Article written in collaboration with Pixis Conseil
Sundhar Pichai, CEO of Google, recently said that AI will have a greater impact on society than electricity or fire. The contribution of AI to the global economy is estimated to be around $15 trillion by 2030.
Since the mid-1950s, AI has had some impressive successes (chess, go, machine translation, language understanding, autonomous driving, chatbots) but their day-to-day applications remained relatively unclear and their societal impact marginal.
The first significant shift in artificial intelligence is actually quite recent, coinciding with the multiplication of the data we produce and the explosion in the computing power of our equipment. We are talking about the datafication of our society and Machine Learning (ML).
This shift is crucial because it allows us to move from the "rear-view mirror" management of our activity that has prevailed until now (learning from the past so as not to reproduce it) to a predictive vision: knowing what will happen and acting accordingly. The opportunities for Machine Learning seem infinite.
Prerequisites to master to get the best out of AI
The AI fantasy machine has been revived but the technological reality still has some limitations:
1. Data availability
- The data available is not always of sufficient quality and quantity to develop a successful AI. For example, the Stanford University AI developed to diagnose skin melanoma is as successful as a panel of 21 oncologists. But to do this, it needed a huge database of 130,000 images to develop its learning.
- In addition, the data may also be biased. For example, sales from one year to the next may have been influenced by random or integrable factors (weather, strikes, etc.) that are not always easily identifiable. Thus the predictive power of AI may be distorted by these biases which it is not always possible to neutralise.
- The other major bias of the data is that the data, generated by our behaviours and interactions, is likely to reproduce the discriminations of our society. In order to create equitable artificial intelligences, it is necessary to clean up the data that feeds the AI so that it does not reproduce these biases (ethnic, sexual or religious discrimination) on the exponential scale that it does.
2. Scaling up AI algorithms
The mathematical performance of machine learning algorithms cannot be an end in itself, the issue is the performance of the algorithm, its efficiency and its resilience when subjected to a colossal flow of data. This skill, which enables us to move from a prototype to an industrial application, depends on a high degree of maturity in the management of AI, its algorithm and its infrastructure. This maturity is still largely lacking.
According to a Gartner study, 85% of AI projects never get beyond the prototype stage.
This ability to do :
- business predictions from the data,
- at an industrial level, i.e. capable of generating a reliable ROI in the long term,
- do so in a timely manner
- at a reasonable cost is the real challenge of artificial intelligence today.
This is possible, especially for retail, thanks to MLOps.
A contraction of Machine Learning and Operations, MLOps is a set of tools and practices that allow Machine Learning projects to move from the prototype stage created from a controlled amount of data to the stage of large-scale operations, processing exponential streams of data. As a result, the AI industry is finally coming of age. MLOps is now the sine qua non that will finally allow the massive deployment of AI by companies.
AI meets the retail world
This large-scale prediction tool allows four main types of applications particularly adapted to the supply chain, customer relationship management or customer / point of sale marketing:
- Operational excellence
- The creation of innovative products
- Risk management
From these four main families of applications, an infinite number of actions can be undertaken on a case-by-case basis, including
- Supply chain: the use of AI to predict flows within the supply chain would allow the optimisation of upstream/downstream order volumes and the sizing of teams (in particular in the warehouse) by taking into account variables such as the weather or consumer trends
- Customer relationship management: handling the simplest customer interactions would relieve human operators of the burden of handling the most complex cases
- Customer/Point of Sale Marketing: Analyzing customer flows, evolving the layout and predicting which products a specific customer would be likely to buy the next time they make a purchase (in-store, online...) would boost conversion rates and grow in-store revenue/m2.
The average shopping cart of e-commerce sites could also be targeted. For example, a customer advisor in contact with the customer could propose complementary products based on the recommendations of an AI that would have analyzed the customer's profile and history.
Identify what you want to target
Because of this proliferation, it is essential to choose the right application cases to aim for a tangible return on investment.
As an example, the startup Craft AI has developed a 6-step methodology to support its customers in record time:
To be able to use AI in your processes, you will finally have to consider these last conditions:
- What data and what performance can we expect from AI?
- What data is available? In sufficient quantity and quality?
- What level of prediction is expected?
- Is this enough to make a POC?
- What benefits are expected? What is the scope to justify a POC?
- What business plan for AI projects? What ROI?
It is important to have cross-functional teams from the start of the project to have a global vision of the problem and to develop an efficient AI.
Fears about AI
Often perceived as a replacement for existing teams, it can inspire fear. AI is at the service of decision support teams. The confidence of the teams in the results provided and the clarity of the models are paramount.
Pixis Conseil is an independent consulting firm, expert in retail strategy and organisation, created in 2005. It is made up of experienced men and women, with a double operational and consulting experience, to bring the necessary efficiency to our missions.