Aivot: AI-Powered Customer Support for Ecommerce Teams
Aivot: AI-Powered Customer Support for Ecommerce Teams
Aivot: AI-Powered Customer Support for Ecommerce Teams
Overview
Overview
Overview
Aivot is a B2B AI-powered customer support platform, set up as a CRM built for ecommerce support teams. It gives frontline reps and support managers real visibility into how AI actually works, not just what it says.
Every AI suggestion in Aivot comes with a confidence score and a cited source, so reps can see exactly where an answer came from before they send it to a customer. When something's off, reps can flag it directly, turning mistakes into a visible, ongoing correction loop instead of a black box nobody trusts. Managers get the same clarity from their side, with insight into how often AI suggestions are trusted, overridden, or flagged, helping them know where to step in and coach.
The result is a support tool that treats AI as something reps can verify and rely on, not something they have to take on faith.
Aivot is a B2B AI-powered customer support platform, set up as a CRM built for ecommerce support teams. It gives frontline reps and support managers real visibility into how AI actually works, not just what it says.
Every AI suggestion in Aivot comes with a confidence score and a cited source, so reps can see exactly where an answer came from before they send it to a customer. When something's off, reps can flag it directly, turning mistakes into a visible, ongoing correction loop instead of a black box nobody trusts. Managers get the same clarity from their side, with insight into how often AI suggestions are trusted, overridden, or flagged, helping them know where to step in and coach.
The result is a support tool that treats AI as something reps can verify and rely on, not something they have to take on faith.
Aivot is a B2B AI-powered customer support platform, set up as a CRM built for ecommerce support teams. It gives frontline reps and support managers real visibility into how AI actually works, not just what it says.
Every AI suggestion in Aivot comes with a confidence score and a cited source, so reps can see exactly where an answer came from before they send it to a customer. When something's off, reps can flag it directly, turning mistakes into a visible, ongoing correction loop instead of a black box nobody trusts. Managers get the same clarity from their side, with insight into how often AI suggestions are trusted, overridden, or flagged, helping them know where to step in and coach.
The result is a support tool that treats AI as something reps can verify and rely on, not something they have to take on faith.
The Scenario
The Scenario
The Scenario
🔁
🔁
Ecommerce support reps keep getting repetitive issue loops, things like order status, returns, and shipping questions, causing workflows that are too long and inefficient to complete.
Ecommerce support reps keep getting repetitive issue loops, things like order status, returns, and shipping questions, causing workflows that are too long and inefficient to complete.
Ecommerce support reps keep getting repetitive issue loops, things like order status, returns, and shipping questions, causing workflows that are too long and inefficient to complete.
My Hypothesis
My Hypothesis
My Hypothesis
💡
💡
Ecommerce SaaS support reps waste time on repetitive troubleshooting loops, so an AI agent that suggests resolutions based on ticket and order history will reduce manual work and speed up case closure.
Ecommerce SaaS support reps waste time on repetitive troubleshooting loops, so an AI agent that suggests resolutions based on ticket and order history will reduce manual work and speed up case closure.
Ecommerce SaaS support reps waste time on repetitive troubleshooting loops, so an AI agent that suggests resolutions based on ticket and order history will reduce manual work and speed up case closure.
Meet the Rep
Meet the Rep
Meet the Rep
To validate my hypothesis and understand the core user on a deeper level, some research included: competitive analysis, user flow mapping, and pain point identification.
To validate my hypothesis and understand the core user on a deeper level, some research included: competitive analysis, user flow mapping, and pain point identification.
To validate my hypothesis and understand the core user on a deeper level, some research included: competitive analysis, user flow mapping, and pain point identification.
Competitive analysis
Competitive analysis
Competitive analysis
Looking at what already exists across three major players in customer support software, what's going well, where they fall short, and the resulting user pain points, helped me understand the current market. That made it possible to find where Aivot could actually add value, instead of rebuilding a feature set that already exists.
Looking at what already exists across three major players in customer support software, what's going well, where they fall short, and the resulting user pain points, helped me understand the current market. That made it possible to find where Aivot could actually add value, instead of rebuilding a feature set that already exists.
Looking at what already exists across three major players in customer support software, what's going well, where they fall short, and the resulting user pain points, helped me understand the current market. That made it possible to find where Aivot could actually add value, instead of rebuilding a feature set that already exists.
User Flow
User Flow
User Flow
Mapping the flow made it possible to catch where the rep could get stuck or drop off before any screens were designed, so those risks could be solved for upfront rather than patched later. The user flow for the rep incorporates:
Drop-off risks paired with corresponding solutions
A color legend distinguishing flow states
Clear end states
Screen-type node labels
Mapping the flow made it possible to catch where the rep could get stuck or drop off before any screens were designed, so those risks could be solved for upfront rather than patched later. The user flow for the rep incorporates:
Drop-off risks paired with corresponding solutions
A color legend distinguishing flow states
Clear end states
Screen-type node labels
Mapping the flow made it possible to catch where the rep could get stuck or drop off before any screens were designed, so those risks could be solved for upfront rather than patched later. The user flow for the rep incorporates:
Drop-off risks paired with corresponding solutions
A color legend distinguishing flow states
Clear end states
Screen-type node labels
Lo-fidelity Directions
Lo-fidelity Directions
Lo-fidelity Directions
This low-fidelity wireframe was used as the foundation for the layout and content priority before any visual design, keeping the focus on structure and information hierarchy rather than polish.
This low-fidelity wireframe was used as the foundation for the layout and content priority before any visual design, keeping the focus on structure and information hierarchy rather than polish.
This low-fidelity wireframe was used as the foundation for the layout and content priority before any visual design, keeping the focus on structure and information hierarchy rather than polish.
The AI in Action
The AI in Action
The AI in Action
This is where the hypothesis gets tested: does the AI actually cut down the repetitive manual work reps were stuck doing before?
A rep opens a high-priority ticket, a customer's ceramic planter arrived damaged, meant as a birthday gift. Before Aivot, resolving this meant the rep manually digging through order history, checking the carrier status, and finding the right policy before drafting a reply. Instead, the AI panel does that work automatically and shows its receipts:
Order history pulled: surfaces the customer's order and lifetime value instantly
Carrier checked: confirms delivery status directly from FedEx
Policy matched: identifies the customer is eligible for a free replacement under the damaged item policy
Draft reply generated: writes a response at 91% confidence, citing the damaged item replacement policy as its source
The rep isn't asked to trust this blindly. They can see exactly why the AI landed on this answer, approve it in one click, and it drops straight into the reply composer ready to send. If the AI got it wrong, the same panel gives them a flag-as-incorrect option instead.
This is the core pain point solved: the repetitive troubleshooting loop, pulling order history, checking shipping, finding the right policy, gets compressed into a single AI-assisted step, without asking the rep to give up visibility into how the answer was reached.
This is where the hypothesis gets tested: does the AI actually cut down the repetitive manual work reps were stuck doing before?
A rep opens a high-priority ticket, a customer's ceramic planter arrived damaged, meant as a birthday gift. Before Aivot, resolving this meant the rep manually digging through order history, checking the carrier status, and finding the right policy before drafting a reply. Instead, the AI panel does that work automatically and shows its receipts:
Order history pulled: surfaces the customer's order and lifetime value instantly
Carrier checked: confirms delivery status directly from FedEx
Policy matched: identifies the customer is eligible for a free replacement under the damaged item policy
Draft reply generated: writes a response at 91% confidence, citing the damaged item replacement policy as its source
The rep isn't asked to trust this blindly. They can see exactly why the AI landed on this answer, approve it in one click, and it drops straight into the reply composer ready to send. If the AI got it wrong, the same panel gives them a flag-as-incorrect option instead.
This is the core pain point solved: the repetitive troubleshooting loop, pulling order history, checking shipping, finding the right policy, gets compressed into a single AI-assisted step, without asking the rep to give up visibility into how the answer was reached.
This is where the hypothesis gets tested: does the AI actually cut down the repetitive manual work reps were stuck doing before?
A rep opens a high-priority ticket, a customer's ceramic planter arrived damaged, meant as a birthday gift. Before Aivot, resolving this meant the rep manually digging through order history, checking the carrier status, and finding the right policy before drafting a reply. Instead, the AI panel does that work automatically and shows its receipts:
Order history pulled: surfaces the customer's order and lifetime value instantly
Carrier checked: confirms delivery status directly from FedEx
Policy matched: identifies the customer is eligible for a free replacement under the damaged item policy
Draft reply generated: writes a response at 91% confidence, citing the damaged item replacement policy as its source
The rep isn't asked to trust this blindly. They can see exactly why the AI landed on this answer, approve it in one click, and it drops straight into the reply composer ready to send. If the AI got it wrong, the same panel gives them a flag-as-incorrect option instead.
This is the core pain point solved: the repetitive troubleshooting loop, pulling order history, checking shipping, finding the right policy, gets compressed into a single AI-assisted step, without asking the rep to give up visibility into how the answer was reached.
Reflection
Reflection
Reflection
💭
💭
The AI transparency gap surfaced from two independent angles, competitive analysis and persona research, which made it a strong signal rather than a guess. Focusing the case study on the rep kept every decision, from the flow to the wireframes, grounded in one clear day-to-day experience. The working AI flow shows the hypothesis holding up in practice, the repetitive research work reps used to do manually now happens automatically, without sacrificing the visibility reps need to trust the answer.
Beyond the product itself, this project pushed my own process forward. Starting from a single low-fidelity wireframe, I used AI agents to build out the platform itself and to generate the animations shown in the demo, essentially using AI to build AI. That experience taught me a lot about directing AI tools as a design partner rather than just a feature I was designing for, prompting, iterating, and knowing when to step in versus when to let the AI carry a task. In the end, this project was as much about learning to design with AI as it was about designing for it. Next steps would include usability testing the flag-as-wrong flow and validating whether the confidence score actually changes rep behavior before sending a response.
The AI transparency gap surfaced from two independent angles, competitive analysis and persona research, which made it a strong signal rather than a guess. Focusing the case study on the rep kept every decision, from the flow to the wireframes, grounded in one clear day-to-day experience. The working AI flow shows the hypothesis holding up in practice, the repetitive research work reps used to do manually now happens automatically, without sacrificing the visibility reps need to trust the answer.
Beyond the product itself, this project pushed my own process forward. Starting from a single low-fidelity wireframe, I used AI agents to build out the platform itself and to generate the animations shown in the demo, essentially using AI to build AI. That experience taught me a lot about directing AI tools as a design partner rather than just a feature I was designing for, prompting, iterating, and knowing when to step in versus when to let the AI carry a task. In the end, this project was as much about learning to design with AI as it was about designing for it. Next steps would include usability testing the flag-as-wrong flow and validating whether the confidence score actually changes rep behavior before sending a response.
The AI transparency gap surfaced from two independent angles, competitive analysis and persona research, which made it a strong signal rather than a guess. Focusing the case study on the rep kept every decision, from the flow to the wireframes, grounded in one clear day-to-day experience. The working AI flow shows the hypothesis holding up in practice, the repetitive research work reps used to do manually now happens automatically, without sacrificing the visibility reps need to trust the answer.
Beyond the product itself, this project pushed my own process forward. Starting from a single low-fidelity wireframe, I used AI agents to build out the platform itself and to generate the animations shown in the demo, essentially using AI to build AI. That experience taught me a lot about directing AI tools as a design partner rather than just a feature I was designing for, prompting, iterating, and knowing when to step in versus when to let the AI carry a task. In the end, this project was as much about learning to design with AI as it was about designing for it. Next steps would include usability testing the flag-as-wrong flow and validating whether the confidence score actually changes rep behavior before sending a response.