Inspired by the seven “wastes” popularised in lean manufacturing, Artefact has adapted this concept to the field of artificial intelligence. This study is based on more than 30 artificial intelligence projects over the last three years.

Inspired by the seven “wastes” popularised in lean manufacturing, Artefact has adapted this concept to the field of artificial intelligence. This study is based on more than 30 artificial intelligence projects over the last three years.

For each identified “waste” – extra processing, energy, inventory, overproduction, waiting, failures and talent – the causes are explained through concrete examples.

Extra processing: Over-delivering against customer expectations

AI for the sake of AI

Making AI because it’s the new and shiny thing. This occurs when technical teams work to build an efficient and innovative but complex technical solution. Business considerations are relegated to the background in favour of the technological challenge; technical specifications take precedence over imprecise business specifications, and the core value-generating features are not clearly identified. Extra- processing often results in complex “black boxes”

which need to be maintained by technical teams and only then understood and adopted by business teams.

“Reinventing the wheel”

Building custom AI is time-consuming. Sometimes a small, perfect detail which the client hasn’t prioritised and isn’t expecting becomes the focus of a project. When this happens the true purpose of the project is abandoned in favour of solving minor problems.


An insurance company wants to create a churn prediction algorithm. Traditional off-the-shelf algorithms allow specifications to be met. However, the data team wants to design a personalised deep learning solution. The algorithm they have designed is a “black box”, which takes much longer to evaluate and understand.

An FMCG company wants to set up an allocation algorithm to help manage media buying. Rather than capitalising on traditional attribution models (last touch, linear, first touch, etc.), the company wants to set up a data-driven allocation model, delivering an unnecessarily complex model which is difficult to understand and use.

Energy: Human or machine efforts that do not increase profits

Lack of integration

Often companies lack a technical AI ecosystem and have legacy IS and tools in place. The different components needed to build the AI solution are not integrated, increasing cost and time in the development process. On a human level, skills are not centralised and communication is not fluid, resulting in misunderstandings and wasted time.


A retail company wants to set up a customer recommendation algorithm. The data necessary for the project (CRM, transactional, navigation etc.) is scattered over several databases without centralised access. The time-consuming task of centralisation is necessary before launching the project.

To benefit from increased computing power, the data team decides to train the algorithm on the cloud, and to deploy to a local IT infrastructure. The technical ecosystems are not already integrated, so extra integration work is required.

Inventory: The creation of non-autonomous intelligence

Specific versus standardised culture

Models are designed individually to meet a particular need. It is not possible to capitalise on what has been built before so each new need is met starting from scratch, slowing down production and development of new models.

The burden of past choices

Poor initial technology choices result in a data product that requires substantial maintenance. The team spends too much time maintaining the existing system rather than iterating and improving.


A cosmetics company wants to create a Natural Language Processing (NLP) solution to analyse sentiment on social networks. Rather than using an existing component and adding a Specialised NLP element, the solution is built bespoke. Three months later, another team wants to use NLP for another type of analysis; it is forced to start from scratch, creating more AI “inventory”.

An Internet of Things ( IoT) company wants to understand how consumers use its connected devices, the data coming back to the company however is erratic and poor quality. The company decides to outsource the implementation of a new data product to address this issue and allow the data to be processed in real time as well as anticipate any quality problems. Upon delivery of the new data product the internal teams are then unable to maintain the new (specific and complex) solution. They have too many specific, separate and complex AI systems.

Overproduction: AI arrives too early for a company’s maturity level, or in an unstructured way

Too many non-industrialised Proof of Concepts (POCs)

With no integrated strategy, many different initiatives (POCs) outside the main purpose of the company pop up. These initiatives often arise in anticipation of customer expectations and many respond to the same problem.


An FMCG company decides to launch a project team focused on artificial intelligence. The team’s first action is to organise an ideas workshop to identify use cases for further development. The submitted use cases focus on promotion and marketing, and other valuable areas are overlooked, including supply chain optimisation, energy consumption reduction and in-store cost reduction.

Waiting: Unoccupied time, waiting for a delivery or a managerial decision

Lack of C-level sponsorship

Artificial intelligence is not considered a strategic tool by the company’s managers. The time spent convincing management that artificial intelligence can create a competitive advantage leaves little time for action.

The weak sponsorship of the leaders leads to a lack of participation by all teams. Processes are not measured, leading to quality problems, defects and technical bugs. Processes are also not fluid, with significant time lost waiting for approval between successive stages.


A company in the travel sector wants to develop a voice app on Google Home and Alexa. Three months are devoted to the identification and mobilisation of technical resources. A clear need to increase the skills of the team is identified. The company decides to use external resources, which are then put on hold for one month until access is granted to the databases and the technical environment. Once the project is ready for launch, it is paused for several weeks, due to a delay in the CEO signing an NDA.

Failures: The AI product does not meet the needs of the final customer

A focus on execution speed versus quality and consistency

  • Poor understanding of business needs; specifications not accurate enough at the beginning of the programme
  • Non-standardised processes; poor bug detection during the production phase
  • Broken features as infrastructure, and connectors to other systems are not updated
  • This may be caused by rewarding teams on quantity and speed of execution rather than on quality


A company in the food and beverage sector wants to replace its software for diagnosing failures in its refridgerated supply chain. They had previously used a human built deterministic decision tree to analyse failures but wanted to replace this with a machine learning algorithm. Rather than letting the algorithm define the most relevant solution to solve the problem, the data team limited the algorithm by including human developed constraints. At the end of the programme the results are the same as the previous system.

People: Failure to capitalise on internal skills

Lack of an inclusive culture with humans at the centre

To succeed in its AI transformation, the company must recognise the strengths and contributions of each employee, and make the AI transformation an inclusive one. Today, too little time and resources are devoted to increasing the skills, training and development of employees, making deploying AI solutions challenging and limiting the transfer of knowledge.

Three years after the launch of a major AI transformation plan, a banking company consults its employees. Three main points come back from the consultation:

  • Problem solving done in the room by the experts, ignoring the contributions of the other team members
  • Improvement ideas imposed by management or experts on the rest of the team rather than designed with them
  • Products designed “on the cheap” due to a lack of skills and knowledge
  • Teams not incentivised enough on the delivery of the project
Artefact Newsletter

Interested in Data Consulting | Data & Digital Marketing | Digital Commerce ?
Read our monthly newsletter to get actionable advice, insights, business cases, from all our data experts around the world!