Choosing the right combination

There's no single winning combination: it depends on your sector, call volume, and brand positioning. Below are some typical profiles to help you choose.

Typical profiles

• B&B or small business, standard voice + standard AI. Low volume, predictable requests (hours, check-in, room availability). Low per-minute cost, more than adequate quality for that kind of interaction.
• Mid-range hotel, premium voice + standard AI. A business hotel or 3-star with mixed clientele benefits from a higher-quality voice (brand impression) paired with a standard AI model, sufficient to handle bookings and common requests.
• Luxury hotel, premium voice + advanced AI. The phone experience must mirror the stay experience. Premium voice for the welcome, advanced AI model to handle articulated requests (packages, special needs, different languages).
• High-volume restaurant, standard voice + advanced AI. Fast bookings but with variables (party size, allergies, specific tables). A standard voice keeps costs down; an advanced AI prevents booking errors.
• Multi-property chain, multi-language premium voice + reasoning AI. When the assistant must identify the correct property from the call, manage different rules for each location, and speak with international customers, top-tier models are an investment that pays off.

Optimising cost without losing quality

• Refine instructions. A well-instructed standard model often outperforms an advanced model with generic instructions. Invest time in the Instructions section before moving up a tier.
• Reduce call duration. Shorter sentences, avoiding unnecessary repetition, getting to the point: the AI speaks less and costs less. The agent's instructions can guide a leaner style.
• Use targeted criteria and actions. Have the assistant handle only requests it's good at, and transfer the others immediately, to avoid unnecessarily long conversations.
• Differentiate inbound and outbound. Some businesses need a top-tier voice/AI for inbound calls (customer impact) but can use more economical models for outbound reminder calls.

Monitoring your choice over time

After choosing the combination, monitor regularly: the percentage of calls handled autonomously, the average duration, the outcomes recorded by evaluation criteria, and customer feedback. If you see things getting worse, the cause is rarely the model: more often it's a seasonal change, an active promotion, or an outdated instruction. Check the context first, then possibly the model.