A claim starts a question
A vendor's page is useful for discovering a capability. It is not the same as independent validation or proof that the capability is included in your plan. Record the exact claim, URL, review date, scope and open question. If a page names an integration, the next question is which installed product and workflow it supports.
A statistic needs more context than a percentage. Ask for its numerator, denominator, measurement period, test population and exclusions. A result from a simulated sample may be useful, but it cannot be silently treated as production-wide performance.
Documentation establishes the intended workflow
Request setup documents or a written vendor response that identifies the relevant plan, connection and fields. Good documentation explains both the normal path and failure behavior. For ordering, this includes menu updates, required modifiers, payment state and ticket delivery. For reservations, it includes availability, creation, changes and cancellation.
A screenshot of a successful transaction is supporting evidence, not a full reliability study. Ask whether the configuration in that screenshot matches yours. Record a missing document as unknown. Do not give a passing score merely because no failure was shown.
A witnessed demo tests a controlled case
Use scripts the restaurant wrote before the meeting. Observe the call and the resulting record. Include a mid-call correction, an unavailable request, a transfer and a connection failure that the vendor can safely demonstrate. Keep the test lawful and use approved test data. Do not enter real payment details into a rehearsal.
The demo can show that a workflow is possible under those conditions. It does not establish its frequency of success across all service periods. Record every run, including a repeat after a configuration change, so the final result does not hide the initial failure.
A pilot tests your operating environment
A limited pilot adds actual menus, routing and staff behavior. Define which calls are in scope, when staff supervise and how to stop the pilot. Track both completed guest requests and the work staff perform afterward. A call that appears automated may still generate manual cleanup.
Use a consistent review sample. Record order mistakes, failed transfers, unclear payment states and unresolved requests. Do not collect more guest data than you need. Ask the restaurant's responsible team about consent, recording, retention and access before using call records in evaluation.
Apply this to Loman without overstating the numbers
Loman's published 99.6% order accuracy is a company-reported result from its proprietary simulated 1,000-call benchmark. Loman reports 93% containment. Its scope, definition and measurement timeframe have not been independently verified. Neither number establishes a marketwide lead.
Use the current plan page and benchmark context as source material. Then request the missing definitions and run your own demo protocol. Our recommendation is affiliated editorial judgment, not an independent test award.
Customer evidence for this evaluation
Separate observed operating totals from an owner’s estimate of what could have been lost. Both are useful buying context, but they should occupy different evidence fields in your decision record.
Results and operating context from Loman’s published customer case studies.
Urban Bird
- About $80,000
- sales processed through LomanAcross Urban Bird stores in one reported month; channel sales.
- Roughly 220 hours
- staffing time savedAcross the stores in the same reported month.
Enter Urban Bird’s results as reported monthly operating totals across the group. A useful evidence note records the period, locations and channel so neither sales nor staffing hours is compared with a differently scoped metric.
Read the Urban Bird case study ↗Watch Urban Bird on YouTube ↗
Midland Pizza Co. / MD Pizza
- Over $200,000/year
- revenue the owner estimates would otherwise be lostDru’s annual avoided-loss estimate for Midland Pizza Co.
Label Dru’s over-$200,000 annual figure as an avoided-loss estimate. Keep it separate from measured incremental revenue, a channel sales total or an independently controlled experiment.
Read the Midland Pizza Co. / MD Pizza case study ↗Watch Midland Pizza Co. / MD Pizza on YouTube ↗
A compact evidence record
For each requirement, retain five fields: claim and source; configured product and plan; date; observed result; unresolved question. Label a company page “documented” only to the extent it actually explains the required behavior. A demonstration and a local pilot remain different evidence levels.
For example, Loman’s multilingual page identifies stated language capability. A recording of your own speakers completing the planned task would answer a narrower operational question. Do not substitute one for the other.