AVESTARA

Research and measurement · Ontario, Canada

Know what your feedback can support before you decide.

Avestara helps public-service and tourism organizations judge what their customer feedback can and cannot establish about a decision they have to make.

Our first service is a feedback audit. AI does the heavy reading. Senior researchers check it, sign it, and tell you plainly what the evidence supports.

  • AI-assisted analysis
  • Researcher reviewed
  • Every step on record
Contact us

What we do

Most organizations have plenty of feedback. Few can say how far it can be trusted.

Complaints, survey comments and reviews arrive every day. They are useful, but they are not a fair sample. Some customers write often. Some never write at all. A theme that appears a lot is not always the thing that matters most.

We read that feedback as measurement. Before any analysis, we agree the decision and what exactly is being estimated. You receive estimates with their limits stated, labels that have been independently checked, and a plain account of what the evidence supports, what it does not, and what would change the answer.

Who the feedback speaks for

Customers who wrote inEveryone else you serve

Illustration, not data. Feedback comes from some customers and not others. An audit states which group each figure describes.

Who it is for

Organizations that collect feedback and need to act on it.

Public services

Teams responsible for customer service in municipalities, agencies and other public bodies.

Feedback you may hold
Complaint and contact-centre records, service requests, open-ended survey comments.
Decisions it can inform
Whether a complaint theme is a real pattern. Whether a service change shows up in what people report. Whether the evidence is strong enough to take to a committee.

Tourism and visitor experience

Destinations, attractions, parks and associations who look after the visitor experience.

Feedback you may hold
Online reviews, visitor surveys, comment cards, enquiry and complaint emails.
Decisions it can inform
Whether a visitor concern is widespread or voiced by a few. Whether reviews and surveys tell the same story. Whether there is enough evidence to support a funding or planning case.

An audit is a good fit when

  • One person owns the decision and is able to act on it.
  • The decision has a date.
  • The feedback already exists, and you have permission to share it.

It is probably not a fit when

  • There is no decision in view yet.
  • You need a live dashboard or ongoing monitoring.
  • The feedback has not been collected, or cannot be shared.

How the audit works

One decision and one body of feedback, in four steps.

The scope is agreed in writing at the start: one decision, one source of feedback and one language, unless we agree otherwise. The tags show who does the work at each step.

  1. Agree the decision

    We write down the decision, who owns it, when it is due and the options being considered. We also agree exactly what is being estimated. This happens before any analysis.

    You bring
    The decision, its date and the options on the table.
    You get
    A short written scope that both sides sign.
    YouResearcher

  2. Check the data

    We review whether your feedback can answer the question. We look at what is missing, what is duplicated and who the records can speak for. Sometimes the data cannot answer the question. If so, we explain why in a short memo and stop there.

    You bring
    One export of feedback records under a signed data agreement. Volume figures too, where you have them.
    You get
    A clear answer: go ahead, or a memo explaining why not.
    Researcher

  3. Code and measure

    We remove personal details and fix a codebook. Researchers code reference samples by hand. AI-assisted coding is tested against those samples before it is used on the full set. Then we estimate, with the assumptions written down.

    You bring
    The categories that matter most to your decision.
    You get
    Every record coded to the same fixed scheme, with each code linked to the words it rests on.
    AI-assistedResearcher

  4. Report what holds

    We report what the evidence supports, what it does not, and what would change the answer. We go through the findings with you in a meeting and check back after the decision is made.

    You bring
    Your questions, and one round of comments on the draft.
    You get
    A decision report, the coded data, the codebook and the Evidence and Decision Statement.
    ResearcherYou

What an audit does not include

We do not build dashboards or collect new data. We do not claim cause and effect from a single export, rank what to fix first, or report findings about named individuals.

Why the results can be trusted

Every claim is checked, and every check is written down.

Measurement validity

We check that each category measures what it is meant to measure, and we state which customers or records each figure describes.

Reliability checks

Two researchers code the reference samples independently. Disagreements are resolved and agreement is reported for each category.

Human review of AI-assisted work

AI helps with coding at scale. A senior researcher frames the question, accepts each category, interprets the results and signs the report.

A logged record

Each AI-assisted step is logged and reviewed. Every figure in the report traces back to that record, and headline figures are recomputed a second way.

Evidence and Decision Statement

Delivered with every audit · Outline
Consistency
How consistently the feedback was coded.
Validity
Whether the codes measure what they were meant to measure.
Coverage
Which customers or records each estimate describes.
Decision support
Which decisions the evidence can support, and which it cannot.
Still open
The uncertainty that remains, and what information would change the answer.

The audit record

Example layout
  • Decision agreed in writing✓ signed
  • Personal details removed✓ checked
  • Codebook fixed✓ reviewed
  • Reference samples hand-coded✓ two coders
  • AI-assisted coding tested✓ reviewed
  • Report figures recomputed✓ matched

“The evidence is not sufficient” is an allowed finding. We would rather say so than overstate.