Simplify the quotation process with generative AI

As a company, you can successfully sell products without producing any - the reverse is not really true. Without wanting to diminish the performance of all the work in design, development, production, service and maintenance: No company becomes successful without a successful sales process; everything else can perhaps be outsourced somehow or even left out altogether.

This makes it all the more remarkable that sales processes in many companies still offer a lot of room for optimization, especially through the better use of information that has long been available in the respective company.

Challenges in the quotation process

One example: Suppliers of heating and ventilation technology or building materials make their biggest sales primarily in large construction projects. As a rule, several companies bid for general contractor status for large construction projects, and each of these companies uses the tendering body's specifications to obtain offers from many suppliers and service providers for the individual trades. For the suppliers, this means that they often receive several requests for quotations for a construction project from different potential general contractors and have to respond to all of them. Unfortunately for the suppliers, the general contractors make it easy for themselves: they simply send the entire bill of quantities - often with hundreds of pages - to many suppliers and simply ask them to offer everything that the supplier can offer. This is inefficient in so many ways that it alone may explain many a construction delay.

The solution

Fortunately, this process can be greatly simplified with a little artificial intelligence, and in several places. Let's start with the general contractors: Of course, it's easy to send the same bill of quantities to numerous companies, but with targeted requests for individual trades, you could have a much better control effect. Thanks to Generative AI, this hardly takes any effort nowadays. "Extract the items for the drywall construction from the existing specifications, create a new set of specifications and send this as a request for quotation with a deadline of August 27 to all drywall construction companies that are rated with at least 4 points in our supplier database." That's it.

For this to work, a quick exchange with ChatGPT will of course not be enough. At the very least, the large language model used must be integrated with the supplier database and the communication process automated. But regardless of whether this is a full-blown ERP system or an Excel spreadsheet: this integration takes place once, leads to faster processes, faster delivery of offers and, incidentally, better data about the suppliers.

If the general contractors do not take care of this, the task remains with the suppliers. Similar game: instead of manually searching every incoming bill of quantities for items to which the supplier could contribute, we again leave the AI the task of determining the relevant part, i.e.: "Extract the items for fire protection technology from the bill of quantities" and we only have two pages to evaluate instead of 200. For these 2 pages, the AI can then also determine whether we have already prepared a quotation for this exact project for another company, in which case we can simply reuse it.

Here too: There is a bit more to do than just chatting with ChatGPT, especially if a quote is to be generated immediately after extracting the relevant passages from the bill of quantities. For end-to-end support, the AI-based chatterbox should be connected to our product database, because language models know nothing about the actual availability of components, their prices and their dependencies on other parts. To ensure that the connection between the product database and the service specifications works, it is advisable to integrate special vocabularies, such as the standardized service descriptions from Datanorm.

But here too, this integration with the databases and/or ERP systems takes place once and then not only saves time on an ongoing basis, but also offers enormous potential for optimizing the content of the quotations.

The product database remains as it is, but a knowledge graph can be used to record in granular form which quotation was created for which customer in response to which request, which additional items that were not requested were offered and whether the quotation was successful. This provides the necessary database to continuously improve the offers: Customers who bought A also bought B - whenever I offer C, I don't get an order - whenever I offer D but don't offer E, a complaint comes later and so on.

And Generative AI helps us to extract the relevant parts from requests that are far too long and to cast the items to be offered in the correct offer form, including an accompanying letter if required.
This drastically reduces routine work and optimizes the sales process based on empirical data, even when experienced colleagues are on vacation.