Optimal optimisation with AI: How to get a chatbot without expensive subscription costs

Never skimp on equipment – said no CFO ever. Because you should only have the equipment you actually use, at a price you are willing to pay.  

This art – finding the right balance – also applies to AI. In a business context, AI solutions often come as part of an entire suite of productivity tools that your company needs to purchase.

And it can quickly become expensive. 

So how can you provide your employees with the tools they need that will make a difference to their efficiency, while keeping license expenses down?  

At Basico, we have approached the task exactly as we would if we were advising a client. We have analysed what the best solution would be for us, and as part of this analysis, we have calculated the costs for two different options: 

This calculation has led us to choose not to invest in productivity tools from the major providers. We have assessed that what most employees use from the new technology is the raw language model – the one that can evaluate content in a text, help write a memo or provide creative input in a thought process.  

This also means that we have determined that employees currently do not use or need the (still largely experimental) integrated solutions available from especially Google and Microsoft – which we naturally would not get without purchasing their suites.  

Instead, we have built an interface on top of the raw language models that exist in the cloud – for example, on Azure – so employees can access a language model anytime and anywhere. Additionally, we have connected some relevant functions ourselves, and we have set up some specialised chatbots for employees.  

The solution is private, as long as we use cloud solutions that are secure, and we can create audit trails on usage and control the prior knowledge the language model should have about, for example, our company.  

This generally means that we have more control over the solution – we can choose which models we make available to employees, and we can control who, when and how access is granted and to what, all within our own environment.    

The price for creating a private model naturally depends on the level of ambition for the solution. But the very basic language model can actually be set up quite simply and without major costs. And once it is set up, your AI no longer runs on a price per user, but a price based on consumption. And that is the biggest difference.  

Let us now dive into the calculation and analyse the concrete figures.

Below, we have taken the consumption figures and prices that we at Basico have paid for the fall of 2024 and transferred them to a company with 250 employees:

Taking an average of the list prices for productivity suites from the largest providers right now, they cost 166 DKK per month per user – or 1,992 DKK per year. For a Danish company with the mentioned 250 employees, this would represent an additional cost of 498,000 DKK per year. Not an insignificant amount for a company with many employees.  

At Basico, approximately half of our employees use a GPT4-size language model daily in their work, and they write an average of 4.6 messages to our AI assistant daily. For each message, we pay 0.23 DKK in raw consumption costs, which amounts to 23.07 DKK per user per month.   

This is – as you have probably already figured out – only about 14% of what an average suite solution costs.   

For a company with 250 employees, this amounts to an annual expense of 69,210 DKK – quite far from the 498,000 we would have to pay for a productivity suite from one of the largest providers.  

At the same time, with this solution, we only pay for actual consumption. If only half of the company actually uses the solution, then the expense becomes only half – 34,605 DKK annually.  

Had we chosen a suite solution, we would either have to pay for all employees, regardless of whether they used AI or not, or choose that some employees should not have access to the solution. With a custom-built solution, we can both give all employees access to AI and keep the expense low.   

If we invest in developing a private, consumption-based solution, the cost can be shifted to an investment case, as opposed to the subscription to a productivity suite, where the fixed, flat license costs are an ongoing operational expense.