It has been reported this week that South Wales Police is receiving improved performance from their facial recognition system, this after negative media reports over the summer on the accuracy of the system. (What?! Facial Recognition at Champions League Final in South Wales Generated a “92% False Positive Rate”? What Does that Even Mean?)
Improved performance of the algorithm is always desirable. However, if South Wales Police had chosen an algorithm agnostic platform, they would be free to choose the best of breed algorithm from the market, rather than being constrained to algorithms supplied by the vendor of the platform.
However, procurement strategy has been steadily maturing.
There is a progressive shift towards separating platform, workflow, presentation layer and integration layer from underlying algorithms, the latter of which are increasingly becoming commoditised anyhow. This trend has been evidenced by multiple significant recent government procurements.
Whilst algorithm developers may not be overly fond of this as it erodes their ability to shoehorn themselves in higher up the value chain, it is to the benefit of the customer as it avoids vendor lock in at the algorithm level and ensures proper competition at the application layer. i.e. a) not being forced to buy Vendor A’s application to access the capability of Vendor A’s algorithms and b) not being forced to write off their investment in Vendor A’s application and reinvest in Vendor B’s application solely for the purpose of being able to access Vendor B’s algorithms when another vendor produces a superior algorithm, which inevitably will happen.
The argument that only the algorithm vendor can get the best out of their own algorithms at the application level fails the smell test for a few reasons: a) properly architected software with robust API’s should sufficiently modularise and abstract capability. b) most vendors undoubtedly have completely different development teams working on the application layer and algorithms anyhow; it should be irrelevant if they are within the same company or not. c) If this is not true, then the vendor either is not architecting their software properly or is reserving much more functional APIs for internal use only, which clearly provides them competitive advantage but equally clearly is not in the best interest of the customer as it diminishes competition.
This is not to say that algorithm vendors should not play at the application layer. In fact, as algorithms are increasingly commoditised, this is where the main ability to differentiate and demonstrate your understanding of the customer, the customer’s problems and the customer’s needs will occur. But, if organisations want to supply full-stack solutions, they should demonstrate confidence by not forcing vendor-lock across different tiers of the solution.
As stated at the outset, this is increasingly being mandated by customers anyhow.