Our recent webinar on the use of AI in creating public sector bids was one of our most popular yet, with Sam Aaron, the CEO at mytender.io, and Dave McCormack, Creative and Bid Services Manager at Hobs Repro and the host of the ‘Bids and Pieces Podcast’, giving their views on the current state of AI in bidding, the pros and cons, plus a demonstration of just how useful AI can be when creating an effective bid.

Unsurprisingly then, we received a bumper number of questions from those in attendance, and although we addressed several at the end of the session, we soon ran out of time.

Not perturbed, we asked Sam and Dave, plus our own Chris Williamson and John Newton, to answer those we didn’t have time to answer on the day.

So, without further ado, let’s get into it!

Preparing to Use AI for the First Time

Is there a manual on how to effectively use AI, such as for compliance—like whether to upload the ITT and what prompts to use? I understand the benefits, but implementing it seems like trial and error.

Although there isn’t a single manual you can consult (not surprising given that this is a rapidly evolving field with many competing players), solutions such as mytender.io do have product-related guidance on how to get the best results.

The implementation of AI often involves a degree of trial and error, as each organisation’s needs and workflows can vary significantly.

However, we’ve put together some general pointers that can help guide the process:

Define Objectives – Clearly outline what you want to achieve with AI. For instance, are you looking to improve compliance checking, enhance proposal writing, or streamline research processes?

Prepare Your Library – Ensure that you have a well-organised and comprehensive library of documents, templates, and previous bids. This is crucial as AI relies on quality data to generate useful outputs.

Select the Right AI Tools – Choose AI tools that align with your objectives. Some AI solutions focus on document analysis, while others may assist with writing or compliance checking – or all three!

Upload Relevant Documents – Don’t forget to upload relevant documents such as Invitations to Tender (ITT) or previous bid submissions. This helps the AI understand the context and requirements.

Create Prompts – Through trial and error, start to develop specific prompts or questions to guide the AI in generating responses. For example, you might ask the AI to identify compliance issues based on the uploaded ITT or to suggest improvements to a previous bid submission. If you get an answer you like, keep a note of how you posed the question to use next time.

Iterate and Refine – Use the AI-generated outputs as a starting point. Review and refine the results to ensure they meet your standards and objectives. This may involve multiple iterations. However, as your bids improve, and are fed back into your library, you should start to see the overall quality of the AI-generated outputs improve too.

Train the AI – Some AI systems allow for training based on feedback. Provide corrections and suggestions to improve the AI’s future performance.

Monitor and Evaluate – Continuously monitor the effectiveness of the AI in meeting your objectives. Evaluate the results and adjust your approach as necessary.

Seek Expertise – Consider consulting with AI experts or speaking to consultants who already have experience in implementing AI solutions with other firms.

Whilst at first it is likely to involve some trial and error, sticking to the above steps can help you start to improve your use of AI in bidding. Good luck!

As an SME with a growth strategy that increasingly relies on public tenders, how do we begin to understand what information will we need to create a library?

To create a bid library that supports your growth strategy and works well with AI, here are some possible steps to follow:

Assess Your Current Processes – Start by mapping out your existing bid processes and identifying the types of tenders you respond to and their common requirements.

Review Past Submissions – Examine previous successful bids to identify what worked well and any recurring themes or questions.

Categorise Your Information – Organise your content into categories such as company background, case studies, testimonials, methodologies, pricing structures, and compliance information.

Create Document Templates – Develop a standard “master” template for bid responses to ensure consistency and to use as the starting point for future tender responses.

Gather Quality Data – Collect up-to-date and relevant information, including metrics that demonstrate your capabilities.

Involve Your Team – Get input from your bid team, but also from different departments to ensure your library covers all relevant perspectives and information.

Use Technology Wisely – Store and organise your library using tools like SharePoint or other document management systems. Ensure your library is structured so AI tools can easily retrieve relevant information.

Maintain a Record – Keep a spreadsheet or database listing all documents, their file names, and content descriptions to facilitate quick location and organisation.

Regular Updates – Continuously review and update your library to keep it current and relevant. AI systems rely on up-to-date data to perform effectively.

Iterate and Experiment – Test different ways of organising your library to see what works best for your team and adapt based on their feedback.

Make It Accessible – Use tools like SharePoint to store and organise your library. This will help keep track of documents and make retrieval easier. Generic AI tools like CoPilot may be able to access a library stored on your own network, while AI bid writing tools may need you to upload the library to the tool’s own storage.

Ultimately, the quality of your library rests with you and your bid team, but by following these steps you can start to create a structured, comprehensive bid library that will serve as the basis for your future tender submissions.

Are there known data security concerns about pasting content from a bid library into AI?

The short answer is yes, you need to be very careful.

One major concern is the potential for proprietary or confidential information to be exposed if the AI platform is not secure. For instance, using open AI models like ChatGPT can lead to the risk of sensitive data being stored or used in ways that are not controlled, as these models could, in theory, use the input data for training purposes, or make it accessible to others.

To mitigate these risks, it is crucial to use AI solutions that ensure data security, such as those hosted on a virtual private cloud, which restricts access to only authorised users. Additionally, organisations should avoid uploading any confidential information to AI platforms unless they are confident in the security measures in place.

It is always advisable to consult the specific data handling and security policies of an AI service before uploading anything to it to ensure that sensitive information remains protected.

Are you able to train AI on questions, responses, and customer feedback?

Yes, it is possible to train AI models using a combination of questions, responses, and customer feedback. This approach allows the AI to learn from real interactions and improve its ability to generate relevant and high-quality answers. By incorporating customer feedback, the AI can adapt to user preferences and enhance its performance over time, leading to better outcomes in tasks such as scoring and analysing tenders.

mytender.io currently train their customer models based on tender feedback and scores, and there is an early-mover advantage for businesses training their models already. mytender.io are also developing a solution to provide this capability, automatically if you subscribe to their newsletter you’ll be kept up-to-date on this features development.

How does AI address the issue of semantics (the meaning of words)?

AI handles meaning in language using techniques that improve its understanding of words and context. Here are six key methods most AI’s will use:

Natural Language Processing (NLP) – AI uses NLP algorithms to analyse and interpret human language. This involves breaking down sentences into their components, understanding grammar, and recognising the relationships between words.

Contextual Understanding – Advanced AI models, such as those using transformer architectures, can consider the context in which words are used. This enables them to distinguish between different meanings based on the surrounding words, phrases, and overall context.

Word Embeddings – AI uses something called word embeddings, which are mathematical ways of representing words as points in a space. This helps the model understand how words are related and grasp subtle differences in meaning.

Training on Diverse Data – By training on a wide range of texts and contexts, AI can learn various usages and meanings of words. This exposure helps it grasp how language is used in different scenarios.

Attention Mechanisms – In transformer models, attention mechanisms help the AI focus on the most important parts of the input data, so it can understand which words or phrases are key to determining meaning.

Feedback Loops – AI models can improve over time through feedback, learning from corrections or adjustments made by users, which helps refine their understanding of semantics.

By using these techniques together, AI can better understand language and meaning, resulting in more accurate interpretations and responses. However, AI is not perfect, and is certainly no match for an experienced member of your bid team, so make sure that you check the ITT, and your response, for any ambiguities.

Buyer Attitudes to AI

Do the panel members think buyers will start ‘marking down’ bid responses where suppliers have used AI?

Given that we are at the start of this journey, it is currently uncertain how buyers will treat bid responses that have been created, even in part, by AI.

PPN 02/24 shows that the government is alive to the use of AI in the bidding process, but we need more water under the bridge to see how they will respond to its use in practice.

As we discussed during the webinar, technology, including AI, can improve bid responses. However, if suppliers rely on generic AI models without qualifying the information, it may become obvious to the person (or AI!) evaluating the bid, potentially leading to a negative evaluation.

However, if AI is used effectively and tailored to make the most of your firms bid library, you can likely avoid ever having to worry about that risk.

Are you finding that more procurement teams from commissioning authorities are using AI to evaluate bids, and what are the long-term impacts?

We are still in the early days of AI usage, and to our knowledge no commissioning authority has yet stated publicly that they are using AI to evaluate bids.

However, from speaking with procurement staff, we know that buying teams remain stretched, and so it is possible we will start to see this shift, driven by the need for efficiency, consistency, and the ability to process large volumes of data quickly. AI can assist in analysing bid submissions for compliance, scoring responses based on predefined criteria, and even identifying anomalies in the submissions.

The long-term impacts of this trend could be significant. Firstly, it may lead to a more standardised evaluation process, reducing the subjectivity involved in bid assessments. This could enhance fairness and transparency in procurement. However, there are potential pitfalls, such as over-reliance on AI, leading to the devaluation of unique, high-quality submissions that may not fit neatly into algorithmic assessments.

Additionally, as AI becomes more integrated into the evaluation process, there may be a shift in the skills required for procurement professionals. They may need to focus more on strategic analysis and less on manual evaluation tasks. This could also lead to a change in how suppliers prepare their bids, as they might need to tailor their submissions to align better with AI evaluation criteria.

While AI has the potential to streamline procurement processes and improve decision-making, it is crucial to balance its use with human oversight to ensure that the nuances of bid quality and innovation are not forgotten.

Do you foresee procurement teams setting a percentage tolerance for how much of a bid can be generated by AI?

As AI becomes more integrated into the bidding process, buyers may want to ensure that the submissions they receive maintain a certain level of originality and human input. In theory, this could help prevent situations where bids are too similar due to over-reliance on generic AI models. As a result, it is possible that buyers may establish a limit on how much of a bid can be generated by AI.

Setting a tolerance could also encourage bidders to use AI as a tool to simply enhance their bids rather than replace human expertise altogether, preserving the integrity of the bidding process.

However, measures like this would depend on the specific policies and objectives of each buyer, as well as how AI is perceived within the field of procurement as a whole. On balance, we think a limit would be difficult to manage and enforce, so for now at least we don’t think this will be common in ITT’s.

Will the new Procurement Act 2023 regulations include guidance on AI? How far off do you think the Cabinet Office is from endorsing the use of AI to read and score bids?

As of now, the specific guidance provided by the Cabinet Office on the Procurement Act 2023 does not mention AI.

However, the Procurement Act 2023 is not the only game in town, and the Cabinet Office’s PPN 02/24 ‘Improving Transparency of AI Use in Procurement’ specifically deals with this subject, albeit primarily from the buyer’s perspective.

Oxygen recently pulled together some thoughts on the implications of that advice, which could be a good starting place if you’d like to know more, and we would recommend continuing to keep an eye on the Cabinet Office website for further updates, of which we’re sure more is to come.

What is your understanding and view on buyers using AI detector programs, similar to those used by universities for student submissions, and their impact on scoring models and scores?

The discussion about ‘AI detector’ tools brings up several important issues about how they are used in client bidding and scoring models.

Firstly, it is crucial to note that, as mentioned during the webinar, there currently isn’t a ‘magic bullet’ tool that can accurately determine whether a bid was written by AI or a human. Existing AI detection tools often produce false positives and false negatives, which means they can mistakenly classify human-written content as AI-generated and vice versa. This unreliability could lead to unfair scoring outcomes in bid evaluations if buyers start relying on these tools without understanding their limitations.

Secondly, if buyers begin to use AI detector tools as part of their scoring models, it could create a bias against submissions where AI has been used, even when those submissions are high-quality and answer the ITT effectively. This could discourage firms from using AI to improve their bids, reducing the ability of AI, when done well, to lift the quality of responses. The fear of negative repercussions from using AI could stifle innovation and lead to a less competitive bidding environment.

As we touched on during the webinar, using AI that is drawing on internal bid libraries and proprietary information is quite different from using a generic AI tool that is pulling random information from the internet. If suppliers use AI responsibly, drawing from their own resources, it should not reduce the quality of their bids – indeed, it should improve what they eventually submit.

While the idea of using AI detector tools in bid evaluation might seem appealing for ensuring originality, their current limitations are likely to lead to unintended consequences. It could impact scoring models negatively, potentially penalising firms for using AI tools that, when applied correctly, can significantly improve the quality of their bids. On balance we think a more nuanced approach is needed, one that recognises the value of AI while also ensuring fair bid evaluations.

During the webinar Sam mentioned that using AI in bidding within the defence sector is more challenging. What are the specific challenges?

It is true that defence bids can be more complicated that the typical bid, and Sam recommends in the first instance using AI on other use cases where it can be more effective.

Why? Well, typical challenges suppliers face when using AI for bidding in the defence sector include:

Regulatory Compliance – The defence sector is heavily regulated, and so suppliers will need to ensure that any AI tools used comply with all government standards and regulations.

Security Concerns – Sensitive information is often involved in defence contracts, making data security a significant concern. There may be apprehensions about using AI tools that could potentially expose confidential data, let alone data with national security implications.

Complexity of Requirements – Defence contracts can have very specific and complex requirements grounded in information that may not be publicly available – just the kind of requirements that are not easily addressed by generic AI tools.

Quality Assurance – Ensuring the quality and accuracy of AI-generated responses is crucial, as mistakes can have serious implications in defence contracting.

We aren’t saying that AI has no place in defence contracting, simply that most AI solutions, even those specifically developed to assist in bid creation, will need to be used with caution, and will require diligence from the bid teams who use them.

How many organisations have you encountered that are blocking the use of AI in their bids? We have found that some of our clients specifically request that we do not use AI.

We have also heard of cases where buyers have requested that AI not be used, but in truth most buying organisations are still catching up.

Looking forward, if suppliers start to see significant improvements in their win rates due to using AI, and if the resulting service delivery meets the promises made in those AI-supported bids, both buyers and suppliers will have a vested interest in allowing AI. Increased market engagement, encouraged by the Procurement Act 2023, will give suppliers the opportunity to make this case.

We believe that, after a few years of real-world experience, we will see general acceptance of AI in bidding, even if some buyers choose to ban it. Buyers are people, and if the Head of Procurement has had negative experiences with AI bids, they might exclude AI-generated bids from tenders. It is up to the supplier community to ensure this does not happen by ensuring that only thoughtful, honest, and compliant bids are submitted. This responsibility lies with us controlling our own people and processes, not with the buyers.

It’s clear that the use of AI in public sector bids presents both opportunities and challenges. While this field is still evolving, the key takeaway is that careful application, clear objectives, and ongoing refinement are crucial to leveraging AI effectively.

A huge thanks to Sam, Dave, and Chris for their time and, of course, to all those who attended the seminar and asked such good (and some quite tricky!) questions. What’s certain is that this is going to continue to be a hot topic over the coming years, and we’re here for it!