Productivity and customer experience are usually treated as separate projects: one owned by operations, the other by sales or service. AI is unusual because when companies use AI to improve productivity, the same capability can improve customer experience too — the assistant that answers a customer at 11 p.m. also frees an employee from answering it at 9 a.m. This article looks at where that double benefit is real, where it is oversold, and how Saudi companies can pursue it with clear measurement and sensible safeguards.
It deliberately contains no invented statistics. Vendor claims of "60% faster" or "3× productivity" depend on the task, the baseline and how the test was run; the way to know what AI does for your company is to measure it yourself, and this guide shows how.
The context: AI is a national priority
Saudi Arabia's National Strategy for Data and AI (announced 2020) sets the Kingdom's aim of becoming a leader in AI by 2030, and the Saudi Data and AI Authority (SDAIA), established in 2019, coordinates the national agenda. This matters to a business because it shapes talent supply, public-sector demand for digital services and the expectations customers bring: many now assume a company can answer them digitally, quickly and in Arabic.
Where AI helps productivity
Reducing "work about work"
Much of a working day goes to searching for information, updating status, writing routine messages and compiling reports. Assistants that can read your company's own data help most here: "which invoices are overdue by more than 30 days?", "what is the status of the Al-Rashid project?", "draft a reply to this supplier." The time saved per query is small; the cumulative effect across a team is not.
First drafts, not final answers
AI is strongest at producing a good first draft — a quotation, a proposal section, a meeting summary, an email — that a person then checks and refines. This keeps quality and accountability with people while removing the blank-page effort. See quotations and proposals.
Routing and prioritisation
Classifying incoming requests, tagging them and assigning them to the right team stops work from sitting in the wrong queue. The helpdesk and shared inbox patterns build on this.
Consistent execution of routine procedures
Reminders that always go out, approvals that always route to the right person and reports that always run on schedule reduce the variation that comes from busy people forgetting things.
Where AI helps customer experience
| Customer moment | What AI can do | What must stay human |
|---|---|---|
| First contact, any hour | Answer routine questions instantly in Arabic or English from approved content | Complaints, sensitive situations, exceptions |
| Choosing a product or service | Explain options, gather requirements, suggest the right team | Custom pricing, negotiation |
| Waiting for a response | Acknowledge, set expectations, keep the customer informed | Resolution of complex issues |
| Getting a quotation | Prepare it faster and consistently | Approval of commercial terms |
| After the sale | Send reminders, status updates and follow-ups | Relationship building, account reviews |
| Repeating themselves | Hand over to a person with the full conversation history | The conversation itself |
Customers care about three things above all: getting an answer quickly, getting the right answer, and not having to repeat themselves. AI is good at the first, only as good as your knowledge on the second, and useful for the third when hand-over includes context.
The trade-offs to manage honestly
- Speed versus accuracy. A fast wrong answer is worse than a slower right one. Ground the AI in approved content and let it say "I will check with the team".
- Automation versus warmth. Some customers want a person. Make that easy and never disguise the assistant as a human.
- Personalisation versus privacy. Using customer data to tailor service is valuable but is regulated. Personal Data Protection Law obligations, enforceable since 14 September 2024, apply (overview).
- Efficiency versus employee trust. If staff feel monitored or threatened they will route around the tools. Involve them.
- Hallucination. Generative AI can state false things fluently. Design for it: grounding, approved sources, review, and a clear escalation path.
A measurement plan that separates fact from hype
Choose a small set of metrics per goal, record a baseline for four weeks, then compare after the pilot:
| Goal | Metric | How to capture it |
|---|---|---|
| Productivity | Hours per week on the targeted task | Sample time logs or task counts before and after |
| Productivity | Turnaround time (e.g. enquiry to quotation) | Timestamps in the system |
| Productivity | Rework rate | Count of items returned or corrected |
| Customer experience | First-response time and resolution time | Conversation and ticket timestamps |
| Customer experience | Customer satisfaction (short post-interaction question) | One-tap rating after resolution |
| Customer experience | Share resolved without repeating information | Review of hand-overs |
| Quality | Wrong or unhelpful AI answers per 100 conversations | Weekly review of a random sample |
| Commercial | Enquiry-to-quotation and quotation-to-order conversion | CRM and quotation records |
Include the quality metric. Without it, efficiency gains can hide a decline in customer experience.
A pilot design in detail (four to six weeks)
A pilot is an experiment with a hypothesis, a control and an end date — not a soft launch. Here is a template you can adapt.
| Week | Activity | Decision or output |
|---|---|---|
| 0 | Choose one customer journey (for example, first contact on WhatsApp) and one internal process (for example, quotation drafting). Record the four-week baseline metrics you chose. | Written hypothesis: "If AI answers routine first-contact questions from approved content, first-response time falls and reception hours fall, without lowering satisfaction." |
| 1 | Prepare approved knowledge; define what the AI may answer alone, what needs review and what it must never do; name the reviewer. | Approved answers, boundaries, hand-over rules |
| 2 | Go live for a limited group (one channel, one team, or business hours only). Review every conversation or draft daily. | Log of wrong or unhelpful answers; quick fixes |
| 3–4 | Keep running; review a random sample of at least twenty conversations or drafts per week; collect staff and customer feedback. | Weekly quality score; gaps in knowledge closed |
| 5–6 | Compare with the baseline; check quality and satisfaction did not fall; estimate total effort including review time. | Decision: extend, adjust, or stop — and a written reason |
Two rules keep the pilot honest: decide the success measures before you start, and count the review effort as a cost. If the pilot only looks good when review time is ignored, it has not succeeded yet.
Designing the customer experience around the AI
- Be transparent. Tell customers when they are talking to an automated assistant and how to reach a person.
- Keep answers short and specific. A two-sentence answer with a next step beats a paragraph of policy.
- Ask before assuming. When a question is ambiguous, the assistant should ask a clarifying question rather than guess.
- Match the language. Reply in the language the customer used; do not force a switch.
- Close the loop. After a hand-over, the customer should know who will respond and when.
- Learn from failures. Every escalation and every 'that did not help' is data for improving the knowledge.
Six steps to get real benefit
- Pick one customer journey and one internal process where delay or repetition is most visible.
- Prepare approved knowledge and clean data. This decides accuracy more than the choice of model.
- Set the rules: what AI answers alone, what needs review, what it must never do.
- Pilot with real users for four to six weeks and read real interactions.
- Compare to baseline; decide; extend, adjust or stop.
- Assign ownership: a named person maintains knowledge and reviews quality monthly.
Safeguards worth having from day one
- AI answers are grounded in approved company content, and staff can see the source used.
- Customers can always reach a person, and the assistant never pretends to be one.
- Access is role-based; an assistant cannot reveal data the asking user is not allowed to see.
- Actions that touch money, contracts or personal data need a human approval or an audit trail.
- Conversations and automated actions are logged and retained for a defined period.
- A person reviews a sample of AI answers regularly and updates the knowledge.
- The company has considered its obligations under the Personal Data Protection Law.
For a structured view of AI risk, see the NIST AI Risk Management Framework and the OECD AI Principles. SDAIA also publishes guidance on the responsible use of AI in the Kingdom on its website.
Examples of what this looks like in WaselTech
- A customer messages on WhatsApp at night; the WhatsApp AI assistant answers from the company's own website, PDFs and FAQs, and hands over to a team member with the history if asked.
- A manager asks the AI assistant which invoices, tickets or tasks are overdue and receives an answer based on data they are permitted to see.
- A salesperson receives a drafted quotation to review and approve rather than starting from a blank page.
- Reports are exported as branded PDF or Excel in Arabic or English from the reporting module.
Plans and included modules are listed on the pricing page.
Measure before you believe anyone — including us. Talk to WaselTech about a pilot with a baseline, a target and a review date, so the decision is made on your own numbers.
