Choosing the right AI tool for tender work

The market for AI tools for tender work has grown quickly over the last few years. For those of you who are decision makers for purchasing at a supplier company, the choice of tool is a strategic decision that affects both the data you handle, how the team works, and how you meet requirements from buyers and regulations.

What we see in the suppliers we talk to is that those who have chosen the right tool have asked some concrete questions before choosing. And those who have chosen wrongly often have not asked them. Here is the checklist we tend to recommend.

What is at stake when you choose an AI tool

A company's tender and bid documents often contain sensitive information: technical solutions, pricing strategies, references, security measures. What you feed into an AI tool is handled by an external party in one way or another. If the handling is not thought through, you can end up sharing trade secrets or breaching regulations you yourselves are bound by.

In addition, the choice is long term. An AI tool quickly becomes a central part of the workflow, and it can be both costly and cumbersome to swap out after one or two years. That is why it is worth spending time on the assessment up front.

Data handling: what happens to the documents you upload?

The first thing you should find out is what happens to the data you feed into the tool. Concrete questions to ask the vendor:

• Where is the data stored, geographically? Inside the EEA, or elsewhere?
• Who has access? Only you, or also the vendor of the AI tool? Any third parties?
• Is the data used to train AI models, and can you opt out?
• How long is data stored, and what happens if you end the agreement?
• Who owns the responses and documents that are produced?

There is a difference between a vendor who stores everything in Europe and guarantees that the data is not used for training, and one who sends the data on to American language model providers without clear boundaries. That difference can be decisive for what you can lawfully use the tool for.

Security and certifications

Certifications are not everything, but they are a good indicator of seriousness. Check whether the vendor has ISO 27001 (information security management) or equivalent, and how they handle GDPR compliance. Ask for a Trust Center page or other security documentation, and read it.

Questions that give insight beyond the certifications:

• What is the procedure in the event of a security breach?
• How is access management handled internally at the vendor?
• Is the data encrypted both in transit and at rest?

For an example of how an AI vendor can document this, you can look at Cobrief's Trust Center. It is a good illustration of the kind of transparency you should expect from the vendors you evaluate.

Professional grounding: is the tool built for tenders?

A general chatbot (such as ChatGPT) can answer a lot, but that does not mean it is deeply grounded in the tender field. There is a big difference between a tool that is built for public procurement, and one that is a chatbot with a tender filter on top.

Ask specifically:

• How many supplier companies use the tool for public tenders?
• How long has the vendor of the AI tool worked with tenders specifically?
• Does the tool understand Norwegian and European procurement regulation, including the Public Procurement Act and its regulations?
• Can you see a demo with a real tender document from your own industry?

A tool that is built for tenders knows the language, the requirement structure, and the document patterns. A general tool has to be taught by you every time.

Hallucinations and quality control

All AI models can make things up. The question is what the vendor has done to reduce the risk. That is what you should ask to see concretely.

• Does the tool always work against a knowledge base, or can it generate freely?
• Does it reference sources, so you can verify what is stated?
• Does it have built-in quality assurance steps that check things like requirement coverage, facts, and structure?

A dedicated AI bid manager with a knowledge base anchors the AI in the company's own documents, instructions, and previous responses. That reduces hallucinations because the system works against real sources instead of guessing. The same kind of system can also perform quality assurance in several layers, including checking that all requirements are covered and that the content hangs together. That is the kind of functionality you should look for when comparing tools.

The vendor's approach and values

A good AI tool is not enough if the vendor behind it is not trustworthy. Check how the vendor talks about their own approach to AI.

• Are they transparent about which models they use, and where they run?
• Do they take privacy seriously, or is it something that "will sort itself out"?
• How do they handle the risk of hallucinations and errors?
• Are they building their product to solve a real problem, or to be able to call themselves an AI company?

For an example of how an AI vendor can describe its own approach, see the article Our approach to AI. It covers data storage in Europe, data ownership, source references, and how AI is integrated in the product. It can be a good template for what kind of questions you should ask the vendors you are considering.

How to test the tool before you decide

Demos and trial periods are underrated. It is easier to discover weaknesses in a tool by using it on a real competition than by reading a product brochure.

Practical things you can do:

• Ask to test the tool on an actual tender document from a previous or existing bid
• Compare how much time you spend on each step compared to what you usually spend
• Check how much you have to correct after the AI has delivered a draft
• Ask for reference customers in the same industry, and talk to them directly
• Also test support and follow-up: how quickly do you get a response when you have questions?

What you are looking for is not a perfect tool. It is a tool that fits how you actually work, with a vendor you can trust over time.

Closing thoughts

Choosing an AI tool for tender work is a strategic decision that affects more than a single competition. It is about how the team works, how the data is handled, and how much you can trust what is delivered. Spend time on the assessment, feel free to ask more questions than you think you need, and talk to reference customers before you sign.

We at Cobrief are happy to talk about what is important to consider, even if you are not going to choose us. It is better to help you ask the right questions than to push a tool on you that you do not need.