A new “call to suppliers” encourages providers to treat buying AI like buying anything else
I expect you’ve had more than a few emails or messages that run along the following lines.
You’ve been happily running a piece of technology or software for ages – maybe it is that perfect tool that does one thing exceptionally well, or maybe it is the core of your daily workflow. But the manufacturer is happy to tell you that it now comes with a new benefit – they’ve “added AI”: by which I mean a new garish button right where the stuff you actually used was, and at first look not all of that existing functionality is immediately available in the same way without moving to the new “pro” subscription.
Maybe you are excited to try it. Maybe you are angrily muttering – “AI? On my fridge?” and contemplating setting about the thing with a stout stick. Either way, it isn’t your choice and you’ve incurred either a time (you have to learn to use the thing all over again) or financial (apparently “pro” is no longer short for professional and instead refers to someone who pays through the nose in order to actually do stuff themselves rather than churn out slop) cost.
If this is a pain for you, imagine how it feels if you are purchasing or subscribing on behalf of a college or university. You run huge, enterprise-level, systems that need to plug in to multiple other huge systems: and do so via a range of special agreements with manufacturers and homebrew fixes. You run PCs that control one of a kind research equipment with software that hasn’t been meaningfully updated since the turn of the millennium, and provide an interface to scary hot furnace hardware in the engineering department via an open source tool knocked up by a 73-year-old guy in Finland apparently named “deepthought42”. You have to pay ever-increasing annual sums to Google, Microsoft, Anthology, or even Tribal – for reasons you don’t always understand.
If one of those things, or one of the things that talks to those things, “adds AI” the first thing you want to do is evaluate where you actually get any meaningful benefit from the new tool. Would you be better off purchasing your AI tool elsewhere and sticking with the “dumb” version of what you have – or is this even possible if you want to include luxuries like ongoing maintenance or security updates?
Or perhaps you feel like there may be an AI answer to a problem you are facing, but you need to do a bunch of testing and feasibility work in order to ascertain whether you have identified a solution or read one too many Anthrophic press releases. Or maybe your staff or students are clamouring for the one cool thing they read in that press release, and you want to know whether or not self-aware agents will proceed to sell your student data to Russian spies on the dark web as a result.
Enter, inevitably, the cool heads at Jisc. The sector’s technology experts, supported by the Association of Colleges, GuildHE, and Universities UK, have put together the temptingly-named A call to suppliers: governance and cost control for generative AI in education and research to help you. The premise is simple, and familiar to anyone who has worked with Jisc in the past: there’s a new category of technology out there, and you need national support in evaluating the cost, utility, risks, and overall value of a whole host of things your ever-vigilant IT team may only know about because they read The Register of a lunchtime.
The curious title of the report emphasises that many of these new behemoths are not responsive to the traditional ways in which Jisc and the sector have worked with what we used to call the “EdTech sector.” UK higher education gets a good deal from Microsoft, for example, because Jisc has been able to negotiate a very good price for the sector. Other suppliers have similar agreements in place. OpenAI and Anthropic do not, and neither do the 23-year-olds who vibe-coded an “intelligent data analysis tool kit” while micro-dosing mushrooms one afternoon.
These agreements don’t just cover costs: they cover functionality and interoperability too. A company that offers timetabling solutions can be prevailed upon to provide hooks to link in your estates systems and your SIS. And as long as software develops in a predictable way these kinds of agreements and understandings hold. AI changes the game, and in many ways we are back in a dot-com boom style wild west – with competing claims and unheard-of promises offering an opportunity for suppliers to hike up costs and refactor agreements.
The call to suppliers, therefore, could be considered a preamble to an attempt at getting things back on track. It sets out a series of expectations that those making purchasing decisions should expect from those doing the selling: clear and complete information, active choice rather than “bundling”, the right user metrics and charging mechanisms to allow predictability and transparency over cost all form part of a section on institutional choice and control.
With most enterprise software now hosted remotely and paid for via an annual subscription, institutions need to know that the value will be preserved into the future: both in terms of maintaining existing functionality, and adding relevant new technologies as they become both available and reliable. Metrics, on usage but more generally on value added, should be transparent, timely, and usable. This might include helping you understand whether staff and students are using the tool, how much it costs when they do, or what it is that they are actually doing. The key words are “granularity” and “enterprise-grade”: there also needs to be an ability to constrain activity where costs are running away from projections, at departmental, research-group, or project level.
With AI coming as an add-on to everything, it may end up that you are paying multiple times for what is effectively access to ChatGPT, Claude, Copilot, or Gemini. Purchasers should know what platforms power which tools, and have the option to consolidate costs via a single subscription or credit-sharing. And should your university choose to move between AI suppliers, the tools that draw on that source of compute should be able to move alongside you. Likewise as tools and platforms change at a supplier level, there needs to be safeguards for the interests of the purchaser.
Though you don’t have to look far for very well-informed evidence that we are in an “AI bubble” funded via overcooked projections of revenue growth and committed to purchasing more GPUs than currently exist in the world with more money than actually exists in the world, the concept of generative large language model AI is unlikely to go away any time soon. My cynical side notes that selling to quasi-public sector industries like education is generally the last gasp of the first phase of excitement before the costs kick in (how’s that blockchain you bought in 2016 getting on, eh?) but this does not and should not change our observation that there is a lot of value carefully deployed AI tools can bring.
The most responsible suppliers will know that universities expect the same considerations and have the same demands on AI deployment as they do on anything else that they buy. UK universities are not new to purchasing complex technology, and a lot of the existing processes – backed up by this helpful Jisc statements of expectations should enable sensible procurement decisions to be made. No matter how excited that your VCs that he can make a plasticine model of himself on a chatbot, a glance at this publication should remind senior staff that for provider-level enterprise subscription purposes AI is really just a tool – a cost, and a set of opportunities – much like any other.