How to improve the ROI of your AI stack with real customer context

The impact of customer context on AI workflows

Most AI tools are powerful pattern engines.

They can generate code, summarize documents, create design variations, write copy, propose product flows, and produce research plans. But unless they have access to the right context, they are often optimizing from general patterns rather than specific customer understanding.

That is a problem for any team using AI to create customer-facing work.

A model may know what a typical checkout flow looks like. It may not know why your customers abandon yours.

It may know how to generate a landing page. It may not know which customer objections your sales, product, and research teams have already uncovered.

This is where customer understanding becomes AI infrastructure.

Customer insight gives AI systems better input:

• real customer needs
• recurring friction points
• verbatims
• behavioral patterns
• usability findings
• audience-specific objections
• evidence from prior studies

Having this information gives us the tools to identify exactly what outcomes need to be improved, and how we can improve them.

The new dimensions for evaluating AI outcomes

AI ROI is often discussed in terms of cost and productivity. Those factors matter, but they are incomplete on their own.

A more useful way to evaluate AI is to consider five dimensions together: 

• The cost of generating work, 
• The speed at which it is produced, 
• The quality of the output, 
• The impact on customers
• The amount of rework required along the way.

Most organizations have visibility into the first two. The latter three determine whether AI is creating meaningful value.

Customer context influences all of them. It improves the quality of initial output, reduces unnecessary iteration, and increases the likelihood that what is produced will perform.

We created the following 5-step operating model to consistently evaluate these dimensions, giving us a clear and actionable method for proving the impact of the AI tools we use.

A practical operating model for improving AI ROI

Improving AI ROI requires changing how work flows through the AI tools you’re currently using.

The organizations seeing the most value from AI are not simply generating more. They are increasing the percentage of AI-generated work that is usable, validated, and aligned with real customer needs. That shift can be operationalized.

What follows is a simple model for doing that consistently.
 

1. Identify your highest-cost AI workflows

Start by creating a simple inventory of where AI is actually used.

Pull in data from engineering, product, design, marketing, and support, and list:

• the workflow (e.g., “generate landing pages,” “prototype new flows,” “write backend code”)
• the tools used (ChatGPT, Claude, Cursor, etc.)
• approximate usage (high / medium / low)
• estimated cost (seats, tokens, or time spent reviewing output)

Then ask:

• Which workflows consume the most AI spend?
• Which ones generate the most output?
• Which ones lead directly to product or customer-facing decisions?

Put this into a simple table. You are not trying to be perfect—you are trying to identify the 3–5 workflows that matter most.

Decision:
 Prioritize a small set of high-cost, high-impact workflows. Ignore the rest for now.

2. Measure output quality, not just activity

For each prioritized workflow, audit what happens after AI generates something.

Take a recent sample (last 1–2 weeks) and review outputs with the team that uses them.

Track:

• Was the output used as-is, lightly edited, or heavily rewritten?
• How many revision cycles did it go through?
• How much time was spent reviewing or fixing it?
• Did it move forward or get abandoned?

You can capture this in a simple scoring system:

• 3 = used as-is
• 2 = minor edits
• 1 = major rewrite
• 0 = discarded

Then ask:

• What % of outputs are actually usable?
• Where are we spending the most time fixing AI output?
• Which workflows consistently produce weak first drafts?

Decision:
 Identify the workflows with high rework and low acceptance. These are your biggest ROI opportunities.

3. Diagnose the context gap

Now look at the workflows with the highest rework and ask a different question:

What did the AI know when it generated this?

For each workflow, review:

• the prompts or inputs used
• the brief or instructions given
• what customer information (if any) was included

Ask:

• Did this include real customer evidence (verbatims, research, usability findings)?
• Or was it based on general instructions and internal assumptions?
• Are teams relying on “best practices” instead of known customer behavior?
• When does customer feedback enter the process—before or after generation?

Document this simply:

• “Context used: none / limited / strong”
• “Customer evidence included: yes / no”

Patterns will emerge quickly.

Decision:
 Flag workflows where output quality is low AND customer context is weak or missing. These are context gaps you can fix.

4. Inject customer context upstream

For the workflows you flagged, change how inputs are created before AI is used.

Start by gathering existing customer evidence:

• recent user interviews
• usability findings
• support tickets
• common objections from sales
• behavioral patterns or drop-off points

Then standardize how that context is used.

For example:

• Add a “customer context” section to product briefs
• Require prompts to include 3–5 real customer insights
• Create reusable context blocks (e.g., “Top onboarding friction points”)

Instead of:

“Generate a signup flow”

Teams should be working from:

“Generate a signup flow using these known user frustrations and behaviors”

You do not need new research to start—use what you already have.

Decision:
 Update workflows so that AI is always prompted with customer evidence, not just instructions.

5. Validate early and reduce rework

Once output improves, introduce validation before further investment.

For each workflow:

• Select a small set of AI-generated outputs (designs, flows, content)
• Test them with real or representative users
• Compare variations where possible

Ask:

• Do users understand this?
• Where do they struggle?
• Which version performs better, and why?

Track:

• number of revision cycles before approval
• time to final decision
• major issues found before vs after launch

Then compare against your baseline from Step 2.

Decision:

• Stop investing in outputs that fail validation early
• Scale the workflows where validated outputs move forward with fewer revisions

Over time, you should see:

• fewer discarded outputs
• faster approvals
• less downstream rework

How MCP enables this workflow

Model Context Protocol, or MCP, is an open standard that allows AI tools to connect to external services and data sources.

For UserTesting, MCP creates a way to bring real customer understanding closer to the AI tools teams already use — including tools such as Claude, ChatGPT, Figma Make, Cursor, and other MCP clients.

That changes the AI workflow in three important ways.

1. AI can work from customer evidence, not just general patterns

Instead of prompting an AI tool from memory or assumptions, teams can bring in real UserTesting findings, customer themes, verbatims, and behavioral patterns.

That means the first draft, design direction, prototype, or product brief can start from a stronger understanding of what customers actually think, feel, and do.

2. Teams can move from question to test faster

If a team does not have enough evidence yet, MCP-supported workflows can help move from a question to study creation faster. Teams can define what they need to learn, generate or refine a test plan, and connect to recruiting through User Interviews.

That matters because AI-generated work needs fast validation. Otherwise, teams can create faster than they can learn.

3. Insights can show up where decisions happen

Insights create more value when they reach teams at the moment of decision. MCP helps bring UserTesting closer to the workflows where product, engineering, design, and marketing teams are already generating ideas, reviewing data, and making tradeoffs.

This is the shift from UserTesting as a research destination to UserTesting as a customer-context layer inside AI-assisted work.

Make customer understanding part of your AI infrastructure

AI tools are becoming embedded in how organizations build, design, decide, and communicate. The return from those tools depends on the quality of the inputs they receive and the feedback loops that shape their output.

Improving AI ROI does not start with buying more tools or limiting usage. It starts with increasing the percentage of work that is useful, validated, and aligned with customer needs.

UserTesting’s MCP functionality helps bring real customer understanding into AI-assisted workflows so teams can generate stronger first drafts, test ideas faster, recruit the right participants, and make decisions grounded in what customers actually think, feel, and do.

That is how organizations move from AI output to AI impact.