The solution is simple: leverage external data, and simplify your predictive model.

Back in 2000 I was working with GE's analytic team to improve sales forecasts for NBC Internet, a web portal owned by NBC. The sales / finance people were using a very basic formula to predict next month sales, based mostly on sales from previous month. With GE, we started to develop more sophisticated models that included time series techniques (ARMA - auto regressive models) and seasonality, but still was entirely based on internal sales data.

Today, many companies still fail to use the correct methodology to get accurate sales forecasts. This is especially true for companies in growing or declining industries, or when volatility is high due to macro-economic, structural factors. Indeed, the GE move toward using more complex models was the wrong move: it did provide a small lift, but failed to generate the huge lift that could be expected switching to the right methodology.

So what is the right methodology? 

Most companies with more than 200 employees use independent silos to store and exploit data: data from the finance, advertising / marketing, and operation / inventory management / product departments are more or less independent and rarely merged together to increase ROI. Worse, external data sources are totally ignored.

Even each department has its own silos: within the BI department, data regarding paid search, organic search and other avertising (print, TV, etc.), is treated separately by data analysts that don't talk to each other. While lift metrics from paid search (managed by SEM people), organic search (managed by SEO people) and other advertising are clearly highly dependent, from a business point of view, interaction is ignored and the different chanels are independently - rather than jointly - optimized.

If a sales results from a TV ad seen 3 months ago, together with a Google ad seen twice last month, and also thanks to good SEO and landing page optimization, it will be impossible to accuratly attribute the dollar amount to the various managers involved in making the sale happens. Worse, sales forecasts suffer from not using external data and econometric models.

For a startup (or an old company launching a new product), it is also important to accurately assess sales growth, using auto-regressive time series models that take into account advertising spend and a decay function of time. In the NBC Internet example, we've found that TV ads have an impact for about six months, and a simple but good model would be

Sales(t) = g{ f(sales(t-1, t-2, ... , t-6), a1*SQRT[AdSpend(t-1)] + ... + a6*SQRT[AdSpend(t-6)] }

where the time unit is one month (28 days is indeed better), and both g and f are functions that need to be identified via cross-validation and model fitting techniques (the f function corresponding to the ARMA model previously mentioned).

Pricing optimization (including an elementary price elasticity component in the sales forecasting model), client feedback, new product launch and churn should be part of any basic sales forecasts. In addition, sales forecasts should integrate external data, in particular:

• Market share trends: is your company losing or gaining market share?
• Industry forecasts (growth, decline in your industry)
• Total sales volume for your industry, based on competitor data (the data in question can easily be purchased from companies selling competitive intelligence).
• Prices from your top competitors for same products, in particular, price ratios (yours vs. competition)
• Economic forecasts: some companies sell this type of data, but their statistical models have flaws, and they tend to use outdated data or move very slowly about updating their models with fresh data. 
• Analyticbridge plans to provide econometric forecasts, and to create an econometric index about the future of the economy. We've identified metrics connected to some of our internal sales, that are very good 30-day predictors of the stock market indexes and of the economy in general. More on this later when we have completed all our calibration tests. We will probably make a version of our economic index (30-day forecasts) available for free.

A very simple model

Identify the top four metrics that drive sales among the metrics that I have suggested in this article (by all means, please do not ignore external data sources - including a sentiment analysis index by product, that is, what your customers write about your products on Twitter), and create a simple regression model. You could get it done with Excel (use the data analysis plug-in or the linest functions) and get better forecasts than using a much more sophisticated model based only on internal data coming from just one of your many silos. Get confidence intervals for your sales forecasts: more about this in a few days; I will provide a very simple, model-free, data-driven solution to compute confidence intervals.

How to hire a good sales forecaster?

You need to hire some sort of a management consultant with analytic acumen, who will interact with all departments in your organization, gather, merge and analyze all data sources from most of your silos, integrate other external data sources (such as our forthcoming economic index), and be able to communicate both with executives and everybody in your organization who owns / is responsible for a data silo. He / She will recommend a solution. Conversations should include data quality issues, which metrics you should track moving forward, and how to create a useful dashboard for executives.

Are these data gurus expensive? Yes, they usually cost more than $150K/year in base salary, in United States. If your budget is limited, feel free to contact me at vincentg@datashaping.com: I work for free, and yes, there's a catch: I only work for projects that I am very interested in, and my solutions are eventually published in the Data Science book by Analyticbridge  (although your company name will not be mentioned).