There’s a version of the AI-in-research story that goes like this: technology arrives, promises to replace human research expertise, and delivers varying degrees of success. In this story, research practitioners feel worried about their relevance as technology seems to out-perform them on speed and quality.
Canadian Viewpoint Inc. (CVI), a data collection and fieldwork company, is taking a more practical approach.
CVI’s programming and tech teams have been building AI into their workflows in ways that are genuinely useful; not flashy, not experimental for its own sake, but practical integrations that make their work faster, sharper, and more reliable. And throughout all of it, the throughline is consistent: AI handles the repetitive while humans handle the judgment calls.
AI Applied to Data Quality
CVI implemented tools such as the physical cheque mailed to panelists on an intermittent basis and tracked cheque cashing to monitor data quality. Even with these measures, they decided to add additional checks in-survey. A human can still be a poor-quality respondent if they continually fly through a 15-minute survey in four minutes, or answer attention-checking questions incorrectly.
CVI’s tech team built a layered response to this. It starts with digital fingerprinting to catch bots and duplicate respondents. It adds an AI-powered review of open-ended responses, checking whether answers actually fit the questions and whether respondents engaged meaningfully. Red herring questions, or short-answer items designed to catch inattentive respondents, get verified by AI as well. And all of that gets combined with each respondent’s history: in the last six months, what percentage of their responses have been flagged as problematic?
Those signals feed into an algorithm that tells human reviewers which rows deserve the most scrutiny. The humans make the final call on what stays and what to replace.
As Ben Hilleli, CVI’s Tech Lead, explains it: “By combining multiple different tools and angles of attack, we can scrutinize data in a fast and comprehensive way that would be infeasible without the technology.” The cost and time required to do this manually, simply wouldn’t work. But the answer isn’t full automation, either. It’s narrowing the field with AI so humans can apply judgement where it actually matters.
That’s a meaningful distinction. The AI doesn’t decide who’s a good respondent. It applies rules to flag who’s worth a closer look.
Getting More from Open Ends via Conversational Probing
If you’ve ever analyzed open-ended survey responses, you know the frustration. Some respondents write paragraphs. Most people write three words – if you’re lucky. And the three-word responders aren’t necessarily less thoughtful; they just didn’t have anything prompting them to go deeper.
CVI’s conversational probing tool changes that dynamic. Bryon Seale, CVI’s programming lead, describes the process simply. “[It] elicits improved open-end feedback from respondents by prompting users with AI-driven, real-time, follow-up questions that look at the question subject, review the respondent’s typed responses, and continues to probe them to go deeper.”
CVI tested this in a split-design Halloween study last October. Half of respondents answered open-ended questions without AI probing; the other half received real-time follow-up prompts. The overall number of thoughtful responses was roughly similar across both groups, but among those who did engage thoughtfully, the AI-probed respondents consistently gave more detailed answers with more considered responses.
AI Doesn’t Equal More Engaged Respondents
It’s worth noting what that finding means: AI probing doesn’t transform disengaged respondents into engaged ones. But for respondents who are willing to share, it gives them a structure to share more. That meaningfully deepens data that was already good quality.
And while most industry conversations from insights tech leaders focus on researchers being reluctant to adopt AI-led conversational probing due to fear of being replaced, Ben has had a different experience. As he puts it, researchers sometimes hesitate to use new question approaches because they’re worried about introducing methodology bias. They worry that how you ask a question will change the question scores, not just what you ask. CVI’s approach to testing conversational probing in a controlled split design is exactly the kind of rigor that addresses these concerns by letting clients test new AI-enabled approaches without losing comparability to their existing norms.
Speed Gains by AI-Assisted Survey Programming
AI is being applied beyond data quality and more detailed open-ended responses. CVI’s programming team has also found it helpful for their day-to-day workflow.
Survey programming requires constant custom coding in jQuery, CSS, JavaScript, or Python to handle specific setup requirements that the platform doesn’t cover out of the box. Before AI, the process meant finding examples from other coders by searching for Stack Overflow and other developer forums. Bryon estimates that AI has made this process roughly four times faster.
Four times. That’s not an incremental improvement. That’s a fundamental change in how much the team can get done and a significant reduction of time and cost for implementing customizations.
AI Beyond the Coding Support
Beyond coding support, CVI has been experimenting with training models to take survey design documents and translate them directly into Decipher platform programming. The goal is speedier programming without sacrificing the review and quality-checking that CVI builds into every project. That work is still in development, but it’s a window into how CVI is looking at applying AI to their work.
And then there’s the questionnaire scrutinizer, an AI assistant that reviews incoming questionnaires before the kickoff call, flagging issues that might otherwise get missed. Mobile responsiveness problems. Skip logic that doesn’t account for certain responses. Instructions that aren’t clear enough for interactive question types. Ben describes clients responding well when CVI raises issues that were not on their radar because it signals that CVI is paying attention in ways that go beyond the letter of the design document.
The Human Judgment You Can’t Train Away
Here’s where both Bryon and Ben land, independently, when you ask them about AI and the future of their work: human judgment isn’t going anywhere.
Bryon frames it this way, “Only a human testing a survey can understand the human experience that individuals will have while taking the survey. AI is great at following rules and directions, but it can’t be a human or feel what a human would when interacting with a survey.”
Ben’s version of the same idea comes from a different angle. He’s not worried about people being replaced by AI. He’s focused on what people are actually good at and what they’re not. “I don’t believe that most people enjoy or excel at repetitive, mindless tasks,” he says. For CVI’s tech team, the goal is getting AI to take on the work that humans aren’t great at and don’t love, freeing up the team to focus on the higher-level creative and analytical work that requires human intelligence. “Where AI really shines is when it enables people to do what they already do better, and in a more enjoyable way.”
That’s not a defense of the status quo. It’s a blueprint for a more balanced way to work with AI.
Innovation That Doesn’t Sacrifice What Matters
What makes CVI’s AI story interesting isn’t any single tool. It’s the pattern across all of them.
Every integration they’ve built keeps humans in the loop at the moments that count: final decisions on data quality; judgment calls on survey design; the lived experience of taking a survey. AI handles the volume, the repetition, or the first pass. Humans handle the parts that require context, experience, and the ability to understand how a real person will feel clicking through a questionnaire on their phone.
That balance is harder to achieve than it sounds. A lot of organizations either automate everything, and risk losing the human touch or hesitate to automate anything and fall behind. CVI has done the harder work of figuring out where automation makes the most sense and building accordingly.
For clients, this translates into cleaner data, richer open-ended responses, faster programming turnaround, and surveys that get multiple reviews even before kick-off. The AI isn’t the point. Better partnership is. The AI just helps get there.
