From restaurant data to real decisions: what we learned from our first pilots

We launched our first pilots at two Mandraki restaurant locations during busy Saturday services.

The first goal was straightforward: could guests use Zvelto to browse the menu, place an order and receive exactly what they asked for without disrupting the restaurant?

They could.

Guests used the product successfully, orders arrived accurately and the feedback was strongly positive. One member of staff told us that they no longer needed to keep going downstairs to pass on and clarify every order.

That comment stuck with me.

Restaurant technology is often presented through dashboards and feature lists. The value during a live service can be much simpler: fewer unnecessary trips, fewer interruptions and more time focused on guests.

But the pilots also gave us something else.

They gave Zvelto real ordering behaviour to learn from.

During one of the pilot services, Zvelto picked up a pattern: many guests ordering wraps and portions were not adding a drink.

There could be several explanations. Some guests might already have had a drink. Others might not have wanted one. But the ordering journey might also have made the addition too easy to overlook.

Zvelto identified a straightforward action:

Prompt guests to add a drink after selecting a wrap or portion.

The change was written directly into the Zvelto ordering flow. The restaurant team did not need to study a report, update the menu manually or ask staff to remember another upselling instruction.

Based on the restaurant’s typical volume, we estimated that this single change could generate £1,000+ in additional monthly revenue. More than £12,000 a year from one small change.

This is the difference we want Zvelto to create.

The data produces a signal. Zvelto interprets it, recommends a specific action and helps put that action into practice.

Zvelto can then measure how well the change worked, adjust it and apply what it learned to the next decision.

The drinks prompt was only one of the opportunities Zvelto identified.

The same review found several more potential improvements across add-ons, signature dishes, shared items and opportunities for guests to order again.

None required the restaurant to attract more customers, add more tables or create an entirely new menu.

They focused on helping existing guests discover more of what the restaurant already sold.

Individually, each decision was relatively small. Together, after adjusting for overlap between them, we estimated that the actions could generate:

• Around 4% to 10% potential revenue uplift
• £20,000 to £50,000 in additional annual revenue

These are modelled estimates based on early pilot data. They are opportunities to test, not promises.

We are now preparing to run these actions for longer, measure the results and understand which changes genuinely improve the restaurant’s performance.

Experienced operators already know that small decisions can make a large difference.

Move the right dish higher on the menu. Make a popular addition easier to find. Promote something different when demand is quiet. Stop pushing an offer that is reducing margin. Adjust the recommendation when guest behaviour changes.

The problem is rarely a complete lack of ideas.

The problem is finding the right opportunity at the right time, deciding whether it is worth acting on and following it through while running the rest of the business.

That becomes even harder across several locations. Each restaurant can have a different menu mix, customer base and trading pattern.

A useful action for one venue, one day or week, might be irrelevant at another.

Zvelto is being built to do this work continuously, using each restaurant’s own data and operating context.

A system making commercial decisions needs clear boundaries.

An operator may be comfortable allowing Zvelto to change the position of an item automatically, but still want to approve a price change or promotion. Others might want to approve every action or fully automate certain types of decision.

The level of control should reflect each operator’s needs.

For each action, the operator should be able to understand why Zvelto suggested it, see the expected impact and decide whether to approve, adjust or reject it, then review the result.

As confidence grows, the operator can choose which decisions Zvelto is allowed to execute automatically.

Trust should come from measured results, not from asking a restaurant to hand over control on day one.

A recommendation is only useful if the restaurant - and Zvelto - can see what happened afterward.

Did the drinks prompt increase the number of drinks sold? Did it improve revenue without damaging the ordering experience? Did it work better with certain dishes or at particular times? Should the restaurant keep it, change it or remove it?

If an action works, Zvelto learns from it. If it does not, Zvelto can move on quickly and still use the result to make a better decision next time.

Over time, every measured result can make the next decision more relevant to that specific restaurant, opportunity or challenge.

This creates a simple loop:

1. Zvelto identifies an opportunity.
2. It recommends or executes a change within the operator’s boundaries.
3. It measures what happened.
4. The result improves the next decision.

That loop is more important than any individual recommendation.

The early ordering product gave us a way to test Zvelto inside real restaurant operations.

Since then, we have continued building the wider intelligence and execution layer behind it.

Zvelto Advisor, our internal decision engine and the core product modules are now running. We are also working with payment and technology partners so Zvelto can connect with more of the systems restaurants already use.

Zvelto does not require operators to replace their existing technology stack. It is designed to work across the systems they already use, while providing its own modules where they add value.

Over time, it can take responsibility for more decisions as the operator becomes comfortable with the results.

Our first pilots do not prove that every recommendation will work or that every restaurant will achieve the same uplift.

They gave us an early indication that live restaurant data can produce practical actions with meaningful commercial potential.

Now we need to prove that those actions consistently deliver results across more services, menus and venues.

The question remains the same one that led me to start Zvelto:

After looking at all the numbers, do you know what to do next?

We want the answer to become much simpler.

Zvelto should identify the opportunity, explain the decision, make the change and show the operator what happened.