We build the AI that runs automotive retail.

Selling and servicing cars is conversation work: advising a buyer, booking a service slot, chasing a quiet lead, answering at 11pm in the customer's own language. Salesteq builds AI agents that carry that work end to end, in production, inside the systems dealers already run.

Where we operate

Live in production

Salesteq agents run customer conversations for BMW, MINI, Ford and Hyundai retail in Saudi Arabia and for dealers in Europe. Live at the group behind ~20% of the Saudi car market. 8,000+ customer conversations a month, around the clock, with roughly 70% of the work completed by AI and clean handover to people for the rest.

How we think about people

Not layoffs. Capacity. A five-person team carrying the volume that would otherwise take a fifty-seat operation, with the AI doing the repetitive coverage and your people doing the judgment calls. The platform works with Keyloop, Autoline and your DMS. It does not ask you to replace anything.

Built for enterprise

European infrastructure

Hosted in the EU with GDPR discipline and data-residency options. Every tenant is isolated: your inventory, your customers, your conversations stay yours.

Auditable by design

Every conversation, booking and handover is recorded and reviewable. Your team sees what the AI did and why, and can take over any conversation at any moment.

Works with your stack

Keyloop, Autoline, WhatsApp Business, Genesys, telephony and your CRM. Appointments land in the live DMS diary, not in a parallel tool your team has to reconcile.

Proven on the hard cases

Gulf-dialect Arabic voice calls, mid-sentence language switches, real inventory questions. 17.4% word error rate on real BMW service calls where general-purpose engines score 21.5% and worse.

Research

The research behind the product.

Salesteq is built by a team with PhDs in artificial intelligence and original work in speech recognition, transfer learning and agent architectures. That is why the hardest parts of automotive retail, dialect Arabic voice and reliable long-running agents, are the parts where we measure best.

Speech and Voice AI

Real-time speech recognition and synthesis for dialect-heavy, code-switching conversations. Benchmarked on real service calls, not lab audio.

Agentic Systems

Multi-step autonomous reasoning with error recovery and human handover, engineered for month-long production runs rather than demos.

Human-AI Collaboration

Trust-building agent behavior, oversight patterns, and the operating model that lets a small team supervise a large AI workload.