Best Practices for Creating Intelligent Intake Routes
Intelligent Intake is a landing page for procurement requests through an intuitive interface. This single starting point for requesters provides smart assistance, minimizes data input, and simplifies the request process, showcasing intelligent Spend AI capabilities. This page will help you craft efficient and accurate prompts for intelligently routing customers to the correct path for your business needs
Follow the configuration in the Set Up Intelligent Intake section of the Administrator Guide to enable Intelligent Intake for procurement.
- Note: You must have Modify access to the Set Up: Intelligent Intake domain to set up the routing rules.
Optimize Rule Specificity and Avoid Overlap
- Avoid Overlapping Items: Don't use the exact same Purchase Items across multiple routes.
- Resolve Ambiguity: Be specific enough to resolve ambiguity, especially for terms that have different meanings across various departments.
- Example: Define one route with Category: IT Hardware + Item: Monitor (Computer), and a separate route with Category: Medical Supplies + Item: Monitor (Cardiac), since Monitor is ambiguous.
Leverage Text Instruction Routes for Guided Intake
Use the Text Instruction destination type as an alternative to direct form routing, especially for complex or multi-path requests.
- Provide Canned Context: This route allows you to embed precise, controlled information and external links (URLs and mailto) directly into the Intelligent Intake Assistant's response. This eliminates the risk and limitations of the Large Language Model (LLM) suggesting an incorrect or hallucinated path.
- Address Multi-Route Needs: This is an excellent solution for scenarios where the user's initial query could lead to several different forms or processes. By listing multiple links (for example, Software License Request, Training Request, Consulting Services Form) in the text instruction, you overcome the Assistant's typical limitation of suggesting only a single route.
Write Compelling and Specific Route Descriptions
The route description is the primary context the AI uses for decision-making. Think of the AI as a human employee; explain the route in natural language just as you would to a person—the more context and information given, the better the decision-making will be.
Avoid Broad Descriptions: Broad, generic route descriptions leave too much room for misunderstanding and conflicting route suggestions.
Be Specific: The more details a route description has, the more opportunity the AI has to make the right decision.
Example: A generic description is Computer Hardware. A specific description is: Requests for computer hardware, including desktops, laptops, and all peripheral devices like monitors, keyboards, and mice.
- Use Exceptions: Use exceptions to help further define routes that deviate from broader rules (for example, all furniture requests except chairs).
- We have included sample starting prompts in the files section of this page.
Example:
- Poor Route Description
- Requests for services or one-time purchases.
- Great Route Description
- Requests for services or one-time purchases, including: Consulting services, Legal services, Temporary staffing, Advertising and marketing services, Maintenance and repair services, Online subscriptions and data sets, Custom software development, Event planning and execution, Specialized training, Capital equipment.
Use Multifaceted Criteria for Robust Routing
Build rules using a combination of criteria for maximum precision.
- Keywords and Text Analysis: Analyze the text in the user’s Business Justification or What do you need? field. Use synonyms and root words to group related requests.
- Example: Keywords like PC, Computer, and Laptop should all be routed to the same Hardware Request Form.
Implement Testing and Refinement
Ensure your routes work as expected in a real-world setting, as user behavior is often unpredictable.
- Test and Refine Iteratively: After the initial configuration, test the routes in the Testing Playground with real use cases. A rule may appear logically sound in configuration but fail to capture the nuances of a user’s natural language or terminology.
- Monitor these results and update the route description with more details when the response is incorrect.
Set Up a Fallback Response
If you follow the guidelines above you should have good success with directing your end users to the correct path for your organization. However, sometimes the user may not get a good result or be able to find what they are looking for. Set up a Fallback Response that enables your end users to take the next step.
You can use the Create Email Link button (envelope) to create a link that will automatically start crafting an email to your procurement department's email address.
- Example: Can't find what you are looking for? Email procurement@yourorg.com for additional purchasing assistance.
Intelligent Intake Sample Starting Prompts
- Filename
- Intelligent Intake Sample Starting Prompts.xlsx
- Size
- 13 KB
- Format
- application/vnd.openxmlformats-officedocument.spreadsheetml.sheet