A news item about AI

The recent news item that few people will have missed was the launch of DeepSeek, an artificial intelligence (AI) tool set for use in commercial environments. Its total cost of ownership is reported to be substantially lower than the cost of currently available AI products developed in America.

This announcement provides a long-term benefit for businesses, because it shows that development of AI tools can happen anywhere where a good quality technical team can be assembled (and in the case of DeepSeek, an equivalent humanities focussed team). Development also requires sufficient financial investment, although not the enormous amounts spent by early AI developers.

The business model of large US based technology companies concerning AI is now under threat from a competitor that needed to use lower powered computer chips and therefore people’s brains were applied to the challenge of developing a different approach to AI.

Benefits for end users

For end users (including supply chain professionals), an important factor has been to make distribution of an AI product ‘open source’, under an MIT licence. This requires that the source code is provided free of charge to customers (likely to be developers of applications), with documentation concerning construction of the software and models. The software can be improved and additions made, which are provided to the technical community to grow the product.

Using open-source as the distribution channel for a software product enables low-cost promotion, sales and sales support for an AI business. Collaboration across application developers is attractive as it enhances trust through the user community, provides faster enhancement of the product and a quicker cycle for bug fixes.

This AI tool (and those from Mistral AI in Europe) can be incorporated into current and new products sold and supported by software applications providers. While the AI product is ‘free’, there will be costs to incorporate it into an application, build models for end users and provide on-going support. However, for end users the overall ‘total cost of ownership’ (TCO) will be lower than the cost to develop AI based ‘solutions’ in-house.

The focus will be on the needs of all enterprises, regardless of size. That requires building a variety of models, each suited to a specific task and structured to process specific data sets of an industry, business or function. This approach requires less resources to train a model. For example, the approach of DeepSeek is to provide text-based large-language models where computing power is supplied for the few ‘expert’ tasks, while less-critical tasks will be less trained. The models will therefore provide a high capability in selected areas that are important to end users.

This approach is called Narrow AI, where models are trained that address the specific needs of a business. General AI (Gen AI) will still be required for large and challenging problems, but most (or all) needs through supply chains will not require this level of power and complexity.

To provide better protection for corporate data, AI models for business must be capable of being hosted within an organisation’s internal (and controlled) computer environment, without any connection to the Cloud. Mistral AI has this facility and some commentators have identified that DeepSeek also has this capability. Eventually, as computers increase in power and model size is reduced, users should be able to run large-language models (LLMs) in house.

AI use in supply chains

AI is an umbrella term for a group of technologies and tools, including robotics and analytical systems, such as machine learning (including pattern recognition), deep learning and natural language processing. These are identified by either algorithmic driven (select a ‘maths’ routine to find an answer) or those that are data driven (find patterns in a set of data).

The more likely requirement in supply chains for AI tools is to capture and analyse transaction data and data from machines and other equipment to identify patterns. To efficiently achieve this, the data must have automated collection, be catalogued and scalable, with data mining enabled.

An organisation’s supply chains are a ‘complex adaptable system’ with many interacting elements, which make it near impossible to predict the behaviour of the system. So, planning supply chains at the tactical level is not repetitive work, nor a ‘real-time’ process. It is therefore preferable that AI is not used to develop the ‘perfect’ plan or outcome, as detailed accuracy to enable ‘better decisions’ is not required. Instead, AI can be used to provide information for supply chain professionals and other decision makers about trends and identifying constraints in the process. Then provide planning and scheduling options (with reasons), that enable knowledgeable people to improve their decisions.

Likewise, If there have been attempts to integrate APS (Advanced Planning) and ERP (for operations scheduling), then also question the value of running AI tools. Doing so will likely introduce noise and error into the complex adaptive system, which makes the system more nervous, increasing the ‘bullwhip effect’.

It is best to introduce AI in supply chains where the are defined patterns. Examples in Operations are: inventory planning and control, predictive maintenance, warehouse management, freight management and fleet management. Also purchase orders, invoices and compliance records.

The release of DeepSeek and the availability of its methodology will spur competitors to commence AI development and application providers to provide focussed applications across many areas, including supply chains. But that will not happen overnight. The Gartner Hype Cycle showed that by 2024, AI had reached Stage 2 – the Peak of Inflated Expectations. There are a further three stages in the cycle before AI may achieve general acceptance and Gartner expects this to be around 2030, when climate change is likely to be getting all the headlines. So there is time for your organisation to evaluate the data and information needs of the supply chains, learn more about the capabilities of ‘low-cost’ AI and evaluate prospective application suppliers.