Contact usENعربي

Case studyE-commerce technology

Saudi E-commerce Technology Company: AI That Answers From the Catalogue, and Admits What It Doesn't Know

Answer trace

The shopper asks

المولّد هذا يشتغل بالديزل؟ وكم مدة الضمان؟

Does this generator run on diesel? And how long is the warranty?

The catalogue record

Standby generatorGN-7500

Fuel type
Diesel
Rated output
7.5 kVA
Warranty term
Not in the catalogue

The agent replies

نعم، يشتغل بالديزل حسب بيانات المنتج. أما مدة الضمان فالكتالوج ما يذكرها، فما أقدر أؤكّدها لكم.

Yes, it runs on diesel, per the product data. The catalogue does not state the warranty term, so I can't confirm it.

Cites the fuel fieldMissing field: it says so instead of guessing

Illustrative record on a fictional product.

A Saudi e-commerce technology company wanted a sales agent that answers shoppers only from the store's own catalogue, in Arabic and English. After a paid AI Solution Blueprint, Ting built the platform and handed it over deployed and working; real merchants can install it once the store marketplaces approve it.

Industry
E-commerce technology
Location
Saudi Arabia
Status
Platform delivered; merchant installs pending
Challenge
Accurate answers on the product page, from the store's own catalogue

By the Ting teamUpdated Client name withheld for confidentiality.

Shopper journeysQuality check before handover
  1. 01Known product mismatchPendingPass
  2. 02Honest trade-off between two productsPendingPass
  3. 03Missing fact, admitted, not guessedPendingPass
  4. 04Memory across follow-up questionsPendingPass
  5. 05Prices written in ArabicPendingPass
  6. 06–08Three further journeys in the same scriptPendingPass ×3

8 of 8 journeys passedHandover check, 10 Oct 2026

Journeys as named in the quality script, run on the deployed platform.

The challenge

A shopper's question on a product page is usually specific: what does it run on, what does the warranty cover, how does it compare with the model beside it. A general-purpose chatbot can answer that fluently and still be wrong, and for a merchant a confident wrong answer about a product is worse than no answer at all.

The company's goals set the bar for the build. The agent had to answer only from the store's own product data and say plainly when that data did not hold the answer. It had to speak Arabic the way Saudi shoppers write it, and follow a shopper who switches to English. And the merchant needed to see the conversations: which questions repeat, which objections come up, and which product descriptions leave shoppers asking.

Our approach

  1. An AI Solution Blueprint first. Ting delivered a paid Solution Blueprint for the sales agent, so the company could judge the idea before funding the platform.

  2. A grounded agent with guardrails. The agent answers from the catalogue and falls back honestly when a fact is missing. Its quality check runs eight shopper journeys, including a known product mismatch, an honest trade-off, a missing fact, memory across follow-up questions and prices written in Arabic.

  3. A merchant dashboard. Merchants see their KPIs, every conversation in detail, insights drawn from repeated questions and objections, suggested product-page improvements, and their integrations and settings.

  4. Connected to the stores. Connections to two store platforms were built and verified on developer demo stores, with live answers drawn from synchronised catalogues.

  5. Handed over as a product the team can run: tested, documented, and deployable without us.

Outcomes and impact

Kept apart on purpose: what was handed over, what has been proven, and what is still to be measured.

  1. DeliveredHandover, 10 Oct 2026

    The platform, handed over deployed and working. When the catalogue does not hold an answer, the agent now says so instead of guessing, and merchants can see every conversation.

  2. ProvenHandover, 10 Oct 2026

    All eight shopper journeys pass on the deployed platform, and both store connections were verified on developer demo stores.

  3. Still open

    Approval on the two store marketplaces, and installation by real merchants. Merchant, conversion and revenue figures will be reported here once they are measured.

No conversion or revenue estimate is published for this platform.

Why this matters

An AI agent that talks to customers is only as good as the data it is allowed to use. The boundary here was simple to state and hard to build: answer from the catalogue, admit what the catalogue does not say, and leave a record the business can learn from.

That pattern reaches well beyond e-commerce. Any company that puts AI in front of customers, from dealers and distributors to service desks, needs the same boundary. A paid AI Solution Blueprint is the cheapest place to settle it: it defines how the agent must behave on real questions, written as acceptance tests, before the full build is funded.

More case studies

All case studies

Sound like your operation?
Let's map yours.

Contact usSee Custom Implementation