A wrong recommendation starts with the question AI never asked.
Most ecommerce assistants match keywords instead of understanding the need. They skip the question, miss the constraint and often recommend the wrong product. isabeto does the opposite: it asks first, remembers the context and recommends the right product with a clear reason. The same advisor works across your website and WhatsApp.
Real queries, real results
Müşteri
"I'm looking for a keyboard"
Result
Keyboard Pennant Ribbon
Live AI assistants in Turkey · July 2026 · brands withheld
Each conversation solves a difficult buying moment specific to that industry. The side panel shows the sales move behind every message. Change the tone: the voice changes, but the reasoning and safety boundaries do not.
Hard momentAn iron supplement can be a poor recommendation when medication timing is unknown. The useful question is not 'which brand?' but 'what are you already taking, and when?'
MMayaWellbeing Lab · isabeto
Type a message…↑
Brands and products in this demo are illustrative. Your demo is prepared with your catalog and your preferred sales tone.
Evidence file
Not an edge case. A pattern.
These are real queries and real results. In Baymard's 2026 benchmark, product-type searches performed poorly on 20% of the ecommerce sites and apps tested. Asking for a keyboard and getting a keyboard ribbon is a visible example of that wider product-finding gap.
A live ecommerce AI assistant in Turkey · July 2026 · brand withheld
The keyword matched. The product need did not.
An AI assistant
I'm looking for a keyboard
Here are the keyboards you asked for:
Marketplace seller
Keyboard Pennant Ribbon
$7.50
product image
Marketplace seller
61-Key Mechanical Gaming Keyboard
$38.00
product image
Marketplace seller
Full-Size Hot-Swap Mechanical Keyboard
$62.00
product image
I hope these products help. Can I help with anything else?
isabeto, same question
I'm looking for a keyboard
The right keyboard depends on how you use it. What is the main setting?
Office and writingGamingMixed home use
It does not dump products. One useful question narrows the need.
×The first result is not a keyboard. It is a keyboard ribbon; the words matched, the product type did not.
×Not one discovery question: work or gaming, budget, layout, noise or switch preference.
×A compact 61-key model and a full-size model appear together without explaining the trade-off.
×The generic closing repeats without moving the sale forward.
Evidence B
A live ecommerce AI assistant in Turkey · July 2026 · brand withheld
It sells a brake part, but cannot tell whether it fits the car.
An AI assistant
I need brake pads for a Megane
Here are suitable brake pad options for Megane:
Auto parts seller
Megane 1.9 DTI Brake Vacuum Pump
$115.00
product image
Auto parts seller
Laguna II Brake Master Cylinder
$98.00
product image
Auto parts seller
Megane IV Front Brake Pad Set
$31.00
product image
I hope these products help. Can I help with anything else?
Do these fit my 2024 Megane?
I cannot help with that. I can only help with shopping and promotions.
isabeto, same question
I need brake pads for a Megane
Megane spans several generations with different brake pads. What model year and engine is yours?
Before 20102010–2016After 2016
Fitment cannot be guessed from a model name, so it asks first.
×The first result is a vacuum pump, not a pad. The second is for a different vehicle.
×The assistant never asked for model year, generation, engine or axle.
×The only actual pad is for Megane IV, which still may not match the customer's car.
×It gives up on the most important question: does it fit?
And the pattern continues
Evidence C
"I need sunscreen for my baby"
SPF 50 Baby Sunscreen
It sells baby sunscreen without asking the baby's age.
Evidence D
"I need an iPhone case"
iPhone Case Back Charm
It never asked the iPhone model; 'cheaper' made it leave the category.
Evidence E
"I need 16 GB RAM"
16 GB DDR4 Desktop Memory
It sells RAM without asking about the motherboard.
Evidence F
"I need a cable"
CAT6 Ethernet Cable
It matched 'cable' and ignored what the cable needed to do.
Test an AI assistant with three questions.
A good answer does not start with a product list. It asks for the missing decision input first.
“I need a gift. What do you recommend?”
It should askFor whom, for what occasion and at what budget?
“Is there a cheaper one? What is the difference?”
It should askWhich feature can the customer give up?
“I use X. Will this work with it?”
It should askWhat information determines compatibility?
Research on AI-assisted shopping
Customers are ready for AI. Not blind recommendations.
People have started using AI to shop. Interest is not the problem; trust is. Many users still verify the information from scratch. Research shows that explaining the reason behind a recommendation helps rebuild that trust.
0125%
Adoption has started.
One in four consumers used generative AI shopping tools in 2025; another 31% plan to use them.
The isabeto differenceIt does not decide for the customer. It narrows the options with the right question and explains why the recommendation fits.
The same advisor on WhatsApp
Your customer left the site. The conversation did not.
The same customer returns a week later. They do not repeat the vehicle details; the advisor continues from the earlier sale and closes the next order without sending them back to the website. Turkey context: 88.6% of people aged 16–74 use WhatsApp, the highest measured rate among social and messaging apps (TÜİK, 2025).
AAlex · Partline Garageonline
One week later
The squeal is gone after fitting the pads. My friend also has a Megane. Can we send him the same set?19:38✓✓
Glad to hear it. Yours is a 2015 diesel Megane III, front axle. Your friend's may still differ, so what model year and engine is his?19:39
A week passed. The customer did not need to explain their car again.
Also 2015 diesel. We bought them the same year.19:41✓✓
Then the same set fits. It is the exact reference you bought last week:19:41
Megane III Front Brake Pad Set, OEM 410608481R$85
Ten seconds of discovery prevented a blind fitment assumption.
Add it. I will pay for him.19:43✓✓
Added. I am leaving the payment link here. He can show the OEM number on the box to his mechanic.19:43
Payment linkVerify with VINThat's all for now
The second order closed without sending the customer back to the website.
Type a message↑
The customer is already there
No new app, account or password. They message from a familiar place and receive an answer in seconds.
The tab closes. The conversation does not.
A website visitor can disappear. A WhatsApp conversation stays on the customer's phone until they are ready to return.
The second sale does not start from zero
When the customer returns, the advisor can continue from the relevant product, fitment and constraint instead of asking everything again.
Deployment uses WhatsApp's official business channel; the number and conversations remain in your business account.
Where revenue leaks
Two silent leaks.
Adding AI is not the finish line. The data points to two quieter failures: customers cannot find the right product, or they leave when a question goes unanswered. Neither appears in the cart.
A They cannot find the right product
46%
Share of desktop ecommerce sites rated mediocre or worse for overall search experience.
A good sales advisor asks the right questions and guides the customer to a suitable product. A human advisor cannot serve every customer at every hour. isabeto brings the same approach to your website and WhatsApp: understand the need, explain the recommendation and leave no question unanswered.
Now use your own numbers.
The fields on the left describe your store today. The sliders on the right model a possible improvement. This is not a promise; it shows what a small change means in your own revenue.
A benchmark, not a targetIn a broad current panel, trailing 12-month category conversion averages range from 0.71% in luxury and jewelry to 5.37% in beauty and personal care. Because the range is 0.71%–5.37%, we do not insert one generic industry norm into your calculation. Dynamic Yield by Mastercard · 2026
Current monthly revenue$20,000400 orders × $50
Conversion scenario0.30 point increase
2% → 2.3% scenario
Order value scenario8% growth
A relevant add-on can grow the order naturally
Monthly gain from conversion$3,000
Monthly gain from order value$1,600
Annual total$55,200
We do not promise a percentage. The improvements above are assumptions you control. Actual impact varies by industry, catalog and traffic, and is measured in a pilot. This calculator only shows what a small improvement means in your own revenue.
Turkey market context: ecommerce volume exceeded TRY 4.5 trillion in 2025 (T.C. Ticaret Bakanlığı, 2025), while WhatsApp usage reached 88.6% among people aged 16–74 (TÜİK, 2025). These are Turkey-specific figures; the demo and calculator above can be adapted to another market.
The isabeto difference
It does more than chat. It builds the right decision.
The difference comes down to three behaviors: ask the right question, explain the recommendation and stop where selling would be wrong.
The right question01
Understands without turning the sale into a survey.
It asks only for information that can change the answer, then clarifies the customer's description and use context in a short conversation.
Reasoned recommendation02
Narrows the catalog around the need.
It takes price and availability from the real catalog, then connects the recommendation to the customer's own words.
Safe boundary03
Stops where it should not sell.
It respects known constraints, does not invent uncertain information and directs real risk to the right human support.
Pre-release checks
Trust is tested before the demo.
We test different customer profiles.
Each demo scenario runs end to end across 18 distinct customer profiles.
Risky recommendations do not pass.
A flow is not approved when it invents information or crosses a defined safety boundary.
Sensitive data stays limited.
Sensitive information such as health data is not written into a persistent customer profile.
