The VVI™ utilizes data sets available in the public domain which are processed through a series of multivariate algorithms to provide input to an AI-driven analytic engine which will generate expected end user pricing with probability assignment. Utilizing machine learning, it autonomously improves its accuracy by continually evaluating results and data points.
The Virtual Vendor Index (VVI™) is designed to gather information on pricing base on a combination of SKU (Stock Keeping Unit) numbers, Part Numbers, Product Description, Product Names or similar designations, and associated published list prices. The VVI™ employs Robotic Process Automation (RPA) to gather the information and is supplemented by manual input as necessary. Each request requires a new run of the RPA’s as there is no history preserved. The position of the item within it’s natural ‘product life’ is also factored in, for example End of Support announcement pending. A corollary of this is the actual maturity of the technology itself on the natural tech maturity S curve.
Another input source is the available financial information of the manufacturer, publisher and where applicable reseller and distributor etc. Of particular importance are measures that are imputed through the various margins and ratios. Other obvious data points are fiscal quarters and the system does not rely on just the current financial statements but uses multiple quarters worth of data.
While the data for public organizations are readily available most privately-held technology companies do provide sufficient guidance to serve as input sources to the VVI. In addition, there are often proxy measures – for example if the company is held by a venture capital fund – then the funds run rate, conversion rate and history etc. are available.
For organizations that transact their B2B through channel partners, value added resellers, integrators and correspondingly through distributors as well – these entities financials also become another set of data. Within this area there is what used to be a murky world of deal registrations, opportunity identification etc. This is now much more codified and as such can be built into the VVI model.
One other key data feed is the sales team, at each point of the sales chain (i.e., publisher, distributor, channel managers, reseller sales and professional services/integrator). Their respective sales performance, commissions, total compensation, stage in the fiscal or half fiscal year etc. are factored into the model.
Geographic information with respect to state taxes, equivalent pricing portfolios, legal registrations that may limit competitive access and funding sources are also factored in and appropriately weighted.
The construct developed above is configured into a model that can be run through the analytics engine. There are 11 common models which if a construct can be fit to, will allow for a quicker turn around. Custom models require additional time to be both developed and validated. The model is subjected to not fewer than 215 simulations, which we have identified as the minimum number to generate a 90% confidence in the output of expected market competitive price for each specific proposal that is analyzed. These runs are very rapid, but after each series of simulations, the output is reviewed and usually needs modification for curves that project to extremes. These critical human/expert interactions are essential as we replicated the simulations to drive to stay within the parameters of the known selling strategies.
Even though the analytic engine and overall process is designed to deliver a 95% confidence interval for the pricing, it is a static run and if the actual purchase takes place later (typically 4 weeks or more) there are possible changes. These changes follow a typical Poisson distribution linked to fixed time events. Incorporating those events allows us to get to 5σ range of possible total price offering that will be accepted by the vendor.