In games, artificial intelligence is on a winning streak against the best human brain power of the world. In 1997, IBM’s Deep Blue already beat Gary Kasparov, the chess world champion. In 2016, Google’s AlphaGo beat grandmaster Lee Sedol in the much more complex game of GO. The past year, Deepstack and Libratus – two AI algorithms – have beaten top-ranked professional poker players in 2 person no-limit Texas hold-em poker. It is only a matter of time before AI is deployed in the yearly negotiations in food retail and FMCG. After all, these negotiations can be considered as the most exciting high-stakes poker game in the world, with billions of dollars on the table.
AI Evolves Exponentially In The Area Of Game Theory
Chess and Go are perfect information games. At any given moment, all information is available and visible for all players. Gaming AI’s calculate the possible results and rank strategy options based on the history of previous games. In 2017 Google demonstrated a new revolutionary approach. Its AI Alpha Zero was programmed only with the chess rules, no historical game data, no chess strategies. It was given 4 hours to learn how to play chess and afterwards beat Stockfish, the reigning computer chess algorithm. This illustrates the state-of-the-art of AI speed of learning and the fact that AI can come up with different and better solutions than the human brain.
Poker is an entirely different animal, much more similar to retail negotiations. Bluffing is often employed as a tactic. Information is imperfect and changes constantly, since both players have two initial private cards and at the introduction of each set of public cards, each player is asked to bet, hold or abandon the money at stake on the table. Poker theorist David Sklansky stated that if you can play poker as closely to the way you would play if you could see everyone else’s cards, you will win money. So winning in this type of game is all about filling the gaps of missing information, and acting on this data intelligently.
The Deepstack poker algorithm only calculates a few steps ahead and continuously updates its algorithms as new information is acquired. Very soon, this type of poker bots will be very competitive in the multi-player poker world series.
Robotic Negotiations In Retail Will Become The New Normal
In a business context, winning poker games is not a terribly valuable problem to solve. Researchers are more and more interested in using AI for business situations such as retail negotiations or buying real estate. This evolution will follow three distinct phases
Phase I : AI-assisted Negotiators Explore The “Negotiation Space”
In FMCG, negotiations between buyers (retailers) and sellers (brand owners) are very complex. First of all because there are many topics, issues and possibilities. Secondly, the rules of the game are not clear, nor are they agreed to in advance.
Data Models Will Reveal The Limits Of The Negotiation Space
The first hurdle is the complexity of multi-issue negotiations. On a high level, one can say that negotiations are about (de-)listing of assortment, promotions and net-net pricing. Moreover, retailers manage consumer price levels as a lever for growth and profitability.
In reality things get even much more complex. For example, assortment discussions involve breadth and depth of assortment of the brand owner versus the competition including private labels. Share-of-shelf and shelf position are key parameters. Furthermore, retailers are interested in exclusive SKU’s and brands to be able to differentiate themselves in a competitive retail landscape.
Price discussions do not only focus on net-net pricing. Various components of pricing are also crucial: recommended consumer price, invoiced pricing versus periodical trade terms, conditional versus unconditional pricing, promotional pricing and participation of both parties in the promotional price decreases, folder and display budgets…
We are also in a situation of incomplete information. Buyers do not always have a full understanding of brand profitability, in-depth category dynamics (like cannibalization between SKU’s) or options for exclusive brands, variant, promotions or other advantages.
Buyers also lack knowledge on their situation relative to key retail competitors. Sellers lack knowledge on retail profitability and do not know the offers of other branded competition. They do not know the state of the detailed future promotional plans of all players.
Today, even the best-in-class companies hardly leverage advanced modelling to simulate different negotiation scenarios. It is even less common that scenarios include simulations of the impact on the negotiation partner in terms of key parameters such as growth, profitability and competitiveness. Negotiations are done in an invisible space, with limited preparation and limited information. Win-win strategies are vague intentions or emotional statements, rather than a calculated expression of a real scenario.
In phase 1, AI can assist negotiators to simulate different scenarios on multiple dimensions. The negotiation space will become more visible. In first instance, alternative options will be simple, even simplified for instance focusing on one category only. Later on, interdependencies between different categories will be included (e.g. shifting category shelfspace) or full store modelling will allow to predict competitive strength versus adjacent stores. The model parameters will be fine-tuned along the way, based on information exchange. Parties could also decide to run tests in order to improve their models. At this very moment, Carrefour France is testing an assortment with 80% private label SKUs in four stores. This will allow them to model cannibalization and store attractiveness.
Amazon is constantly improving the parameters of their models through thousands of personalised A/B tests per week. This allows them not only to improve consumer loyalty but also to become a fierce negotiator, having access to more of the hidden information.
The Rules Of The Negotiation Game Are Unclear And Can Change Along The Way
The rules of the game are not clear. Very often there is no agreed priority of mutual KPI’s to be used as a measurement of success. Sometimes retailers communicate their priorities. However, in most cases this offers little guidance. If retailers declare they want to have lower net-net prices – the best of all competitors, less branded SKU’s in order to reduce complexity and will focus on listing exclusive innovations, then little room for negotiation is left.
Priorities – real or communicated – can also change during the negotiation after exchanging the first bids or under time pressure. Furthermore, the process of negotiation can move from open and co-operative to closed power play, including threats, deadlines and limited communication.
Phase II: Centaur Teams Will Take Over
20 years ago, after the supremacy of chess bots, a new sport was created: advanced chess. Over time it became visible that a team formed by a chess player with a chess computer always wins from the computer. The centaur team was born. With the exponential growth of business process automation in the past years, the quality of the interaction process between man and machine proves to be crucial for success. A key skill and development area for humans will be how to instruct, use and cooperate with artificial intelligence.
These centaur teams will lead the negotiations. On top of content, a new dimension will be added: AI-based negotiation strategies. Textbooks have extensively described negotiation strategies, many hours of consulting are spent on preparing for negotiations and how to respond to every tactic. These strategies can be improved by AI as well. Moreover they can be fully personalized to perfectly fit the underlying personalities and track records of the negotiators. Simple apps already exist today that analyze personality and optimize communication style, such as Crystal Knows. Crystal Knows scrapes the web for data regarding your interactions on linkedin, facebook and other public sources. Facial recognition algorithms by Microsoft and Google are already able to read emotions from the face of negotiators. Imagine a tool with a more precise input and more efficient feedback loops with the sole objective to optimize your negotiation strategy.
Phase III: Machine-to-Machine Negotiations
In a final phase machine-to-machine negotiations will become a reality. This will allow to take many inefficiencies out of the process, such as human emotions and it will minimise the information gaps. As we have seen with Google’s Alpha Zero, AI might come up with a level of creativity beyond human potential. The negotiation space will expand.
It now depends on the programmed ethics if the negotiation robot will only try to maximize its own benefit or if it will take into account the interests of its opponent in order to reach faster, a fairer and more long term sustainable solution.
About Nils van Dam
Nils van Dam is a seasoned business leader, with more than 30 years of experience in the FMCG industry. He has occupied senior roles at global, regional and local level at Unilever, AB-Inbev and Censydiam in marketing, sales and general management. In his last role, as CEO Unilever Belgium & Luxemburg, he has lead the digital transformation in order to build a future-proof company. Nils has a passion for marketing, change management and business transformation.
In 2018, Nils became Global head of the Food, Beverage and Food Retail practice of Duval Union Consulting. He is also partner and non-executive director at Jacoti, a new tech company in the hearing aids industry and non-executive director at the Brewery of the Trappists of Westmalle.