A recent article on Civil Service World noted two things that attracted my attention:
- To manage innovation, governments must fix procurement
- Too often, contracts in AI do not give governments powers to investigate algorithms or the data they are trained on. As a result, they risk taking the blame when things go wrong without the means to find out why.
Public Procurement is expensive. Very expensive. Given that it represents 12% of the annual GDP of an average developed economy, that is a huge amount of spend. Given that the overspend in most departments of most jurisdictions is likely as bad as in the private sector, which means, depending on the category, is likely in the 4% to 6% range at a minimum (based on the results high performing organizations see when implementing best-in-class processes and technology), that means a minimum of 1/2% of GDP is being wasted annually, but based on the fact that most public sector projects exceed initial budgets and timelines, we’d bet that the overspend is double that and at least 1% of the annual GDP. That’s a lot of waste — 770 Billion on the top 10 economies. Furthermore, that assumes that all of the spend is necessary and well planned. (There is likely considerably more savings with better demand planning, more operational efficiency, better project planning, etc. We’re just stating that the savings on committed spend alone is likely 10%.)
The article notes that despite the strategic importance of Procurement, it’s rarely seen as a priority and is more often treated as a standardized compliance function, rather than a tool for strategic investment and, in some cases, has become synonaomous with absurdity, due to an accumulation of rules so complex that even those administering them cannot interpret them creates the perverse incentive of doing the least risky thing to avoid individual liability. As a result, governments end up buying obsolete technologies that make them vulnerable, because innovation evolves so rapidly, and forces them to buy more. The cycle repeats, budgets balloon, and public capabilities diminish.
And, unfortunately, public procurement is a brick-and-mortar process, still more suited to bulk-buying precisely describable goods, accounting for them, and moving onto the next purchase. Innovation is different: you do not know today what is going to be possible tomorrow, even when you are the one inventing the tech. While governments work in one-off projects, innovation is made of ever-changing, always-fleeting products.
Furthermore, those in charge of procuring these technologies are not technologists. Public procurement is professionalized in only 38% of OECD countries, so even if officials had the incentive to experiment, they would not have the expertise.
To combat this, the authors of the article propose that Procurement systems should be like good software, fluid, flexible, and constantly evolving. However, as they note, this will take more than changing rules. As they note, it will take talent that are experts in what they are buying. It will take the treatment of Procurement as a strategic function, with clear lines for advancement for all personnel (as studies have shown that even a marginal improvement in skill can yield significant reductions in costs, times, and contracting complexity). Thirdly, they will need a federated data environment to make use of modern technology. (Especially if they want to use AI.)
This is just the start of what is necessary. There needs to be regular training. There needs to be specialization to different types of functions and purposes. There needs to be a rewrite of rules to focus on the right outcomes, not just a plethora of rules designed to prevent previously undesirable outcomes. There needs to be clear paths from buyer to public organization CPO to department head, not just paths of advancement within the Procurement function. There needs to be a focus on what’s best for the public being served, not best to minimize the risk to the buyer. And a willingness to accept that their may be a few mistakes made here and there as new buyers learn the ropes, while a willingness to weed out anyone that “makes a mistake” in order to give a contract to a supplier who is not the best fit (and do so in exchange for a kickback).
But most importantly, if they acquire AI technology, they also need to acquire the right to investigate the algorithms being used, the data it is trained on, the results of prior training, and the right to inspect any changes to the algorithms, data, and training. Otherwise, you can never trust any AI technology you might want to acquire.
Because governments need to apply the most appropriate AI-enhanced technology more than the private sector, but are the least likely to be able to use them properly.