You are viewing 1 of your 1 free articles
The AI revolution? ‘In 12 months, no one is writing bids or tenders without using software like this’
It has been a little over a year since ChatGPT burst onto the scene, and large language model AI has proliferated since then. While you may have seen AI used to create memes, it is also being used by companies wanting to win contracts from social landlords. Simon Brandon reports. Illustration generated by Bing
How would you feel if this article had been written by artificial intelligence (AI)? What if its algorithmic authorship had been hidden from you?
To be clear, it has not – this article is 100% human-generated – but these are questions facing social landlords today, especially procurement professionals trying to decide on which companies should win their business. It is just over a year since ChatGPT – a large language model AI (see box, right) that can generate text on just about any subject – was released to the public. It reached one million users in just four days, a record for any new tech product.
This technology has proliferated since then. IBM, Google and Meta have all released their own large language models. These tools are already being used in all manner of professional tasks, from drafting lesson plans to handling customer service queries and writing marketing copy. If you have played around with a programme, you will understand why: you input your query, bash the enter key and the AI will deliver as much copy as you need.
READ MORE
This functionality has now found its way into social housing procurement. Suppliers are using AI tools such as these to help them write bids, and it is probably more widespread than you think.
“I think it is being used fairly extensively already,” says Ian Makgill, founder of Spend Network, a firm that gathers and analyses public sector procurement data from around the world. “From the conversations we’ve had with bid writers, they are putting questions into AI, which then returns a first draft… then [they] refine the bid as a team before submission. I think it’s already being adopted quite significantly.”
Sector’s views
Rebecca Rees, a partner and head of public procurement at law firm Trowers & Hamlins, says her clients are generally aware that some bidders are using AI. Their reactions, she adds, are mixed.
“Some people think it’s no better or worse than hiring a professional bidding organisation or employing bid writers,” Ms Rees explains. “But some think it’s not the effort or the views of the bidding organisation and therefore it should… be viewed more suspiciously than people that employ bid managers to write the thing.”
The sector seems wary of this topic, at least for now. A spokesperson for the Chartered Institute of Housing says this issue has not shown up on its radar yet, while most of Inside Housing’s requests for comment from large housing associations, procurement organisations and contractors have gone unanswered.
There are a couple of exceptions, however. One bid writer for a large contractor with many clients in the social housing sector, who asked not to be named, says her organisation has looked into using AI, but has not been very impressed – at least, not so far.
“We’ve been trialling it, but it doesn’t really work,” she explains. “It doesn’t give us specifics – the sort of detail that we would need to make a response bespoke.” She adds: “If we were up against it, it could be a starting point which we could then tweak and refine – but we’re at the very, very early stages at the moment.”
Notting Hill Genesis (NHG), a 64,000-home London-based association, is certainly on the case. “Our procurement team are aware of the intervention of AI on bids and are engaging with our e-procurement provider to incorporate a form of AI tracking on submissions,” says a spokesperson. “In the interim, we are reviewing our qualitative questions, making them far more project and service-specific to deter AI intervention.”
How tracking might work is unknown. Inside Housing approached NHG’s e-procurement provider, Delta, for comment but it had not replied by the time of writing.
Computer says ‘yes’
It is possible that attempts to distinguish human and AI-generated answers in bids might be futile. This is partly due to the sophistication of the technology, but it is also made harder by the kind of formulaic questions often asked in tenders.
“The problem for buyers is that the questions they’re asking are exactly the sort of questions the AI is pretty good at answering,” says Mr Makgill. “It’s marketing copy, and [the AIs] have been trained on literally billions and billions of paragraphs of marketing copy.”
Here is one such question taken from a recent tender for grounds maintenance services for a small housing association: “How will you develop this contract to become more biodiverse and encourage green space (750 words)?” When this question is put to ChatGPT, its response is detailed, clear and divided usefully into sections, such as ‘plant selection’ and ‘pollinator-friendly practices’. It is impressive, albeit generic, but it says nothing about the bidder’s level of expertise in anything other than using ChatGPT.
Where it could be useful is as a framework on which to hang a more detailed, site-specific response: which plants, for example? And why? Used in that way, the difference between asking an AI to structure your answer, or hiring a professional bid writer to do the same, comes down largely to cost.
What are large language models and how do they work?
The new generation of AI tools, such as ChatGPT, are known as large language models. They have been trained on massive amounts of text in order to understand and generate human-like language. They are essentially highly advanced auto-complete engines – they can predict what word comes next in a sentence given the context of what came before.
There is no doubt that these AI tools are very good at aping human communication, but this can obscure their drawbacks.
“ChatGPT is incredibly limited, but good enough at some things to create a misleading impression of greatness,” wrote Sam Altman, chief executive of OpenAI, the company behind ChatGPT, in a tweet last year. “It’s a mistake to be relying on it for anything important right now. It’s a preview of progress; we have lots of work to do on robustness and truthfulness.”
Ms Rees describes the word-heavy answers demanded by questions such as this as a “proxy for effort”, a way to demonstrate the effort and commitment companies are willing to put into the bidding process. According to Mr Makgill, the advent of large language models means that proxy is now obsolete.
“The whole of the public sector has this really weird reliance on text and number of words,” says Mr Makgill. “In the world of AI, all of that is gone. You’re much better off asking, ‘If you are going to respond to this bid, tell me what prompt you would put into ChatGPT,’ and then we can all cut to the chase.”
There is an old saying in computing: “garbage in, garbage out”. In other words, the quality of output depends on the quality of input. For large language models, the user’s prompt is the input, and its quality rests to a great extent on the amount of detail and specificity the user includes. According to Mr Makgill’s argument, a bidder’s level of knowledge and expertise will therefore be evident in their prompts; it is the difference between reading bullet points or an entire essay. But it is not a solution in itself, because these AI models can also help their users write better prompts.
“The whole of the public sector has this really weird reliance on text and number of words. In the world of AI, all of that is gone. You’re much better off asking, ‘If you are going to respond to this bid, tell me what prompt you would put into ChatGPT,’ and then we can all cut to the chase”
Using AI also raises potential issues around data privacy, as the third-party platform may well have ownership of all the data entered in. It is normally assumed that information in tenders will be kept confidential, Ms Rees says, to mitigate risks such as bid-rigging or conflicts of interest.
“Bidders will often be required to sign a non-disclosure agreement or confidentiality undertaking,” she explains. “So at the moment, the use of an AI tool is entirely incompatible with that approach.”
It is not all bad, though. This technology could strengthen procurement by giving smaller suppliers, which may not have their own in-house bid-writing team or the means to hire bid writers, a cheap and easy way to elevate the quality of their submission. This could increase both the number and quality of bids.
After all, if you are a small grounds maintenance firm bidding on the biodiversity tender example, what does the quality of your penmanship have to do with your ability to do the job? “If you’re [a small business] and you’re trying to get up the ladder, AI could be a real help to you as you write your initial draft,” says Mathew Baxter, chief executive and founder of Echelon, a housing procurement consultancy.
Bespoke responses
If large language models are less useful at answering more detailed, case-specific questions, then that could give wary housing organisations a way to subvert their involvement in the bidding process. “It’s quite difficult to use AI to give a bespoke response to a bespoke question,” as Mr Baxter puts it.
As well as this shift towards more focused, case-specific questions, the use of these programmes in bid writing could also be undermined by another growing trend: the use of face-to-face meetings during the bidding process.
The Procurement Act 2023, which is due to go live in October, gives contracting authorities greater flexibility in how they design their tendering processes. Mr Baxter believes one consequence of this will be much more interaction between suppliers and clients. “It’s got much more emphasis on getting around a table with your bidders to understand the commerciality of their bid, to try and identify risks, and to help form the final solution for delivering that work,” he says. “I don’t see how a contractor could use AI in that process.”
Procurement may be changing and developing, but so is technology. AI-focused start-ups have public procurement firmly in their sights, and this niche new sector is moving fast. The chief sponsor of this year’s conference for UK bid-writing professionals was AutogenAI, a British firm that has developed an AI model dedicated to writing bids.
“If I thought people were putting in bids using generic large language models with no human input, that would be a terrible thing”
Founder and chief executive Sean Williams gave Inside Housing a quick tour of his product. When asked the same question on biodiversity as ChatGPT, the AI’s answer covered similar ground – but its functionality was streets ahead. It generated a string of ideas for the user to choose from and order, each one clearly sourced, before stitching them together according to the user’s preferences: tone, word count, structure.
It is not about replacing humans, Mr Williams says, but about augmenting them. “In the same way that a financial modeller uses Excel to write quicker, more sophisticated financial models, our software is used by writers so that they can write more sophisticated, better answers, because they’re able to do a lot of that work faster,” he explains.
Publicly available large language models, in comparison, are simply the wrong tool for the job. “If I thought people were putting in bids using generic large language models with no human input, that would be a terrible thing,” he adds. “Fortunately, anyone who does that is just going to get beaten by somebody who’s doing it properly.”
Should bid writers acknowledge that they have used AI tools to help prepare their bids? “There’s an obligation on people to be honest about what they’re doing,” Mr Williams says.
He acknowledges a wariness about this kind of technology – which might explain why his firm protects its clients’ identities – but that might not be the case for much longer. “Trust me,” he states. “In 12 months’ time, no one is writing bids or tenders without using software like this.”
Other longform articles by Simon Brandon
Becontree: the 100-year-old estate
This year, the Becontree Estate in east London turns 100 years old. To celebrate the largest council estate in the UK, Simon Brandon went round to see what makes the estate so special and how it has stood the test of time
Why has diversity progress stalled?
Simon Brandon talks to housing sector veterans to find out how they think social housing should solve its race diversity problem in top jobs
