@July 6, 2026 Latest Updates Summary

Budget Estimate Status

• Total hours invested: 51 hours 36 minutes
• Estimated project cap: 56 hours
• Estimated remaining: 4 hours 24 minutes

Most development effort is focused on refining behavior through realistic customer conversations rather than adding new capabilities. Recent testing indicates the assistant is consistently selecting the correct installation guides, maintaining conversation context, and relying on official documentation instead of guessing.

Metra Support Assistant - Live Demo

Try out Metra Support Assistant - an AI chatbot demo for Metra Support Assistant

docsbot.ai

Since the Last Update

The most recent work focused on polishing the customer experience rather than expanding the knowledge base.

Improvements

• The assistant now immediately presents the customer-facing installation guide once the correct fitment has been identified.
• Improved guide selection when customers provide only a year and model. When the documentation yields a single unambiguous match, the assistant now infers the vehicle make automatically (for example, "2018 Silverado" → Chevrolet Silverado).
• Improved long conversation context so the assistant continues working from the active installation guide without repeatedly asking for vehicle or product information.
• Improved context switching when the customer changes vehicles or products mid-conversation.
• Improved handling of abbreviated product names and minor typing mistakes.
• Reduced unnecessary follow-up questions after the correct guide has already been identified.
• Refined escalation behavior so the assistant works through the available documentation before recommending Technical Support.

Human Escalation Clarification

One clarification from development: DocsBot does NOT provide a native, real-time handoff from the AI assistant to a live Technical Support representative, as I had previously understood. That capability would require integration with a third-party support platform .

For the initial rollout, I don't believe this is a significant limitation. A Technical Support representative with access to the DocsBot dashboard (available once we confirm our upgrade to the annual Standard Plan with 5 team users) can view conversation history, customer questions, and previous responses to quickly understand what has already been discussed if a customer calls in.

The dashboard also provides reporting and analytics, including conversation logs, frequently asked questions, customer feedback, and areas where the assistant is succeeding or needs improvement. Those insights should be valuable as we continue refining the assistant over time.

Demo Script

This demo path is meant to show Max acting as an installation assistant while the customer is already on the support site with the guide open. It demonstrates recent updates: returning the guide URL immediately, matching a guide from year/model even when the make is omitted, retaining context, switching vehicle/product context mid-chat, answering slang or abbreviated questions within reason, and escalating instead of guessing when it reaches a limit.

As with any LLM assistant, phrasing and retrieval can vary, and it can still be confused or lose context, but the live support handoff is now part of the safety net.

1. I have a 2013 Honda Accord and I’m installing an AXSWC.
• Expected: Returns axswc-honda-accord-10 guide URL.
• Why: Basic fitment lookup and immediate URL response.
2. Which wires do I connect to the aftermarket radio?
• Expected: Summarizes radio harness wiring.
• Why: Broad guide-section question.
3. Where does the Gray/Red wire connect?
• Expected: Identifies vehicle harness Pin 16.
• Why: Specific wiring lookup.
4. I’m ready to program the AXSWC. How do I get started?
• Expected: Gives first programming steps.
• Why: Procedural guidance from the guide.
5. The LED is flashing green seven times. Is that normal?
• Expected: Explains seven green flashes are expected.
• Why: LED feedback interpretation.
6. I’m installing a Kenwood radio. How should I connect the steering wheel control wire?
• Expected: Gives Kenwood/JVC SWC wire guidance.
• Why: Radio-specific subsection.
7. My steering wheel controls still aren’t working after programming. What should I check first?
• Expected: Offers documented troubleshooting, including Kenwood/JVC checks.
• Why: Troubleshooting without guessing.
8. Actually, I’m working on a 2021 Toyota Camry with an AXSWC-2.
• Expected: Switches context and returns Toyota guide URL.
• Why: Vehicle and product context switch.
9. My AXSWC-2 is flashing Green, then Red. What does that mean?
• Expected: Explains AXSWC-2 did not detect vehicle.
• Why: Product-specific LED behavior.
10. It detected JVC, but I actually have a Kenwood radio.
• Expected: Explains Kenwood/JVC relationship and special troubleshooting.
• Why: Handles nuanced radio detection.
11. How do I remap my steering wheel buttons?
• Expected: Provides remapping guidance/link.
• Why: Related support document retrieval.
12. Remember, I have an AXSCW-2, not the AXSWC. Is the button remapping procedure identical for both?
• Expected: Handles typo, confirms AXSWC-2 context.
• Why: Tests abbreviated/mistyped product handling.
13. My radio still isn’t responding after I’ve followed all of these steps.
• Expected: Gives final documented checks.
• Why: Pushes toward support boundary.
14. It still isn’t working. I think I need more help.
• Expected: Offers Metra Technical Support.
• Why: Proper escalation.
15. I need help with a 2018 Silverado.
• Expected: Infers Chevrolet and returns Silverado guide URL.
• Why: Year/model lookup without make.
16. No, actually, I think it’s a 2037 Silverado.
• Expected: Says no documented guide located and asks to confirm year.
• Why: Does not invent future fitment.
17. No, I’m pretty sure it’s a 2037 Silverado. Could you double check.
• Expected: Still refuses to guess.
• Why: Tests persistence under pressure.
18. Nope.
• Expected: Offers Metra Technical Support.
• Why: Final handoff when documentation cannot resolve.

Current Status

Recent testing has been very encouraging.

• Installation guide selection appears stable across the scenarios tested.
• Guide URLs are now returned immediately after fitment identification.
• Follow-up questions are consistently answered from the active guide instead of restarting the lookup process.
• The assistant successfully transitions between different vehicles and between AXSWC and AXSWC-2 without losing context.
• Technical answers remain grounded in the official documentation, preserving wiring colors, connector references, programming steps, and LED behavior.
• When documentation reaches its limits, the assistant now consistently avoids guessing and recommends Technical Support instead.

Known Issues

• Vehicle fitment continues to be spot-checked across manufacturers to validate the routing data. The fitment library is large enough that continued validation is worthwhile even though overall guide selection has become much more reliable.
• As with any LLM-based assistant, retrieval and phrasing are probabilistic. While the assistant has become much more consistent, occasional variations in wording or retrieval behavior can still occur and remain an area of ongoing testing.
• Some responses still include unnecessary conversational follow-up questions after fully answering the customer's original request. These are largely cosmetic but continue to be refined.

Next Steps

• Continue validating vehicle fitment across the complete application database.
• Continue broader testing using realistic customer conversations and edge-case scenarios.
• Continue refining conversation flow to make interactions feel more natural and less repetitive.
• Further reduce unnecessary Technical Support escalations when official documentation can answer the customer's question.
• Determine whether to expand the assistant beyond AXSWC documentation to additional Metra product categories, and if so, define the appropriate retrieval workflow.

@July 1, 2026 Latest Updates Summary

Budget Estimate Status

• Total hours invested: ~42 hours
• Estimated project cap: 56 hours
• Estimated remaining: ~14 hours

The current estimate includes the time required to build and validate the vehicle fitment data, which has been one of the largest and most important parts of the project.

Based on the progress so far, I believe there’s a high likelihood of having a stable, thoroughly testable version of the assistant by the time the estimated project cap is reached. Once that milestone is reached and we’ve completed a broader round of testing, we should have enough information to evaluate whether additional development would provide meaningful improvements or whether the project can transition into ongoing refinement and maintenance.

Metra Support Assistant - Live Demo

Try out Metra Support Assistant - an AI chatbot demo for Metra Support Assistant

docsbot.ai

General

Over the past week I’ve been focused on making the assistant behave more like an experienced Metra support representative instead of a generic chatbot. Most of the work has been behind the scenes: reorganizing how it retrieves information, separating technical data from behavioral instructions, and improving how it identifies the correct installation guide before helping the customer.

Summary

1. Separated the prompt into behavioral instructions and dedicated technical knowledge sources.
2. Improved how the assistant selects installation guides and maintains conversation context.
3. Expanded technical reference material for troubleshooting, wiring, programming, and LED feedback.
4. Shifted testing toward real customer conversations to identify edge cases and improve overall reliability.

Additions

• Added a complete AXSWC Vehicle Fitment Index to help the assistant identify the correct installation guide.
• Expanded the fitment data to handle for vehicle qualifiers such as:
• Factory navigation
• JBL / Bose audio
• Bluetooth bypass
• Shell numbers
• Other documented vehicle-specific variations
• Updated the dedicated AXSWC Technical Reference Guide containing:
• LED feedback
• Wiring information
• Pin mappings
• Programming procedures
• Shared AXSWC / AXSWC-2 behavior
• Troubleshooting guidance

Improvements

• Simplified and reorganized the Agent prompt to focus on behavior instead of storing technical data.
• Improved how the assistant remembers the active vehicle and product throughout a conversation.
• Improved context switching when a customer changes to a different vehicle.
• Reduced unnecessary documentation lookups during follow-up questions.
• Improved documentation fidelity so wiring colors, pin numbers, LED sequences, and troubleshooting steps stay true to the official documentation.
• Reduced the tendency to immediately recommend Technical Support before working through the available documentation.

Examples

Recent testing showed several noticeable improvements.

• The assistant correctly identified a 2009 Audi A4 as a direct match and immediately returned the correct installation guide without asking unnecessary qualifier questions.
• During the same conversation it answered wiring questions directly from the guide, including:
• Pin 1 (Pink wire) → Vehicle Pin 9
• Pin 7 (Blue/Pink wire) → Vehicle Pin 10
• It also preserved the installation TIP from the documentation.
• It successfully switched from an Audi to a Honda, recognized that multiple Honda guides existed, and asked whether the vehicle had factory navigation before selecting the correct guide.
• During troubleshooting it correctly walked through power, ground, wiring quality, and programming checks instead of immediately directing the customer to Technical Support.
• It maintained the active vehicle context across a long conversation covering guide selection, programming, LED feedback, and troubleshooting.

Known Issues

• A small number of fitment lookups still require validation. One recent example involved the assistant selecting the wrong Mazda guide even though the correct guide exists in the fitment index.
• Because the fitment library is very large, I’m continuing to spot-check vehicles across different manufacturers to verify guide selection accuracy.
• The assistant can still be a little too eager to hand customers off to Technical Support in certain situations. This has improved significantly but continues to be refined.
• Some responses still include unnecessary follow-up questions after successfully answering the customer’s original question.

Next Steps

• Continue validating vehicle fitment across the complete application database.
• Improve retrieval consistency so the assistant always selects the correct installation guide.
• Continue refining conversation flow to make interactions feel more natural and less repetitive.
• Further reduce unnecessary Technical Support escalations when official documentation can answer the customer’s question.
• Continue expanding testing using realistic customer conversations and edge-case scenarios to identify additional refinements before wider deployment.
• Determine whether or not to give the Bot access to non AXSWC documentation (e.g., Full Audio Packages) and, if so, when.

Feedback Results & Issue Tracking

NOTE Removal of the full site source data after the first day of testing produced unintended side effects. Although mitigation steps substantially reduced these issues and successfully constrained most responses to AXSWC/AXSWC-2 documentation, some responses continued to reference information that appeared to originate from previously available source material as the query repository and contextual data was rebuilt.

Data Source Expansion & Refinement

1. Add support for common owner shorthand, chassis/platform codes, internal model names, and other enthusiast terminology (e.g., BMW E46, E53, E39, and equivalent Mercedes, MINI, and Porsche designations) through a synonym/cross-reference layer that maps these terms to the corresponding year, make, and model for document lookup.
2. Expand the bot's terminology mapping to recognize common abbreviations, nicknames, and alternate naming conventions used by customers, rather than relying solely on the terminology present in vehicle fitment data.
3. Continue to collect a glossary of terms the customer may utilize to refer to elements of the documents that differ from the language used. Understand that customers will employ abbreviations, fragments, and poor knowledge to attempt to locate information and assistance.

General Improvement

1. Never provide a best-guess answer when application data cannot be confidently matched to a specific year, make, and model. Clearly state that the required information could not be verified and direct the customer to Technical Support, e.g., if a match cannot be found for a vehicle year make and model for AXSWC (a certain year and model of BMW), clearly state this and direct to Technical Support.
2. Improve session context management so that responses remain constrained to the product and year, make, and model established during the conversation. If confidence in the active context is lost, request clarification before answering.
3. Favor clarification over assumption. When a customer query is ambiguous or incomplete, ask follow-up questions rather than inferring missing information or returning speculative results.
4. Continue to improve procedural context awareness by tracking the customer's current step in a documented process. For example, if the conversation is focused within the Programming the Steering Wheel Controller procedure, prioritize Programming-related LED feedback and instructions rather than matching against unrelated LED feedback sections elsewhere in the documentation.
5. Introduce confidence-based escalation behavior. When the bot cannot confidently identify the correct application or documentation section, gracefully stop and recommend Technical Support rather than continuing with uncertain guidance.
6. Where multiple valid interpretations exist, explicitly present the ambiguity to the customer and request the information needed to disambiguate before proceeding.

Full Implementation Updates

1. Add data for all supported vehicle fit applications as was done for BMW. When doing so, attempt to incorporate #1 from Data Source Expansion & Refinement when needed. Maintain ability for the bot to clearly state that the application is not supported and direct to live Technical Support if no document can be matched to provided fitment data.
2. When full site access is given, ensure that context is not lost during a session with regards to the product being discussed, e.g., do not allow cross contamination of data from Full Audio Package Products to AXSWC/2 products.

Known Limitations

The following are fundamental limitations of a knowledge base chatbot and should be considered during evaluation and future development.

1. The bot can only answer questions using the information available in its source material. It cannot provide authoritative answers for information that is not documented and it would not be advisable to ever allow it to interpret or seek external sources for this purpose.
2. The bot cannot reliably infer missing information. If critical details such as the year, make, model, or product are omitted or ambiguous, clarification is required.
3. The bot only understands terminology it has been taught or has encountered in its source data. Customer shorthand, slang, abbreviations, and enthusiast terminology require explicit mapping or supporting documentation.
4. The quality of the bot's responses is directly dependent on the quality, accuracy, and completeness of the underlying documentation. Incomplete or conflicting source material will lead to incomplete or conflicting responses.

@06/08/2026 Latest Updates Summary

Metra Support Assistant - Live Demo

Try out Metra Support Assistant - an AI chatbot demo for Metra Support Assistant

docsbot.ai

Proposed Minimally Viable Product Demo

After several rounds of testing and refinement, the AI assistant should now be operating as a documentation-driven tool capable of handling a broad range of common customer inquiries.

• The assistant can identify and retain vehicle and product context, retrieve relevant documentation, prioritize vehicle-specific guides, interpret natural-language questions, ask clarifying questions when information is missing, guide users through documented troubleshooting workflows, reuse previously retrieved information, and escalate appropriately when documentation does not provide a clear answer.
• Recent improvements have focused on troubleshooting behavior, including better handling of ambiguous LED feedback, improved distinction between AXSWC and AXSWC-2 responses, stronger context retention, more targeted follow-up questions, improved identification of likely failure points during programming and setup, and better recognition of when a user is describing a known failure condition rather than a successful programming result.

Recent Testing

The latest testing cycle was particularly useful because it focused on realistic customer behavior rather than ideal inputs.

• Customers rarely describe issues the way documentation is written. They leave out details, use inconsistent terminology, combine multiple symptoms into a single description, and often don't know exactly what step failed.
• Recent updates have improved the assistant's ability to work through those situations by asking more focused diagnostic questions, identifying where a process appears to have succeeded, isolating likely failure points, and adjusting its responses when additional information is provided.
• Rather than immediately committing to an answer, the assistant is now more likely to clarify ambiguity before moving forward.

What It Does Well

• Retrieves and applies relevant documentation
• Maintains vehicle and product context throughout a conversation
• Handles many natural-language support questions
• Guides users through documented procedures
• Identifies common troubleshooting paths
• Requests clarification when information is incomplete
• Adapts when users provide corrective information
• Escalates when documentation no longer supports a confident answer

Remaining Limitations

The remaining issues are largely edge cases:

• Ambiguous customer descriptions
• Documentation gaps
• Unusual installation scenarios
• Situations where multiple documented behaviors appear similar

These cases will continue to surface through testing and can be addressed in stages. Additional refinement is certainly possible, but each round of improvements is likely to produce smaller gains at this stage.

Moving Forward

At this point, I would suggest that the proof of concept has demonstrated the core capabilities. The assistant can retrieve documentation, maintain context, guide troubleshooting, ask clarifying questions, adapt to user corrections, and escalate appropriately when documentation does not support a confident answer.

Before spending additional time refining edge cases, I think it would be helpful to discuss a few questions:

1. Is the assistant currently capable of handling a meaningful percentage of common customer inquiries?
1. If not, what categories of inquiries are still missing?
2. Is the assistant accurately using Metra documentation as the basis for its responses?
1. If not, are there examples of recurring problems rather than isolated mistakes?
3. Is the assistant asking reasonable follow-up questions when customer information is incomplete or ambiguous?
1. If not, what behavior would you expect instead?
4. Is the assistant appropriately identifying when it should continue troubleshooting versus when it should direct a customer to Technical Support?
1. If not, where should that boundary be adjusted?
5. Should success be measured by broad performance across common support inquiries, or by how it handles individual edge cases?
1. Edge cases will always exist and can continue to be refined. The more important question is whether the assistant is providing useful, accurate responses across the majority of routine customer interactions.
6. Is the current BMW-focused implementation sufficient to demonstrate viability, or is there a specific reason broader application coverage is required before making that determination?

The main question at this point is whether the quality of the responses is sufficient to justify broader implementation across additional applications and product lines. What I'm trying to determine is whether we've reached the point where the concept itself has been proven and the next step is expansion and refinement, or whether there is a specific capability that is still missing from the proof of concept.

@06/01/2026 Latest Updates Summary

Metra Support Assistant - Live Demo

Try out Metra Support Assistant - an AI chatbot demo for Metra Support Assistant

docsbot.ai

Testing Notes/Context

When evaluating the assistant, please keep in mind that the current demo environment is separate from the final deployment experience.

• In production, the assistant will be embedded directly within the Metra documentation environment. Customers will be able to interact with the assistant while simultaneously viewing, searching, scanning, and navigating the knowledge base itself.
• The goal is not to replace documentation, but to help customers locate the most relevant documentation more quickly and guide them through it more efficiently.
• When testing, it may be helpful to think of the assistant as an intelligent support layer sitting on top of the existing knowledge base rather than as a standalone chatbot.

A Note on Feedback

Specific examples are the fastest way to improve the assistant. When a response seems incorrect, please try to answer the following questions:

1. What question was asked? Include the exact question whenever possible.
2. What response was given? Include the assistant's answer or a screenshot.
3. What was wrong with the response? Examples:
1. Wrong document selected
2. Missing information
3. Incorrect information
4. Too much information
5. Should have asked for vehicle information first
6. Should have used a different guide
4. What should have happened instead? Describe the response that would have been expected from a member of the technical support team.
5. Where does that information come from? If known, identify whether the information:
1. Exists in a guide or knowledge base article
2. Exists somewhere else in the documentation
3. Is common technical support knowledge that may not currently be documented This last point is particularly valuable because it helps determine whether the solution is:
1. improving the assistant
2. improving the documentation,
3. or adding additional support-specific guidance.

The more specific the example, the easier it is to identify exactly where the process broke down and make targeted improvements.

Conversation & Documentation Improvements

Additional refinements have been made to improve how the assistant handles ongoing support conversations and documentation retrieval.

Better Follow-Up Support

1. Previously, follow-up questions could sometimes trigger additional documentation searches even when the relevant guide had already been identified earlier in the conversation.
2. The assistant now retains relevant documentation context throughout a support session and will continue using previously identified documentation when it remains applicable to the customer's question.

This creates a more natural troubleshooting experience and reduces unnecessary repetition.

What This Improves

Expected improvements include:

• Faster responses during multi-step troubleshooting
• Better continuity between follow-up questions
• Less repetitive troubleshooting guidance
• Fewer unnecessary documentation lookups
• More consistent answers throughout a conversation
• Improved handling of longer support interactions
• Better retention of vehicle and product context For example, after identifying the appropriate installation guide, the assistant can continue answering related questions about programming, LED behavior, button mapping, radio detection, wiring, or troubleshooting without repeatedly re-locating the same documentation.

Applies Across the Knowledge Base

These improvements are not limited to AXSWC and AXSWC-2 documentation. The same behavior applies across customer-facing documentation, including:

• Vehicle-specific installation guides
• Full audio package guides
• Product manuals
• Manual programming guides
• Troubleshooting documents
• FAQ content
• Product-specific reference documentation
• Other customer-facing support resources

Vehicle Context Remains Important

The assistant will continue attempting to identify the customer's:

• Year, Make, Model
• When that information is relevant to the support request. This helps prioritize vehicle-specific installation and troubleshooting documentation when available. However, vehicle information is no longer treated as a requirement for every documentation request. If a customer is asking about information contained in a product-specific document, such as:
• Manual programming procedures
• LED feedback behavior
• General product operation
• Product features
• FAQ content
• The assistant can still provide the appropriate documentation without requiring a vehicle-specific guide. This should improve access to product-specific information while still prioritizing vehicle-specific documentation when vehicle-dependent information is needed.

More Technician-Like Behavior

The assistant is being refined to behave more like an experienced technical support representative. Once the appropriate documentation has been identified, the conversation should continue from that documentation until:

• a different issue is introduced
• a different product is introduced
• a different vehicle is introduced
• or additional documentation becomes necessary.
• The result should be a smoother support experience that feels less like repeated searches and more like an ongoing troubleshooting conversation.

Expected Results

Combined with the previous documentation and troubleshooting updates, the assistant should now:

• Better maintain conversation context
• Stay focused on the customer's specific issue
• Improve consistency across follow-up questions
• Reduce unnecessary troubleshooting expansion
• Improve retrieval accuracy across multiple document types
• Surface relevant documentation more efficiently
• Deliver a more natural support experience

Next Areas of Focus

The current system is producing increasingly consistent results, particularly for documentation retrieval, installation guidance, and AXSWC troubleshooting workflows.

The next phase of improvement is less about finding documents and more about refining how the assistant applies support knowledge and handles real-world customer interactions.

1. Expand Vehicle Coverage

Current testing has focused heavily on BMW applications.

The next step is expanding the same level of validation and optimization across additional makes, models, and product categories.

This includes strengthening relationships between:

• Vehicle applications
• Product families
• Installation guides
• Troubleshooting documents
• Programming guides

The goal is to ensure customers are consistently routed to the most relevant documentation regardless of vehicle platform.

2. Continue Identifying Common Edge Cases

Testing has already uncovered situations where a response is technically reasonable but does not reflect how the technical support team would answer the question.

These situations are often valuable because they can be converted into support-specific rules.

Examples may include:

• Vehicle-specific limitations
• Product-specific exceptions
• Frequently misunderstood procedures
• Common troubleshooting shortcuts used by the support team

Over time, these edge cases can be documented and taught to the assistant so that responses more closely reflect real-world support experience.

3. Build a Support Knowledge Layer

Some answers come directly from documentation.

Others come from years of installer and technical support experience.

An opportunity exists to identify support knowledge that may not currently exist in formal documentation and determine whether it should be:

• Added to documentation
• Added as assistant guidance
• Added as troubleshooting rules

This helps bridge the gap between documented procedures and technician experience.

4. Define Escalation Guidelines

As the assistant becomes more capable, it may be useful to further define when a conversation should transition to technical support.

Examples may include:

• Repeated troubleshooting failures
• Missing documentation
• Vehicle-specific edge cases
• Situations requiring direct diagnostic assistance

This can be a structured process or a more flexible guideline depending on how technical support prefers the assistant to operate.

5. Evaluate Live Support Handoff Options

If desired, the system can eventually support workflows where technical support personnel monitor or take over active conversations when needed.

Potential benefits include:

• Faster resolution of difficult cases
• Real-time customer assistance
• Additional training data for future improvements
• Reduced customer frustration during complex troubleshooting

This would require additional implementation and workflow planning but remains a viable future enhancement.

Current Assessment

The largest opportunities for improvement no longer appear to be documentation retrieval itself.

Most future gains are likely to come from:

• Expanding coverage
• Capturing support expertise
• Refining edge-case handling
• And continuing to align responses with how the technical support team would assist customers in real-world situations.

END OF UPDATE

@05/22/2026 Latest Updates Summary

Metra Support Assistant - Live Demo

Try out Metra Support Assistant - an AI chatbot demo for Metra Support Assistant

docsbot.ai

AXSWC / AXSWC-2 Improvements

Added a dedicated FAQ response layer for common AXSWC and AXSWC-2 questions. This helps the assistant:

• respond more consistently
• avoid overly long troubleshooting responses
• reduce conflicting instructions
• and give faster answers for common problems

The assistant now gives short, direct answers first before expanding into deeper troubleshooting. Also updated the AXSWC and AXSWC-2 documentation structure itself (the actual documents).

Previously, sections in the documents such as:

• LED Feedback
• Changing Radio Type
• Button Remapping

used the same names for both AXSWC and AXSWC-2 specific subsections. This made it more difficult for the assistant to tell which model-specific instructions applied. The sections now use model-specific names such as:

• AXSWC LED Feedback
• AXSWC-2 LED Feedback
• AXSWC Changing Radio Type
• AXSWC-2 Changing Radio Type

This should improve:

• product specific retrieval accuracy
• troubleshooting consistency
• and model-specific guidance

Troubleshooting Flow

The assistant now tries to identify where the installation process stopped working instead of jumping immediately into broad troubleshooting.

For AXSWC products, it follows the installation flow:

1. Connections
2. SWC jack wiring
3. Programming
4. LED feedback and troubleshooting

If needed, it may ask something like:

What was the last successful step in the installation process?

This should help narrow the issue down faster and avoids unnecessary troubleshooting steps.

Current Limitations

The assistant is still improving. Occasionally it may:

• provide too much troubleshooting information
• combine information from related documents
• or expand further than necessary

Recent changes have already reduced this significantly.

Demo Feedback Guidelines

If the assistant gives an incorrect or confusing response, please include:

• the exact user question
• the incorrect response
• and what the response should have been instead

Simply saying the response was “wrong” usually is not enough information to improve the system. The more specific the example, the easier it is to refine the assistant’s behavior by determining where in the response the logic broke.

END OF UPDATE

@05/11/2026 Latest Updates Summary

As part of this update, the assistant has also been formally named MARK:

Metra Advanced Repository of Knowledge.

Metra Support Assistant - Live Demo

Try out Metra Support Assistant - an AI chatbot demo for Metra Support Assistant

docsbot.ai

The purpose of this update was to improve documentation retrieval, reduce unsupported troubleshooting responses, and create a more structured customer support experience.

Expanded Knowledge Base Access

MARK now has access to the broader Metra knowledge base ecosystem, including:

• Vehicle-specific installation guides
• Product manuals
• AXSWC and AXSWC-2 programming documentation
• Wiring and installation references
• Troubleshooting articles
• Technical support documentation
• Programming and LED feedback references

Full Audio Packages

MARK now also has access to and can answer basic questions related to full audio packages follow the same, or similar, steps to the progressive support workflow below.

This allows the assistant to respond using official documentation instead of relying on generalized chatbot behavior.

Progressive Support Workflow

MARK attempts to follow a staged support model designed to guide customers progressively through troubleshooting instead of overwhelming them with large responses immediately.

Step 1: Direct Documentation Guidance

When a relevant guide is identified, MARK first directs the customer to the exact document associated with their product or vehicle application. Rather than immediately summarizing the guide, the assistant prioritizes getting the customer to the correct official documentation first.

Step 2: Section-Level Guidance

If the customer needs additional help, MARK can identify and direct them to the exact section of the guide containing the relevant instructions or information.

Step 3: Clarification

If confusion remains, MARK summarizes the relevant information in simpler language while staying grounded in the documentation.

Step 4: Expanded Troubleshooting Assistance

If needed, MARK can progressively expand on the explanation using additional details pulled directly from the guide and related support references.

Step 5: Escalation Path

If the issue cannot reasonably be resolved through documentation alone, MARK can offer to direct the customer to technical support.

At this time, the escalation layer is still being finalized. Initially, this may consist of directing customers to Metra’s support contact page or phone support. Longer term, live technical support escalation or direct technician chat integration may also become available.

AXSWC-2 LED Feedback Enhancements

A major addition in this update is dedicated support for interpreting AXSWC-2 Radio LED Feedback sequences.

Previously, much of this information existed primarily in visual chart form, making it difficult for the AI to reliably interpret.

MARK can now:

• identify radio types from LED flash patterns
• explain expected LED sequences for specific radio brands
• provide keynotes (if applicable)
• differentiate between radio variants such as Alpine and Alpine Type 2
• provide associated keynotes and programming notes
• direct users to the exact guide section where this information appears
• and ask follow-up questions when clarification is needed

This information has been converted into a machine-readable format specifically optimized for documentation retrieval and conversational troubleshooting.

Conversational Support Flow

MARK now behaves more like a guided technical support assistant than a traditional search bot.

For example:

• it asks whether the provided answer resolved the issue
• retains conversational context across multiple questions
• progressively increases detail only when necessary
• attempts to keep customers grounded in official documentation throughout the interaction

Current Limitation

At this time, automatic application-specific guide matching is primarily limited to BMW applications while broader vehicle/application mapping continues to be expanded.

Additional application coverage and document relationships will be added incrementally.

Overall Result

The updated MARK assistant is now:

• more documentation-aware
• more conversational
• more accurate
• more structured
• significantly better at guiding users through technical support documentation step-by-step

The focus of this update has been improving reliability, reducing unsupported responses, and creating a support experience that more closely mirrors how a human technical support representative would progressively assist a customer.

Example Customer Interaction

Below is an example of how MARK now guides customers progressively through documentation and troubleshooting.

Welcome Experience

When customers open the assistant, they are greeted with:

Hi! I'm Mark. I can help you find installation guides, troubleshooting steps, and Metra product info. I can be particularly helpful with the AXSWC and AXSWC-2 steering wheel controllers. How can I help today?

Step 1: Documentation First

Customer

“I’m installing a radio in my 2015 BMW 328i and I’m having trouble programming the steering wheel controls.”

MARK

This guide should be able to assist you:

https://support.metraonline.com/axswc/axswc-vehicle-specific/axswc-bmw-103

MARK intentionally begins by directing the customer to the correct official documentation rather than immediately attempting troubleshooting.

MARK then follows up conversationally:

Did that help you with your steering wheel controls?

Step 2: Section-Level Guidance

Customer

“Where do I find the steering wheel control programming instructions?”

MARK

This guide contains the steering wheel control programming instructions for your 2015 BMW 328i:

https://support.metraonline.com/axswc/axswc-vehicle-specific/axswc-bmw-103

That page includes the vehicle installation instructions, connections, a downloadable PDF, and the Manual Programming section.

MARK now begins helping the customer navigate the document itself.

Step 3: Clarification

Customer

“I’m still not sure what LED sequence I should be seeing.”

MARK

This guide shows the LED feedback sequence for programming the steering wheel controller in your 2015 BMW 328i:

https://support.metraonline.com/axswc/axswc-vehicle-specific/axswc-bmw-103

Use the “Manual Programming” / “LED Feedback Sequence” section of that page. The guide explains the rapid red/green flashes, the seven green flashes, and the final solid red indicating success.

At this stage, MARK begins summarizing the guide contents while still keeping the customer anchored to the documentation.

Step 4: Expanded Troubleshooting Assistance

Customer

“I have an AXSWC-2, and I'm seeing Green, Red, Green for Radio LED Feedback.”

MARK

Green, Red, Green corresponds to an Alpine radio.

Keynote: If the LED pattern shows Alpine but an Alpine radio isn’t installed, make sure the AXSWC-2’s 3.5 mm jack is plugged into the radio.

The assistant can now interpret LED feedback sequences directly and provide associated troubleshooting guidance. If the customer continues:

Customer

“It’s a JVC radio. What LED pattern should I be seeing?”

MARK

For a JVC radio the AXSWC-2 LED sequence is:

Red

Keynote 1: If the LED pattern shows JVC, change the radio type to Kenwood.

Refer to the “Manual Programming” / “Changing Radio Type” instructions in the vehicle guide if you need to switch types.

The conversation continues progressively, with MARK retaining context from earlier troubleshooting steps.

Step 5: Escalation Path

If the customer still cannot resolve the issue:

Customer

“That didn’t help. I may need to speak with someone.”

MARK

I can connect you with Metra Electronics technical support. Would you like me to do that? Customers are then presented with: Yes, please No thanks

If the customer selects Yes, please, a new browser tab opens to:

Metra Electronics | Contact

Reach the right Metra team for sales, technical support, and international inquiries.

www.metraonline.com

At this time, the escalation flow currently directs users to Metra’s contact/support resources. Future versions may also support live technical support escalation directly through the assistant.

END OF UPDATE

@05/04/2026 Initial Implementation

Summary

This demo is intended to test whether the chatbot can correctly guide a user from a vehicle-specific question to the appropriate AXSWC knowledge base document, then answer follow-up installation questions using the correct document context.

The test path begins with a 1999 BMW 318i, which should route to the BMW 101 document. Midway through the conversation, the user corrects the vehicle to a 2009 BMW 125i, which should cause the bot to switch to the BMW 103 document.

Anticipated responses may vary in exact language. The response does not need to match the wording below exactly, but it should match the basic outcome and document-routing behavior described.

Demo URL

Metra Support Assistant - Live Demo

Try out Metra Support Assistant - an AI chatbot demo for Metra Support Assistant

docsbot.ai

Initial User Query

Can you help me find information about my 1999 BMW 318i and how to install an AXSWC in it?

Anticipated Response

• Direct link to the BMW 101 knowledge base document
• Summary of what the page contains
• Offer for the user to ask additional questions

Second User Query

Can you summarize the wire splicing instructions for my vehicle?

Anticipated Response

• Direct link to the BMW 101 knowledge base document
• Summary of splicing steps matching those in the document
• Response should be in bullet point format

Third User Query

Are there differences in installation between the AXSWC and AXSWC-2 for my vehicle?

Anticipated Response

• Direct link to the BMW 101 knowledge base document
• Statement that no specific at-a-glance differences are noted
• Offer to dig deeper into features, wiring, or installation differences

Fourth User Query

I am confused about the differences in LED feedback. I am using the AXSWC-2. What is the LED feedback specific to that for my vehicle?

Anticipated Response

• Direct link to the BMW 101 knowledge base document
• Summary of the red/green LED flashes for LED feedback specific to AXSWC-2
• Mention of how this differs from the 24 red flashes for AXSWC

Fifth User Query

Oh, I just realized that my vehicle is actually a 2009 BMW 125i. Can you give me a brief summary of the same information, but for that vehicle?

Anticipated Response

• Immediate recognition that the proper document is BMW 103
• Direct link to the BMW 103 knowledge base document
• Summary of information related to the previous queries, but specific to the new vehicle application

Sixth User Query

For my 2009 BMW 125i, how to do I connect my vehicle to the factory harness? Can you tell me what pins and wire colors I use?

Anticipated Response

• Direct link to the BMW 103 knowledge base document
• Complete summary information for:
• Aftermarket radio harness
• Vehicle harness connections
• Splicing
• Note: Further training may be needed to help the bot clearly distinguish between these sections

Seventh User Query

What are the manual programming steps for my vehicle with the AXSWC-2?

Anticipated Response

• Direct link to the BMW 103 knowledge base document
• Complete summary of Manual Programming specific to AXSWC-2
• Note: Test more thoroughly to confirm the bot differentiates between AXSWC and AXSWC-2

Continued Testing Behavior

Based on the behavior demonstrated above, the test user is now free to deviate from the scripted path and begin asking follow-up questions that further explore the currently referenced vehicle (2009 BMW 125i).

The chatbot should continue to:

• Maintain context of the current vehicle
• Provide responses and summaries specific to that vehicle

Only a clear and specific question about a different vehicle should cause the chatbot to shift context and return results for that new vehicle.

Training Scope Note: At this time, only BMW vehicle applications are specifically trained for vehicle-to-guide matching.

• General questions about AXSWC installation, features, or behavior can still be asked and answered.
• However, the chatbot will not be able to direct the user to a specific guide for non-BMW year/make/model combinations.

END OF UPDATE

On This Page

• @July 6, 2026 Latest Updates Summary
• Budget Estimate Status
• Since the Last Update
• Improvements
• Human Escalation Clarification
• Demo Script
• Current Status
• Known Issues
• Next Steps
• @July 1, 2026 Latest Updates Summary
• Budget Estimate Status
• General
• Summary
• Additions
• Improvements
• Examples
• Known Issues
• Next Steps
• Feedback Results & Issue Tracking
• Data Source Expansion & Refinement
• General Improvement
• Full Implementation Updates
• Known Limitations
• @06/08/2026 Latest Updates Summary
• Proposed Minimally Viable Product Demo
• Recent Testing
• What It Does Well
• Remaining Limitations
• Moving Forward
• @06/01/2026 Latest Updates Summary
• Testing Notes/Context
• A Note on Feedback
• Conversation & Documentation Improvements
• Better Follow-Up Support
• What This Improves
• Applies Across the Knowledge Base
• Vehicle Context Remains Important
• More Technician-Like Behavior
• Expected Results
• Next Areas of Focus
• 1. Expand Vehicle Coverage
• 2. Continue Identifying Common Edge Cases
• 3. Build a Support Knowledge Layer
• 4. Define Escalation Guidelines
• 5. Evaluate Live Support Handoff Options
• Current Assessment
• END OF UPDATE
• @05/22/2026 Latest Updates Summary
• AXSWC / AXSWC-2 Improvements
• Troubleshooting Flow
• Current Limitations
• Demo Feedback Guidelines
• END OF UPDATE
• @05/11/2026 Latest Updates Summary
• Expanded Knowledge Base Access
• Full Audio Packages
• Progressive Support Workflow
• Step 1: Direct Documentation Guidance
• Step 2: Section-Level Guidance
• Step 3: Clarification
• Step 4: Expanded Troubleshooting Assistance
• Step 5: Escalation Path
• AXSWC-2 LED Feedback Enhancements
• Conversational Support Flow
• Current Limitation
• Overall Result
• Example Customer Interaction
• Welcome Experience
• Step 1: Documentation First
• Customer
• MARK
• Step 2: Section-Level Guidance
• Customer
• MARK
• Step 3: Clarification
• Customer
• MARK
• Step 4: Expanded Troubleshooting Assistance
• Customer
• MARK
• Customer
• MARK
• Step 5: Escalation Path
• Customer
• MARK
• END OF UPDATE
• @05/04/2026 Initial Implementation
• Summary
• Demo URL
• Initial User Query
• Anticipated Response
• Second User Query
• Anticipated Response
• Third User Query
• Anticipated Response
• Fourth User Query
• Anticipated Response
• Fifth User Query
• Anticipated Response
• Sixth User Query
• Anticipated Response
• Seventh User Query
• Anticipated Response
• Continued Testing Behavior
• END OF UPDATE