← Resources
Conversations / Guide

Chatbot data collection: design a campaign that produces usable evidence

Work backwards from a decision to the questions, records and follow-up your team needs.

A successful campaign produces more than a response rate. It gives the team information that is relevant, interpretable and connected to a decision. Start by deciding what the team needs to know and what a usable answer would contain.

01

Define the population and the output

Specify who should participate, which decision the campaign supports and the period being described. A competency assessment, a time-allocation exercise and a future-demand conversation need different questions and different evidence.

Agree the output before launching: a profile update, an assessment perspective, a demand record or a theme for analysis. Define how distinct people and responses will be counted. A person may supply several competency records without becoming several participants.

02

Test questions with the people who will answer

Use language that fits the work and the local context. Ask for a specific example where it is needed, and give participants a way to say that they do not know or cannot provide the requested information. Forced answers can make a dataset look complete while making it less useful.

Test the conversation with a small representative group. Check whether participants understand the question in the intended way and whether the responses can actually be used in the planned analysis. Resolve ambiguities before expanding the campaign.

03

Keep the record connected to its source

Preserve who supplied the response, when it was collected and which question or framework version it answered. Distinguish reported capability from manager assessment and independently checked evidence. Make missing information visible.

Access follows the client's chosen use case and role-based permissions. Explain that configuration to participants. Client teams can work with EVA's Customer Success and Business Analysis colleagues to shape campaigns and use administrative SuperBot capabilities within their authorised scope.

04

Review coverage before producing conclusions

Check invited people, responses, usable records and unresolved items separately. If one group has low participation, the resulting comparison may need a narrower interpretation or a follow-up campaign. A higher response rate alone does not establish more accurate proficiency.

The next action might be a targeted clarification, an analysis, a matching review or a planning discussion. Give it an owner and decide when the information becomes too old for that purpose. Refresh cadence is part of the commissioned programme.

Explore the example

Watch one answer become an evidence record

Illustrative question: describe a recent situation where you coordinated work across teams.

What we know

‘I manage stakeholders every week.’

How to read it

The answer names an activity but does not describe the person's contribution or a specific situation.

The next useful action

Ask for a recent example, the teams involved and what the person did.

Compare all 3 cases
All example cases and next actions
CaseKnown informationNext action
Initial answer‘I manage stakeholders every week.’Ask for a recent example, the teams involved and what the person did.
Clarified answer‘I coordinated two country teams during a regional launch.’Attach the response and date to the agreed competency record.
Analysis inputOne person · one dated example · one assessment perspectiveCompare the evidence with the agreed framework and decide what needs review.

Illustrative example. These controls explain the method; they do not access client records or execute workforce actions.

Put it into practice

A clear question, a traceable answer and an explicit next step make a campaign useful.

Explore the relevant EVA capabilities

Sources & further reading

This guide explains EVA's approach and illustrative methods. Follow the current offering and methodology pages for scope and examples.