Introduction

“I need a bag for my laptop and a change of clothes. Something I can take on a short business trip.”

That is a reasonable way to begin shopping. It expresses a need more naturally than selecting a product category and working through a long set of filters.

Now imagine the response: a voice reads the names, dimensions, prices and features of six bags. By the fourth option, the customer is trying to remember which of the first three fitted their laptop.

Speech helped explain the need. It did not remove the need to see the choices.

Useful voice commerce brings those capabilities together. Customers should be able to describe what they want, inspect relevant options and confirm an exact choice, without being forced to conduct every step aloud.

Voice is one part of the customer experience

Voice shopping can mean several things, including smart-speaker purchases and phone-based assistants. This article focuses on a customer speaking with a business’s AI agent while using a screen.

The screen is important because some decisions benefit from simultaneous information. You can compare two dimensions at a glance. You can inspect a colour. You can check a date without asking the agent to repeat it.

Nielsen Norman Group has long described the complementary strengths of speech and visual output: spoken input can make requests convenient, while a display lets people scan information without holding every detail in memory. Its guidance also warns against restricting visual controls simply to preserve a voice-only experience. Read the voice-and-screen analysis.

For a business owner, the relevant question is therefore broader than “Does the AI sound good?” It is “Can a customer understand their options and complete the right next step?”

1. Say the need, then see a useful starting point

Consider an illustrative luggage shop. The customer describes a laptop, overnight clothing and a short business trip.

A helpful agent might first ask which laptop size the bag must accommodate. It should avoid a long questionnaire when one answer would materially narrow the selection.

The next view could show a small set of suitable options with the details that explain their relevance: laptop compartment size, capacity, dimensions and price. The spoken response can summarise the main difference rather than reading every line on the screen.

For example: “These two have a separate laptop compartment. The compact option is lighter; the larger one gives you more room for clothes.”

The customer now has something to inspect. They can ask another question, select an option or change their mind. A useful visual response creates a clearer next step than a paragraph ending with instructions to search the catalogue.

2. Ask a comparison, then see the differences that matter

“Which one is better for travelling with only cabin luggage?”

The agent should use verified product information and avoid assuming that one airline’s limits apply to all flights. It may need to ask about the airline, or explain that the customer should compare the bag’s dimensions with their ticket’s allowance.

The visible comparison should focus on the decision: dimensions, capacity, weight, laptop fit and relevant access features. Repeating a complete specification sheet for each bag makes the customer do the comparison themselves.

If the business does not have a reliable product weight, the agent should say so. Filling a blank with a plausible number makes the experience less trustworthy, however natural the voice sounds.

The customer should also be able to select the two options directly. Nielsen Norman Group’s research on emerging generative interfaces explains how contextual controls can reduce the need to memorise and retype names or choices during an AI conversation. Read its analysis of buttons and checkboxes in GenUI.

In practice, speaking and clicking can support the same task. There is no reason to make someone pronounce a long product name when they can point to it.

3. Choose an option, then review the exact action

“The smaller one, in black.”

Before the customer proceeds, the selected product and variant should be unambiguous. Show the name, image, colour, quantity and price. Where delivery choices or other charges affect the total, make those clear at the appropriate checkout step.

The agent’s words need to match the transaction state. Adding an item to a bag is different from placing an order. Selecting a workshop is different from confirming a date and booking it.

For a service, the review might include the service name, location, duration, date, time and payment arrangement. For a physical product, the important details will differ. The common requirement is that the customer can inspect what they are committing to and correct a mistake before confirmation.

Speaking confidently is not evidence that an action succeeded. The confirmation should reflect the actual outcome.

Let customers change modes without restarting

A person may begin by speaking, switch to text in a quiet office and finish by clicking. Another may prefer reading because an unfamiliar product name is easier to understand on screen.

Treat those choices as normal behaviour. The experience should preserve the relevant selection and conversation context rather than making someone describe their requirements again whenever they change input method.

This also means testing ordinary interruptions: the customer corrects a colour, asks the agent to stop speaking, denies microphone permission or chooses not to use voice at all. A microphone should expand the ways someone can interact with the business, not become a gate they must pass through.

Voice and text alternatives help accommodate different needs, but their presence alone does not establish accessibility. Labels, focus behaviour, keyboard navigation and assistive-technology compatibility still need testing.

What this means for an agent-first experience

In PAKA, customers can speak with the business agent while it uses the Canvas to present products, comparisons and supported actions. Customers can also use visible controls to choose and continue. Explore PAKA’s agent and Canvas.

This framing changes what a good demonstration should prove. A pleasant voice is only the beginning. A convincing example shows how a customer’s request affects what they see, how the next question builds on that information and how the final choice remains under their control.

A short test for your own business

Choose one real customer task and run it from beginning to end. Use an unfamiliar visitor’s words, not the perfect prompt your team has rehearsed.

Check five things:

  1. Does the agent clarify only what it genuinely needs?
  2. Does the screen make the decision easier, rather than merely repeat the speech?
  3. Do the spoken details match the visible offer and actual business information?
  4. Can the customer change a choice or switch to text and clicking without losing the thread?
  5. Is the final selection distinguishable from a completed purchase or booking?

Record where the customer needs to repeat themselves or hold details in memory. Those are useful places to improve the experience.

The goal is a customer who can express a need naturally and make an informed decision. Sometimes they will speak. Sometimes they will read, compare or click. A well-designed business agent leaves room for all of those behaviours.

Further reading: What is Generative UI?

Experience the idea.Try a customer experience