Since the day we introduced Pactum to the world, investors have tried to understand what we are. Is Pactum a chatbot company? Or does it belong to the Robot Process Automation category? Is it legaltech, procuretech, AI or enterprise SaaS? For us it’s always been the same answer – we are all of them. It’s the new category Pactum created – Autonomous Negotiations. Back when we first started the term was not even coined yet. We even registered the domain www.autonomousnegotiations.com because the concept was so paradigm-shifting.
Something that transforms the world as we know it requires a new category as hundreds of new companies will be following the buzz. It’s a category that not only will benefit millions of people but will also transform how global commerce is done. Humans work in collaboration with trillions of bots who never sleep and keep figuring out the deals that are most beneficial for both parties. Negotiation is a form of art that is creating value through understanding each other and coming up with trade-offs that reach us closer to Pareto Optimum deals. Our mission in Pactum is to let the world know that negotiations create value and AI brings that value to everyone.
It’s been a busy 4 plus years working towards the mission and as of today we can say that we have reached a level of maturity where we have confidence in understanding what the key building blocks of Autonomous Negotiations are. Since Pactum still has no competitors, we feel that we have an obligation to reveal what is under the hood at Pactum, to inspire new autonomous negotiation companies to be created alongside us.
The center of everything is Autonomous Negotiations between the AI agent and the supplier. You can imagine thousands of agents collaborating with the suppliers 24/7 to reach the best deals. It’s like a well-orchestrated symphony conducted by a team of virtuoso musicians. Just as each musician contributes their unique expertise to create a harmonious masterpiece, autonomous AI agents orchestrate negotiations seamlessly, leveraging their individual data points and specialized knowledge to achieve optimal outcomes in a synchronized and efficient manner. In this analogy, negotiations work in harmony, guided by the precision and expertise of AI-powered bots, resulting in a masterpiece of successful deals and collaborations.
To make this possible, our clients have two user interfaces – Negotiation Suite and the Supplier Portal. The Negotiation Suite is the client facing interface for their users to configure, launch, manage, access product modules, and see the results of their supplier negotiations. The Supplier Portal is a client-branded interface that engages in real time negotiations. These negotiations are specific to that supplier, taking input from the Negotiation Stack, AI Engine and automation setup, and are adjusted in real-time based.
We have also built two interfaces for internal use. The Negotiation Designer is the first ever software built to build negotiations. Our negotiation designers, professors and data scientists are all using the technology to set up new use cases, A/B tests, and scale improvements that come through machine learning algorithms. The closest software to our Negotiation Designer is used in the video game industry where, although someone is playing in a fixed space (i.e. contract space), every move and decision will open up new paths (i.e. counter offers) while the entire experience must conclude through a clear storyline (i.e. final agreement). The other piece of software we use internally is Workspace. This is specifically built for our deployment and operations. The Workspace is used by Implementation Managers to configure the settings per client (for example, threshold values or e-mail sequences) and Support will be there for any client user or supplier who requires manual intervention.
Let’s now cover the four core pillars of core technology: Negotiation Stack, AI Engine, Automation and Infrastructure.
The core DNA of Pactum is its Negotiation Stack. Pactum negotiations are built of Flow Components (such as “Multiple Equivalent Simultaneous Offers component” or “Commodity-based argumentation component”) that in combination create the world’s largest set of predefined autonomous negotiation use cases. It’s called the Use Case Catalog. Each use case solves specific client problems. For example, Spot Buying Purchasing involves a supplier negotiating a quote before a Purchase Order is released, without human involvement. While Tactical Sourcing use case requests quotes from multiple suppliers in a competitive environment and gives an opportunity for the buying manager to select the winner. These different use cases are grouped into Product Modules that clients can purchase and access via Negotiation Suite. As a co-founder, one of the most challenging parts of the journey had been discovering all the potential applications where the technology makes sense. That is when we faced a positive problem – the technology works across the board. So, what use cases are worthwhile to pursue? You can read more about the Use Case Catalog in this blog post.
The brain behind the Platform is the Artificial Intelligence Engine. “Artificial intelligence (AI) is the ability of machines to perform tasks that are typically associated with human intelligence”. So, AI is like an independent agent who is also at least as smart as the person next to you. Let’s go under the hood and ask – what is needed for this AI to be independent and smart in the domain of autonomous negotiations?
To succeed in enterprise settings, the autonomous negotiations platform needs to use the most modern technology stack. The stack can be divided into two distinct pillars: rule-based AI and generative AI or the subdomain of it – Large Language Models (LLMs). At Enterprise AI business settings both AI stacks are needed for guaranteed scale and governance. Unlike the black box of generative AI, the rule-based AI is explainable and better suited to business-critical negotiation strategy decisions. But Generative AI’s power comes in creating compelling negotiation dialogues with users.
To be more specific, the rule-based AI calculates offers, samples thousands of contracts, creates new counter offers (Multiple Equivalent Simultaneous Offers) and communicates with suppliers via structured and legally approved negotiations scripts. While LLMs enhance supplier profiles with data, find the negotiation opportunities and enable personalized communication of these to thousands of supplier users at scale. The results can be enhanced using LLMs as each supplier has a unique personality and business context. While it is important to remember – as an enterprise, you need full control of everything that goes out with your name. All the communication language is reviewed, and every offer needs to be within the accepted boundaries of the contract space.
The core AI instruments that set the boundaries are: Value Functions, Negotiation Parameters and Trigger.
First, the machine needs to understand what the negotiable terms and their costs and values to our client are. In combination, we call it a Value Function. The client Value Function is mapped for the AI to know which offers and trade-offs create how much value. Putting it another way – we don’t want the AI to offer deals that create loss. The AI Engine also has Negotiation Parameters that form the decision for each strategic path in the negotiation flow and monetary value per offer. Amongst the other things, the parameters include minimum acceptance thresholds, anchor limits and supplier profile metrics. Parameters can also be changed after each negotiation through the concept of Dynamic Pricing. For example, in Logistics Lane Rate negotiations, if suppliers tend to accept the initial offers then the anchor price will decrease, while if no one is ready to serve the lanes then the system increases the initial offers. Finally, the AI is smart if it knows when to send out the negotiation – the Triggers. Negotiation Triggers are discussed and pre agreed with the client and the AI then acts within the boundaries of them. For example, one trigger in Retail Cost Decrease use case can be “IF, commodity dairy decreases 30% within the last 4 months THEN trigger the negotiation for item-level cost decrease negotiations for items in the category.”
So, what makes the AI Engine smart here is orchestrating the combined actions of all of the approaches described above in real time. The AI Engine is responsible for executing the deterministic AI algorithms that drive the autonomous negotiations. The Engine triggers the negotiation, handles the parameters, calculates the offer values, learns about supplier preferences, chooses the right negotiation strategies and actually closes and commits to deals independently. Perhaps most importantly, the AI has plenty of time to learn about the suppliers and previous engagements, and thereby improve every following negotiation. These are perhaps the main reasons why e-auctions have failed us – in the modern world, we need communication that starts from selecting the right supplier to speak to and ends by putting together personalized offers that increase the pie for both parties.
So, you have built the Negotiation Stack and AI Engine, what are you still missing to transform global commerce as it is? It is the scale. To scale negotiations, and make them truly autonomous, we need to establish end-to-end automation. In today’s tech world, this domain is also labeled as Robotic Process Automation (RPA).
It is important to note that in some negotiations humans want to stay in the loop, while others need to happen in the background. For example, buyers may want to review new negotiation opportunities before sending them out, or they may want to select a winner after 3 suppliers have reached new agreements. While buyers may feel comfortable letting the AI initiate the negotiations and update Spend Management system in a single $1,000 dollar purchase. In either of the cases, there are countless steps that need to happen automatically.
The automation starts once the AI detects new negotiation opportunities. For example, Purchase Requisition is submitted in Ariba, or new Full-Truck-Load requisition is added to Transport Management System. Quite often the AI then needs to confirm the contact person from the supplier side who has the rights to enter to negotiations. The AI can source contacts from the client database and if needed then get alternative options from third party providers, like ZoomInfo. If the supplier’s contact person is wrong, they have an opportunity to write the name, title, and email address of the right one.
After a fully autonomous chat has taken place and agreement is reached, the AI takes the agreed terms, generates the new contract automatically and sends it to sign for both parties. Finally, the client’s database is updated in the background and all the relevant negotiation flows and contracts are uploaded as they need to.
The entire end to end process is supported by a smart ticketing and email sequencing system. If some suppliers require manual interference, then automatically tickets are created. Supplier ticketing helps to prioritize manual tasks which Support needs to handle. One of the key RPA services is the email sequencing system as when we include reminder emails, we have on average over 30 standard email templates per use case. It sounds scary, but now think again why one buyer does not find time to negotiate with hundreds of smaller suppliers.
What makes this entire platform powerful is its ability to get insights from the Data Warehouse. And not just any insights, but actual evidence to scientific studies on negotiations and around supplier’s behaviourial science. Data logic needs to be very robust to support A/B testing of theories and offers within and across the use cases. For example, we know that in product module A, we get 10% better savings when negotiations are sent out on Mondays than on any other day. Or if supplier opens the negotiation and answers “Hello!”, not “Great” to our question “How are you?”, then we should offer the supplier to come back another time or we lose 60% of them within the next 10 minutes.
Finally, a clear requirement for the platform includes data input and output integrations to ERPs, Spend Management systems, Contract Lifecycle Management systems and digital signing platforms.
To date, we own the largest repository of behavioral negotiation learnings in the world. The power of the platform requires us to handle data seriously. This includes being certified by SOC 2 and keeping client data separate.
All these components are needed for an Autonomous Negotiation platform. The most difficult part has been figuring out which components work best. With this acquired knowledge, it is our responsibility to share it with the world to encourage industry growth and inspire new companies to join the space. This is the most powerful software platform in the world for accelerating value creation. AI Agents that build up our future companies. Creating value out of thin air while we sleep.