Ask before assuming.
Use the conversation to understand context and constraints, so guidance reflects the customer’s actual need.
Selected work
Beyond search.
AI Shopping Assistant
Beyond search.
Towards a better decision.
Shaping the product experience of an AI shopping assistant: from understanding a customer’s intent to helping them choose with confidence.
- Organization
- Digikala
- Period
- 2024–Present
- My responsibility
- Product experience strategy · Design direction · Roadmap alignment
The experience model
“I know what I need.
I don’t know what to choose.”
-
01
Understand
Intent, context, constraints
-
02
Guide
Relevant options and explanations
-
03
Support
An informed customer decision
The problem
More choice does not always mean more clarity.
Search is useful when the customer already knows what to ask for. The harder problem starts before the query is clear.
In a large commerce ecosystem, people arrive with different levels of certainty. Some need to compare alternatives; others are trying to translate a situation, preference, or constraint into a useful direction.
The assistant’s opportunity is to support that decision–not just return another product list.
My remit
Connect the ambition
Connect the ambition
with the experience.
I lead product experience strategy for Digikala’s AI Shopping Assistant, from ideation and the interaction model through roadmap alignment and rollout.
I define conversational discovery, intent-based shopping, personalized recommendations, and decision-support experiences. The work includes aligning executive, product, engineering, and data stakeholders around customer value, feasibility, and scale.
Product direction
Interaction behavior
Cross-functional alignment
Design decisions
Trust is a behavior.
Make guidance understandable.
Give people a way to understand why an option is relevant and correct a preference when the assistant gets it wrong.
Keep exploration open.
Let customers move between recommendations and browsing. The assistant supports judgment while the customer stays in control.
A representative journey
Move from a request
Move from a request
to a defensible choice.
A customer may ask for “a good phone for my father.” A useful assistant should not jump straight to a ranked list.
Clarify what “good” means.
Ask about budget, eyesight, battery expectations, familiarity with technology, and the tasks that matter most.
Explain the trade-offs.
Show why a smaller set of options fits the situation, where each option compromises, and which assumptions shaped the recommendation.
Keep correction easy.
Let the customer revise a constraint, compare with the wider catalog, or leave the conversation without losing their place.
This is a simplified illustration of the interaction model rather than a published product screen.
The hard product decision
Helpful is not the same
Helpful is not the same
as ready to release.
As the system matured, the central question shifted from whether the assistant could produce a plausible answer to whether it could do so reliably, quickly, and at a cost that made sense at ecosystem scale.
That required clearer ownership between orchestration, conversation state, recommendation logic, and the quality gate before a response reached the customer. It also required testing the combinations that were still uncovered–not treating a large test count as proof of readiness.
Classify failure by consequence.
A minor tone issue can degrade gracefully. An irrelevant product, broken reference, or ignored customer constraint should block or repair the response.
Measure the whole experience.
Quality has to include relevance and constraint-following alongside latency, model calls, repair rate, and cost per interaction.
Connect quality to behavior.
Internal evaluation is only an intermediate signal. The product must ultimately improve meaningful shopping behavior such as exploration, add-to-cart, conversion, and abandonment.
Scale & status
Ongoing work.
Ongoing work.
A broader leadership remit.
40M+
users in Digikala’s wider ecosystem–not assistant adoption
20+
designers & researchers led
15+
product squads receiving direction
The product remains in staged development and rollout. I separate the scale of the opportunity from product adoption: the 40M+ figure describes Digikala’s wider ecosystem, not the number of people using the assistant.