Why Forecasting is important for SEO - Part 1

by Jack Reid on Friday 7 September 2018

Statistical forecasting is an important tool used in all types of business to predict future events and aid planning. Specifically looking at SEO, the ability to predict organic sessions, orders and revenue and then aligning this to return on investment is of great importance, but a forecast of the future can only be as accurate as the data and information available allow it to be.

 

There are ways to make it much more accurate though, so here are some considerations to bear in mind before making that all-important forecast.

 

How to Start Forecasting

To begin forecasting the following needs to be known:

  • What will the forecast be used for?

  • What type of forecast is required?

  • How granular should it be?

  • Over how long will the forecast be required?

  • What data do you have?

  • Past Event History?

  • Future Plans?

  • Assumptions?

 

What will the forecast be used for?

There are a number of reasons to forecast. Does a client require the projected value for maintaining SEO efforts? Is a business case needed for sign off on extra resource? Once the aim of forecasting is known it will be easier to present the required data for its purpose.

 

What type of forecast is required

There are a wealth of models for forecasting. Simple forecasts can be made in Excel, for example, using the “FORECAST” function which uses simple linear regression. More complex models can account for keyword search volumes and customised click-through rate models combined with seasonality.

 

How granular should it be?

Forecasting can be split out as granular as needed, and this is usually dictated by the type of industry and specific business requirements. Quarterly or Monthly data may be used for more top line forecasts, which may also have more assumptions built into them. Weekly and even daily data may be useful to use especially if specific known future event dates are planned (such as sale) and there is clear seasonality e.g. traffic peaks for B2B industry during the week days with lulls on the weekend due to workers being out of office.

 

How long is the forecast required?

This is key to know - short term trends will be easier to predict and less liable to swings. Forecasting for long periods can open up to unforeseen factors. However this is why re-forecasting needs to be assessed after the initial forecast has been made.  

 

What data do you have?

To predict future trends, reliable data from past events is required. At least 2 years’ worth of data is ideally needed to see any common trends (seasonality) in the data series.

 

Past Event History

This is key to know as data may be dependant on past events. Did website migration results affect visibility? Was the site affected by one of the latest Google algorithm updates? Were facets unlocked on an ecommerce site which meant there was a large boost in visibility from long-tail search queries?

If this data is known then past trends can be interpreted accordingly. Data can then be adjusted and factored into future predictions to make them more accurate.

 

Future Plans

The future plans need to be known so that they can be incorporated into the model. Will there be a freeze on development which will hinder implementation of SEO efforts? Is there any planned migration work which may affect visibility? Will organic budget for be cut which could stunt growth?

Knowing scheduled roadmaps from your department as well as wider teams plans will only help to be able to paint an accurate picture of where the future is headed.

 

Assumptions

Possibly the most important part of any forecast is having a clear set of assumptions that the projection is built upon. Many of the prior points covered may factor into the assumptions, but also include wider business considerations. Where is the business in terms of growth? (The business may experience greater growth early on with established businesses at a steadier rate). Is organic visibility becoming harder to maintain due the advent of voice search, richer SERP results and ever increasing competitive landscape? Will brand awareness need to be factored in?

Once a list of assumptions and other data mentioned above is known, forecasting can begin.

 

Example Forecast

The following example looks at a US fashion ecommerce website. First, the questions defined earlier can be answered based on this sample before drafting up a forecast.

  • What will the forecast be used for?

To forecast the next 12 months’ organic visibility.

  • What type of forecast is required?

A deseasonalised linear regression method will be used, with average-percentages for each month.

  • How granular?

Monthly.

  • Over how long will the forecast be required?

12 months.

  • Data?

Access of 3 years’ worth is available.

  • Past Event History?

N/A

  • Future Plans?

No drastic plans e.g. assume similar trend for the future. Hence deseasonalised linear regression may be an accurate forecasting model.

  • Assumptions?

Assume that investment will be the same as in previous years without any growth for the time being.

  

Check out Part 2 of this guide here.