A step-by-step guide to building a ChatBot | Conversational AI in Procurement

Conversational AI, a technology initially focused on external customer-facing processes, is now transforming back-office operations. The rise of generative AI has expanded the range of use cases, offering significant potential to automate repetitive tasks, create additional channels for information retrieval, and enhance the internal customer experience. Procurement teams often spend considerable time handling enquiries from internal stakeholders, many of which could be resolved independently. As a result, introducing conversational AI and chatbot technology can lead to substantial time savings.

A study by The Hackett Group discovered that procurement personnel dedicate up to 60% of their time to tactical activities, such as addressing routine questions from the business, which could be automated through Conversational AI (The Hackett Group, n.d.). This presents a tremendous opportunity for organizations to achieve increased efficiency and productivity by implementing Conversational AI in procurement processes.

It is important to note that the terms conversational AI and chatbots are frequently used interchangeably, but they do not mean exactly the same thing. Conversational AI encompasses the wider domain of artificial intelligence that allows machines to comprehend and respond to human language. Chatbots, in contrast, are a specific application of conversational AI designed to interact with users through natural language formats, typically via text or voice-based interfaces. Chatbots employ natural language processing (NLP) and machine learning (ML) algorithms to understand user intent and respond in a manner that simulates human conversation. Although chatbots represent just one facet of conversational AI, their increasing popularity across various industries, including procurement, can be attributed to their ability to automate routine tasks and deliver a more streamlined user experience.

The value proposition for procurement includes:

1. Enhanced user experience: Chatbots offer an alternative channel for individuals to interact with the procurement process. For instance, conversational AI can assist users during the creation of a purchase requisition (PR), a task that business users might only occasionally undertake. By using a chatbot, users can receive guidance, prevent mistakes, and obtain answers to procurement-related questions. Once the chatbot has collected all the necessary information, organisations can further improve efficiency by employing a robotic process automation (RPA) bot to submit the request on the user's behalf.
2. Boosted productivity: Chatbots can handle various tasks and enquiries, thereby saving procurement teams time. Examples include answering policy-related questions and helping users access budget and expenditure information.
3. Streamlined supplier interactions: Chatbots can be designed to face suppliers directly, enabling them to self-serve. This approach reduces the time procurement managers spend on administrative tasks and non-value-added engagements, allowing suppliers and procurement teams to focus on more strategic topics that foster stronger relationships.
4. Greater effectiveness for procurement staff: By integrating generative AI and large language models, such as ChatGPT, chatbots can produce generative responses. Although this feature may not be suitable for every situation where certainty is crucial, it can aid procurement staff in tasks like researching suppliers during the sourcing process. For instance, during a request for proposal (RFP), a procurement manager could ask the chatbot to obtain information about a specific supplier or compile a list of suppliers with particular capabilities. Tasks like these would typically require considerable time spent searching through various documents and websites.

Understanding Basic ChatBot Architecture

It's important to understand basic ChatBot architecture. By understanding basics about how a ChatBot responds to user queries it can bridge the gap between business and technology and spark ideas on potential use cases.

• Intents: These are the user's goals or intentions when they interact with the chatbot. The chatbot uses Natural Language Processing (NLP) to understand the intent behind the user's message.
• Entities: These are specific pieces of information that the chatbot needs to extract from the user's message in order to fulfill their intent. For example, if the user asks In procurement, the chatbot can extract specific pieces of information from users' messages to fulfill their intent. For example, if a user asks "What is the budget for office supplies?", the intent would be to know the budget for office supplies, and the entity would be "office supplies".
• Slots: These are the places where the chatbot stores information that it has gathered from the user's message. For example, if the user asks "Can you book a flight to New York for me?", the chatbot would need to store information such as the destination (New York) and the date of travel in slots in order to book the flight.

The chatbot architecture works as follows:

1. The user asks a question that is classified via an intent engine.
2. Follow-up questions will fill slots based on entities.
3. The ChatBot will retrieve information from multiple sources, either back-end systems, the chatbot platform database, or generative responses.

The reach of the chatbot depends on the number of intents it can understand and respond to accurately. The more intents a chatbot can handle, the greater its reach. Similarly, the more entities a chatbot can extract, the more personalised and effective its responses will be.

Types of ChatBot Responses

Chatbot responses can be categorised into three main types: hard-coded responses, data-driven responses, and generative responses. Each type has its advantages and limitations, depending on the specific use case and requirements.

Hard-coded (text-based, radio buttons, links):

Chatbots with a natural language understanding (NLU) engine use hard-coded responses like text, radio buttons, or links for predetermined answers to specific user inputs. The NLU engine processes user inputs, allowing the chatbot to comprehend the conversation's context. The chatbot selects a hard-coded response based on the identified intent, providing a structured and controlled conversational flow. However, this approach lacks the flexibility of advanced, generative models.

Data-driven:

Data-driven chatbots retrieve information from back-end systems like databases or APIs. They often combine rule-based or generative techniques with data retrieval, providing users with accurate, up-to-date information. Data-driven chatbots are suited for tasks requiring specific, dynamic data.

Generative:

Generative chatbots, like GPT-4, use machine learning algorithms based on natural language processing (NLP) and natural language generation (NLG) techniques. They generate responses by predicting appropriate word sequences based on user input, enabling more diverse and contextually relevant replies.

Conclusion: Understanding the different types of chatbot responses is essential when selecting the best approach for your specific use case. While hard-coded responses provide a more structured and controlled conversational flow, data-driven responses offer dynamic information retrieval, and generative responses enable a more flexible and contextually relevant conversation.

Use cases and example Intents for conversational AI in Procurement

Procurement Policy & Process

1. Locate specific procurement documentation
2. Retrieve information about the procurement process
3. Find detailed information about a specific step in the procurement process
4. Retrieve evaluation criteria for suppliers
5. Retrieve an overview of the procurement system landscape

Order Management

1. Retrieve details about a specific order
2. Raise a purchase requisition
3. Retrieve information about issues related to a specific order
4. Retrieve information about future demand for products or services
5. Resolve an order query

Supplier Relationship Management

1. Retrieve information on a supplier's performance
2. Retrieve information about the risk associated with a specific supplier
3. Retrieve information on how to optimize value from a supplier relationship
4. Retrieve details about a specific contract
5. Retrieve information about the current status of a contract in its lifecycle

Sourcing

1. Retrieve information about the current status of a sourcing project
2. Retrieve publicly available information on suppliers or market trends
3. Retrieve information that can assist with the sourcing process
4. Retrieve a list of preferred suppliers for a specific product or service
5. Retrieve information about a supplier's capability to provide a specific product or service

Spend and Budget Tracking

1. Retrieve information about spend with a specific supplier or for a specific product or service
2. Retrieve information about how well procurement spend is adhering to the budget
3. Retrieve information on budget utilization by department or category
4. View historical spend trends by category
5. Retrieve information about planned vs actual spend for a period.

Guided Buying & Purchasing

1. Retrieve information about the approval status for a specific procurement request or action
2. Retrieve a list of existing suppliers for a specific product or service
3. Raise a procurement request
4. Retrieve information about the procurement process for a specific request
5. Track the status of a procurement request or action

Supplier Collaboration & Self Serve

1. Retrieve information on setting up a new supplier in the procurement system
2. Retrieve information about the bidding process for procurement
3. Retrieve information about the status of a specific order or shipment
4. Retrieve information about the invoicing process for procurement
5. Retrieve information on how to follow up on unpaid invoices with suppliers.

Solution Options

Using a Chatbot Platform (e.g., Amazon Lex, Dialogflow)

Description: A chatbot platform like Amazon Lex or Dialogflow offers pre-built solutions that can be customised to your organisation's needs. These platforms provide a balance between flexibility and convenience and include built-in NLU (Natural Language Understanding) and NLG (Natural Language Generation) engines, enhancing the chatbot's language processing capabilities.

Pros:

• Offers a balance between flexibility and convenience
• Can be customised to the organisation's needs
• Development costs are typically lower than building from scratch
• Access to pre-built integrations, features, and advanced NLU/NLG engines

Cons:

• May not offer the same level of customisation as building from scratch
• Limited control over the chatbot's functionality and features
• Potential reliance on the platform's ongoing support and updates

Building a Chatbot from Scratch using Python

Description: Developing a chatbot from scratch using Python involves creating a custom solution tailored to your specific needs. This approach allows for maximum flexibility and customisation but requires significant time, resources, and expertise.

Pros:

• Maximum flexibility and customisation
• Complete control over chatbot's functionality and features
• Can be tailored to the specific needs of the organisation

Cons:

• Requires a significant investment of time, resources, and expertise
• Development costs can be high
• Maintenance and updates require ongoing investment

Pre-Packaged Chatbot Solution

Description: A pre-packaged chatbot solution is an out-of-the-box chatbot that can be quickly deployed with minimal customisation. This approach offers the most convenience but may not provide the flexibility or advanced features available with other options.

Pros:

• Fast and easy deployment
• Low development costs
• Minimal time and resources required

Cons:

• Limited flexibility and customisation
• May not be tailored to the specific needs of the organisation
• May lack advanced features, such as NLU and NLG engines, compared to other options

Leading ChatBot Vendors

Here are some of the leading ChatBot vendors:

• IBM Watson Assistant
• Google Dialogflow
• Microsoft Bot Framework
• Amazon Lex
• Rasa
• Tars
• Botpress
• Kore.ai

When selecting a ChatBot vendor, it's important to consider factors such as the vendor's pricing model, features and functionality, customisation options, and integration capabilities. Additionally, it's important to consider the vendor's track record in delivering ChatBot solutions to organisations similar to your own.

Building a Chatbot: Step-by-Step Approach

The Hudson&Hayes ChatBot Delivery approach provide a seven step process for designing, developing, deploying and maintaining a ChatBot.

Phase 1: Initiation

Activities:

1. Define a compelling vision and story for the chatbot
2. Mobilise a team
3. Create the project charter and plan
4. Create a stakeholder and communication plan

Deliverables:

• Vision statement
• Project charter, plan, and communication plan
• Stakeholder and communication plan

Phase 2: Platform Selection

Activities:

1. Capture platform requirements
2. Conduct vendor selection
3. Select a platform that aligns with the chatbot's purpose and requirements

Deliverables:

• Chatbot platform requirements
• Chatbot platform selection report
• Chatbot platform setup and configuration

Phase 3: Chatbot Environment Setup in a Platform

Activities:

1. Configure the platform settings for development, testing, and production environments
2. Integrate with external APIs or services (if required)
3. Set up monitoring and analytics tools
4. Implement a deployment pipeline for test and production environments

Deliverables:

1. Environment setup
2. Integration with external APIs or services
3. Monitoring and analytics tools
4. Deployment pipeline

Phase 4: Iterative Chatbot Design

Activities:

1. Gather and analyse existing demand data
2. Define high-level user journey
3. Create an intent and entity repository
4. Create conversational flows for prioritised intents
5. Capture API details and service connectors
6. Define fallback and error handling
7. Repeat step 4 during each wave/sprint

Deliverables:

• User Journey
• Intent and entity repository
• Conversational flow diagrams
• API list
• Fallback and error handling mechanisms

Example Deliverables 1: User Journey

Example Deliverables 2: Intent Repository for a Procurement ChatBot

Phase 5: Iterative Development and Testing

Activities:

1. Create a wave and sprint plan and allocate intents to sprint
2. Develop, test, and integrate iteratively
3. Conduct user demo
4. Incorporate feedback

Deliverables:

• Sprint plan
• Test plan
• Chatbot development
• User feedback report

Phase 6: Deployment & Change Management

In this phase, the chatbot is deployed to relevant channels and integrated with the relevant systems and APIs.

1. Develop a training plan for chatbot users and stakeholders
2. Develop a maintenance plan to ensure that the chatbot remains up-to-date and relevant
3. Develop a governance plan to ensure that the chatbot complies with relevant regulations and policies
4. Deploy to relevant channels

Deliverables:

• Deployed chatbot
• Integration and deployment report
• Production checklist
• Release notes

Phase 7: Monitoring and Retraining

In this phase, the chatbot's performance is monitored, and the chatbot is retrained based on feedback to improve its accuracy and effectiveness.

1. Monitor the chatbot
2. Retrain the chatbot

Deliverables:

• Performance and user feedback report
• Retraining plan

Common pitfalls and how to overcome them

When implementing conversational AI in procurement, pitfalls can arise at every step of the journey. Here are a few examples:

Lack of Education on ChatBot Capabilities: When organisations introduce a ChatBot, it's crucial to educate users on its capabilities and how it evolves over time. Otherwise, users may become frustrated in the early stages if the ChatBot doesn't answer all their questions. It's important to set expectations and communicate the ChatBot's limitations clearly.

Not Focusing on User Experience: It's easy to get caught up in the technical aspects of delivering conversational AI and forget about the user experience. When users engage with the ChatBot, they should have a seamless experience that feels natural and intuitive. This requires a focus on user experience design and testing.

Going Complex Too Soon: Organisations should ease into the use of ChatBots and show the art of the possible. Going too complex too soon can overwhelm users and lead to adoption challenges. By starting with simple use cases and gradually adding complexity over time, organisations can ensure that users are comfortable and confident using the ChatBot.

Poor data quality: Chatbots rely heavily on data to provide accurate and relevant responses to user inquiries. If the data used to train the chatbot is of poor quality, the chatbot may provide incorrect or irrelevant responses to users. For example, in a procurement context, if the data used to train the chatbot is based on fragmented systems with inconsistent supplier data or duplicate records, the chatbot may struggle to provide accurate supplier information to users. This can result in frustration for users and a lack of trust in the chatbot's ability to provide reliable information. Therefore, it is essential to ensure that the data used to train the chatbot is accurate, consistent, and up-to-date.

Lack of domain expertise: Chatbots require a deep understanding of the domain they are intended to serve. If the team building the chatbot does not have the necessary expertise, they may struggle to create relevant and useful responses. This can lead to a chatbot that is frustrating for users and fails to achieve its objectives.

Insufficient testing: Chatbots require extensive testing to ensure that they are working as intended. This includes both functional testing (i.e., ensuring that the chatbot responds correctly to user inputs) and performance testing (i.e., ensuring that the chatbot can handle the expected volume of traffic). Without sufficient testing, the chatbot may fail to meet user needs and expectations, leading to poor adoption rates and user dissatisfaction.

Maintenance: Like any other technology, ChatBots require ongoing maintenance to perform optimally. They need to be re-trained periodically with new data to ensure that they continue to understand and respond to user messages accurately. Organisations should have a plan in place for maintaining and updating their ChatBot to ensure that it continues to deliver value over time.

In Summary

Conversational AI is rapidly transforming many industries, and procurement is no exception. Despite the fact that procurement spends a large proportion of time dealing with queries from the business that people could have completed themselves, the use of chatbots and conversational AIs has yet to take off. With the implementation of ChatBots, procurement can benefit from improved user experience, increased productivity, ease of business with suppliers, and increased effectiveness for procurement staff. The use of ChatBots and conversational AIs in procurement is expected to significantly grow over the coming years, providing benefits for procurement, budget holders, and suppliers.