Your browser is not supported. Please update it.

22 August 2019

AWS Forecast will spare many blushes, but zero threat to research sector

Remember the barrage of criticism Netflix received after under-forecasting its domestic subscriber outlook in the last quarter, leading to a punishing share price dive? Well, such gross miscalculations could be a thing of the past as AWS announced the general availability of Amazon Forecast this week – a machine learning-based service for delivering highly accurate business forecasts.

Let’s be clear first of all. Amazon Forecast isn’t in the game of rivaling research outfits like our sister service Rethink TV, nor would the system be trusted to forecast emerging markets and innovative technologies without bias. But projecting business conditions like sales, product demand, infrastructure requirements, energy needs and staffing levels is Amazon Forecast’s bread and butter – with the technology bound to trigger some serious fat trimming at in-house forecasting teams.

Amazon of course is a chief rival to Netflix in SVoD, yet many forget that Netflix is in fact hosted on AWS cloud infrastructure. Moving forward with Amazon Forecast to reduce the chances of a repeat forecasting blunder could be a tempting scenario for Netflix, with the technology claiming to make predictions up to 50% more accurate than traditional methods through custom models.

AWS claims achieving high levels accuracy remains elusive for two reasons. Traditional forecasting methods, from simple spreadsheets to complex financial planning software, apparently struggle to account for and process large volumes of historical data, ultimately meaning missing out on important past signals. Secondly, AWS says these traditional forecasts generally fail to incorporate related but independent data, which can offer important context (such as sales, holidays, locations, marketing promotions, etc.). “Without the full history and the broader context, most forecasts fail to predict the future accurately,” claims AWS.

This is why Amazon Forecast is not about to gobble down the lunches of the market research sector. It lacks that fundamental gut instinct which is inherent in forecasting – second to the most important aspect which is talking to a variety of people inside the industry being forecast. Amazon Forecast can equip companies with the tools to build models more accurate and powerful than anything else on the market, but it cannot magically provide forecasting teams with insights into how rivals are performing. Here we have the missing ingredient.

Forecast sounds a formidable weapon nonetheless – developing and feeding its machine learning algorithms with a wealth of data from delivering billions of packages every year for over 20 years. Amazon says this ecommerce experience has translated into advanced forecasting techniques capable of automatically discovering how variables including product features, seasonality and store locations can affect each other. It constructs a so-called data pipeline, where data is ingested and a model is trained, providing accuracy metrics for performing forecasts, meaning developers don’t require any machine learning expertise and can build machine learning models in less than five API calls or clicks.

Professional services firm Accenture is listed as a customer case study, describing Amazon Forecast as “simple to operationalize for production workloads, allowing us to experiment with multiple state of the art learning models for forecasting across our business.”

AWS also name-drops Brazilian firm OMotor, which uses machine learning algorithms, computer vision techniques and cognitive bots to communicate via WhatsApp and other platforms. “Using Amazon Forecast gives us the ability to create and refine various forecasts from time series data without having to build and train a model manually every time. We forecast real sales for the next 12 months, so we can adequately plan for inventory, estimate future profitability, track market share gain or loss, and other insights. This means we can use more contextual data, optimize more frequently, generate forecasts with upwards of 50% improvements in accuracy, and operate at a great speed. For example, we’re helping customers in the automotive industry predict sales across 185 vehicles in Brazil,” according to OMotor CEO Marcio Rodrigues.

Amazon Forecast is the second notably disruptive product move this summer, after the company made its recommendation software Amazon Personalize generally available to AWS customers in June. Personalize was previously an exclusive luxury relished only by the retail platform itself. Amazon’s contentious thinking is that the time has come for recommendations to become ubiquitous in a sense through its self-titled master personalization algorithm – rather than having recommendations siloed into one engine for video, another for music, another for news articles and another for product sales.

Amazon Personalize essentially simplifies the development of applications with personalization use cases, including product recommendations, individualized search results and customized direct marketing. Personalize trains, tunes, and deploys custom machine learning models, while provisioning the necessary infrastructure and managing the full machine learning pipeline. An API provides customers with results on a pay-as-you-use basis – claiming no machine learning experienced is required.

At the time, however, we questioned the credibility of Personalize to effectively serve recommendations outside the retail sector. The same could therefore be true for Amazon Forecast – will the technology remain rooted in retail or will the system branch out and prove pivotal in helping companies in the digital entertainment sector avoid future slip ups akin to Netflix’s recent forecasting howler?

“Given the consequences of forecasting, accuracy really matters. If a forecast is too high, customers will over-invest in products and staff, which ends up as wasted investment, and if the forecast is too low, they will under-invest, which leads to a shortfall in raw materials and inventory; creating a poor customer experience,” states AWS.