example, we want to know the subscription numbers after the first engagement, or we want to understand new member behaviors in 7 days, 14 days after their APP download, or member registration. Normally, we will execute A/B tests, observe a specific ratio, and set the horizontal line as 7 days, 14 days, or some special period for different industries. Improving user journey is also one direction that you can consider while bringing value to your org.
(3) vAOP(Annual Operations Plan)
There is a plan for each indicator in an organization. The plan may be weekly, monthly, quarterly, or annually. One way to highlight the contribution of analysis is to say we brought extra % or growth compared to the AOP, which is extremely connected to the company goals. Our model/analysis, in a sentence, is to help the company achieve goals in a better/efficient way. So understand AOP(or we say phase KPI) and associate our analysis/model to somehow related is the thing we need to learn.
(4) Rolling + forecast
There is usually a seasonal trend of business, and rolling review is a way to eliminate the outliner and make the comparison more reasonable. Hence, the evaluation matrics of analysis sometimes would be taken from a rolling perspective. When we design experiments, do remember identifying the proper time frame(avoid big marketing or annual event) for experiments. The same as for building models, the big events usually bring extra sales or eyeballs and consider making the right treatment while executing data clean before building models. Some concept when you are making a forecast, the high-low season should be carefully considered and adjusted in the forecast.
Here is one more thing
As a leader of an analytics team or experience expert in the business analysis/data science field, you need to understand 1. your industry 2. your company 3. the finance structure of your company. So that you can find a proper cut-through point and leverage your expertise, driving your team to deliver practical analysis or models to help your company better achieve targets, this article covers the structure of internal analysis. I will have another article that touches base on the structure of external analysis(environment) to ensure we fully picture a data-driven mindset on business. Stay tuned.