How We Built an AI Sales Agent for a Saudi Manufacturing Company

Run the same Monday twice. In the first, a contractor in Dammam sends a WhatsApp inquiry for 200 meters of 12mm rebar, grade 60, epoxy-coated — and it sits unread until a sales rep reaches it at 8 AM, after the catalog lookup, stock check, and manual quote assembly that follow. In the second, that same inquiry gets a qualified answer in 23 seconds, in Arabic, with stock verified in Odoo and a real quotation attached. Same manufacturer, same product, same customer — the only difference is the AI Sales Agent Shayntech built for this Saudi steel manufacturer. This is the head-to-head: the human pipeline as it was, the AI-augmented pipeline as it is, and the numbers in between.

At a Glance: The Human Pipeline vs the AI Pipeline

Before we get into how it was built, here is the comparison that matters to a sales leader — task by task, the way work actually flows:

Head to Head: Where the Agent Earned Its Keep

Five capabilities carried the comparison. Each one is a workflow change, not a feature list:

Response speed: the lead that dies overnight

The manufacturer's pre-agent baseline was an eight-plus-hour response — leads arriving after 6 PM effectively waited until the next business day. Industrial buyers reward the first vendor who replies, and in Saudi Arabia the overnight window is long: inquiries land at 10 PM and customers are comparing quotes by 7 AM. The agent cut this from 8+ hours to 23 seconds — the single biggest driver of the +47% conversion lift.

After-hours coverage: Friday is a sales day

Roughly a third of this manufacturer's qualified leads arrive outside office hours — including Friday, when the commercial team is off but contractors are catching up on quotes. Before, that traffic pooled into Monday's queue. Now it converts in the moment, in the customer's language, with the same pricing discipline as office hours.

Bilingual precision: dialect matters

Off-the-shelf Arabic models mangle Najdi and Hejazi business talk — numbers, unit shorthand, and negotiation idioms. Shayntech fine-tuned the extraction layer on the client's real WhatsApp transcript corpus, lifting Arabic intent recognition to 97%. The agent reads "٦٠٠ متر مربع" and "600 sqm" with the same fluency.

Quoting consistency: no more stale price lists

Manual quotes varied with the rep, the day, and the last price-list update. The agent reads pricing tiers, discount rules, and delivery structures from the database — not the prompt — so every quote is consistent, auditable, and generated as a real Odoo document in under two minutes.

Pipeline visibility: the CRM updates itself

Every conversation updates the Odoo record automatically. The weekly ritual of exporting leads into a spreadsheet and asking "what happened with the Dammam contractor?" is gone — the pipeline is live and attribution is unambiguous.

The Before-and-After Numbers

Behind those three headline numbers: the same four-person sales team now handles three times the volume without new hires; quote-related manual work fell by 96%; and the agent paid for itself within eight weeks — attributing over SAR 1.2M in new pipeline value to leads captured and qualified outside business hours, business that would have gone to competitors who answered first.

Beyond Speed: What the Agent Wasn't Built to Do

A head-to-head that ends at "faster" misses the point. The agent was deliberately kept out of the parts of selling where judgment is the product:

• Negotiation and relationships: The agent doesn't close SAR 500K deals or rebuild trust after a bad delivery — it creates the conditions (fast, accurate, responsive) that let humans close.
• Ambiguity escalates: Requests that fall outside confident thresholds route to a human rep immediately, protecting trust over speed.
• High-value sign-off: Anything above a value threshold needs a human signature before it leaves — the agent prepares, the team approves.
• The messy middle: Multi-round negotiation, credit terms, and delivery disputes stay human. The agent hands off with full context, never a cold transfer.

Who Benefits Most

Where to Start: Running Your Own Before-and-After

The build took six weeks end to end. The sequence that made it repeatable:

1. Measure the baseline first: Response time, quote turnaround, and the after-hours leak — you can't claim a before-and-after without the before.
2. Map the sales process as a state machine: Three days diagramming every route a deal can take is what lets the agent navigate without guessing.
3. Collect the transcript corpus: Real sales conversations are the raw material for dialect-accurate extraction — generic models won't cut it.
4. Run shadow mode for two weeks: The agent works beside the team on live traffic, every draft compared with the rep's real reply, before it talks to a single customer.