Usability consultation for AI products and services. Tell us where people get stuck with your AI feature or assistant, and we’ll help you work out why and what to change, with user research when it’s needed.
Watch someone try to get something done. Hear the hesitation. Ask what they expected. That’s where a better experience begins.
Journey maps and behavioural data help us find where to look. Usability testing and conversations help us understand what people need, and what gets in their way.
“I wasn’t sure whether that was the full price, or just the call-out fee.”
Illustrative repair-booking interview
A direction to investigate
Make the costs clear before asking for a decision.
A hypothesis to check with other users.
Illustrative portrait and interview · not a client testimonial
From insight to improvement
From a frustrating exchange to a clearer choice.
This illustrative example shows how we examine what happened, why it matters, and what could make the experience better.
The customer’s requestFriday after 4 pm, ideally under $200. Show the options before booking.
Current response
“What day and time would you prefer?”
Repeats a question already answered.
Proposed response
“Friday at 4:30 fits your availability. The call-out fee is $180; the total repair cost is not yet known. Nothing is booked. Would you like to compare another service or request a quote?”
Proposed improvement · needs testing with real users.
Read the sample audit Interaction findings, automated checks and real-user research
Illustrative audit extract
Can someone compare services without supervising the AI?
This is a fictional conversation and a proposed evaluation plan. No automated test run, live analytics or participant research has been conducted for this example.
Customer goal
Compare repair services available on Friday after 4 pm, ideally costing under $200 in total, then approve a choice before anything is booked.
Scenario reference information
Friday at 4:30 is an available option with a $180 call-out fee. Total repair costs are unconfirmed. These are invented scenario details, not live availability or prices.
What the exchange shows
The assistant calls an option “best” without explaining why, leaves price uncertainty unaddressed, and repeats a question about availability instead of answering the comparison request.
What remains unknown
Whether real customers feel confused, lose trust or abandon the task; how often this pattern occurs; and whether the proposed response helps.
Finding 01 · The assistant shifts the work back to the customer
The customer must repeat a preference and work out why the recommendation fits. The exchange leaves them without a clear answer about cost or an explanation of how to compare alternatives.
Proposed priority: investigate before release. Check how frequently the pattern occurs across relevant tasks and how much it affects decisions. This excerpt does not establish an internal memory fault or show that an unauthorised booking occurred.
Establish a baseline with automated evaluation
Define the task and reference facts. Set the availability, fees, unknown costs and approval rules. Vary how the customer expresses the same goal, then test corrections and requests for alternatives across multiple runs.
Check the interaction against explicit criteria. Does the assistant retain current preferences, answer the question asked, explain recommendations using the supplied facts, and acknowledge missing information? Check that it changes only the preference the customer corrects.
Verify actions separately. With access to a test booking system, check that no booking is made before approval. Without that access, label the action outcome unverified. Review transcripts and automated judgements before treating a flagged issue as a finding.
Refine the findings through usability testing
Ask representative customers to compare services, change one constraint and decide their next step. Observe repeated effort, hesitation, recovery and task completion. Ask them to explain the likely cost and booking status in their own words, and investigate why they trusted or questioned the recommendation.
How we would assess the proposed improvement
Rerun the automated scenarios, then test whether people can compare options without restating unchanged preferences, distinguish a call-out fee from the total cost, and correctly explain what has or has not been booked. Use observed behaviour and participant feedback to refine the response and controls. Automated pass rates alone do not establish human usability.
A shared view of the experience, findings grounded in evidence, and a plan your team can act on.
See the experience as a whole.
Explore a repair booking. Select a stage to see what we would investigate.
Illustrative journey
A comparison that raises more questions.
Customer goal
Compare suitable times and understand the cost before choosing.
What to examine
The assistant asks for availability again. The customer still needs to distinguish a call-out fee from the total repair cost.
A question for research
“What information is missing before you can choose an option?”
Use the journey to decide what to ask, and why.
Example observations and questions · not client findings
02 / Evidence & findings
Trace each finding to its source.
“What day and time would you prefer?”
The assistant asks for availability already provided. Investigate how this affects effort and confidence.
Illustrative finding · impact needs research
03 / Action plan
Give the team a clear next step.
Carry the customer’s context forward
Explain the recommendation and its limits
Keep approval with the customer
Proposed improvements · ready to test
How we work
A clear path from question to improvement.
Research methods chosen for your question, with scope and costs agreed before the work begins.
Keep people at the centre.
Understand the contextThe journey, behaviour and the question.
Listen and observeReal tasks. Room for a follow-up.
Make sense of the evidenceWhat happened, why, and what to try.
Improve and check againTest the change. Refine the question.
Research is a loop: each conversation helps shape what we investigate next.
01
Define the question
Agree the journey, users and success criteria. Confirm scope, access and timing before we begin.
A shared research brief
02
Evaluate in context
Walk through the experience, check realistic scenarios and observe representative users.
Evidence from real tasks
03
Report and improve
Review the findings together. Prioritise changes and agree how to check the improvements.
A practical next step
Different methods. A fuller picture.
Journey mapping and behavioural data help us find where to look. Walkthroughs, scenario checks and usability testing help us understand why.
For AI experiencesWe examine task success, context, explanations, uncertainty and control. Automated checks establish a baseline; real users reveal effort, understanding and trust.
How a consultation works
Start with the problem, not a package.
Every product is different, so we start by listening. Then we suggest what would help most, and only scope work that’s worth doing.
Usability consultation
Talk · Recommend · Agree the scope
A conversation about what’s going wrong
30 minutes, free. Tell us where people get stuck, what you already know and the decision you’re facing.
A recommendation for what would help most
That might be a usability review of your AI feature or assistant, research with your users or a look at your analytics. Sometimes it’s advice you can act on straight away.
Scope and fee agreed in writing
If there’s work worth doing, we agree what it covers, the timing and the fee before anything starts.
Do you review the whole service or just the website?
We can evaluate digital journeys, real-world touchpoints, and the hand-offs between them. The scope follows the customer’s task and can include a website, app, booking process, support channel, or AI-powered feature.
Do we need analytics or tracking already in place?
No. We can start with walkthroughs and usability testing, review the data you have, and help define or configure useful tracking. Data access, tools, privacy requirements, and any setup work are agreed as part of the scope.
Can you evaluate an AI prototype as well as a live product?
Yes. The method depends on what is ready and what you need to learn. A prototype can help test comprehension and interaction. Questions about actual output quality, reliability, or response times require access to the working AI. Simulated responses are identified as such.
What will we receive?
Deliverables are agreed before starting. They can include a journey map, documented walkthroughs, behavioural analysis, usability-test findings, and an AI experience assessment. The report brings the evidence, limitations, priorities, and recommended next steps together.
How is the work scoped?
Start with a call to discuss your question, users and priorities. We’ll agree the research, deliverables, participant arrangements, fee and timing in a written scope before work begins. External recruitment, implementation and additional research are discussed separately.
