Qual sampling
There are many ways in which to sample for a qual project. Let’s break it into the two key areas that agencies are faced with. Client sample or free find.
Client sample, customers or prospects most commonly, are straight forward (to an extent!) A fixed universe so that is what we work with.
But more often than not, a qual project requires free find sample. There's a number of ways you can find people...
1. ‘old school’ recruiters with ‘black books’…which can come with old school respondent issues, but also gives you years of experience in finding suitable qual respondents
2. Panels…which for me are just a modern version of the the old school black books, but with considerably more quality & reliability issues.
3. Lead generation – a tool we’ve found very effective over recent years using social media but also can be very limited in volume and temperamental in its success.
4. 'Cold calling' – gives you a potentially unlimited audience but is very time consuming and costly.
So which works best?
Like most skilled trades it comes down to ensuring you have the right tools in your toolbox. And not limiting yourself to one! Understanding properly how each tool works and what pros and cons it can deliver will help you find the right audience and critically, ensure you have genuine respondents that can be relied upon to attend and give great input.
So traditionally, when market researchers approach a field agency for sample, we turn to established sources like Experian, Dun & Bradstreet, or other GDPR-compliant data brokers. We purchase a list, start calling, and work through the numbers.
The process is time-consuming, occasionally frustrating, and often expensive- yet it has been the standard for years. Now, however, a wave of AI-driven sample providers is emerging, promising more precise targeting and richer profiling.
Like most skilled trades it comes down to ensuring you have the right tools in your toolbox
The Cost–Benefit Equation of Traditional Lists
Traditional list buying can be relatively cost-effective. For example, with a typical 30:1 contact-to-completion rate, a few hundred records might be enough to recruit 10 participants — a fairly standard number for a qualitative IDI project.
While this keeps sample costs low, the real expense often lies in labour: screening, follow-ups, and the inevitable back-and-forth to find the right respondents.
The AI Sampling Proposition
AI-based sample sourcing offers a new approach: pulling data from diverse online sources such as social media like LinkedIn, Companies House filings, recruitment websites, and other public records. The goal? Highly profiled, tightly matched potential respondents - theoretically reducing wasted hours in outreach, which in turn could bring down CPI’s.
Recently, we explored with one such provider to identify retail and hospitality businesses that use specific EPOS (Electronic Point of Sale) systems. Using traditional cold calling or even panels & black books would mean a lot of respondents would be lost to screeners (there was far more screening than type of EPOS system used!) so could this complicated algorithm of 'data enrichment' leave us with a database of qualifying respondents that we simply need to persuade to take part (albeit that is without doubt the hardest part!)
It promised to use platforms such as Companies House to identify businesses in the right sectors and at the right turnover using published accounting data, LinkedIn & other promotional web content to source the right contact names and details and then also other related data on the internet to narrow down the actual brand of EPOS systems used.
Sounds like a golden goose! It could be but there’s one very large downside at the moment. I’ll come to that shortly.
How AI Could Improve Accuracy
Traditional broker data can be outdated or inaccurate, usually at least 30%. AI, by aggregating and cross-referencing multiple live sources, could improve this dramatically. It might even allow deeper profiling;
For instance, if you need to reach businesses in a certain sector, region and employee size then that is all relatively easy to profile from traditional sources, but what if you need these businesses to also export to certain markets. Like we've just explained above, we’d historically have to spend many hours (increasing CPI’s) screening respondents. In theory, AI could spot signs of exporting from website pages or industry press giving us a starting list of qualified businesses. Or what if your project needs high growth businesses. Here, AI could spot signs of growth from account filings or tracking recent hires from recruitment portals.
I suppose you could argue that this has been possible for years. We can use ‘desk research’ to crawl sites and the web ourselves to try and match up to company sample lists. But that takes time and time increases the CPI….again.
It’s worth noting at this point that we should not ignore potential compliance issues. Certain platforms (LinkedIn for example) prohibits web crawling within its T&C’s. Whether AI providers are fully compliant will require a review of their relevant quality procedures and GDPR documents.
A golden goose of qual sampling & recruitment
The Reality Check
So why aren’t we just generating all this perfect sample by AI then? And is it any good?
To answer the latter first: We’ve no idea, we’ve not tried it! Why? This answers the former…You’ve got it…cost! In our EPOS case study, the proposed AI sample cost was around £5,000 - £3,500 more than the entire recruitment & incentive budget.
The question will be whether AI can improve response rates enough to justify its premium cost. At those rates it could never be offset by the required labour hours of traditional free find recruitment, but as costs come down there will be questions to answer.
A word of caution
The approaches we are talking about here still relies very much on a human recruiter/interivewer to 'seal the deal'. There is still the huge hurdle of persuassion. Be very wary of any "AI verified" and automated respondents...in both quant and qual.
Looking Ahead
As AI evolves and more providers enter the market, we expect costs to fall. We may also see more DIY approaches, where researchers build or enrich datasets using general AI/LLM tools like ChatGPT or Copilot. We’ve experimented with this in small-scale trials and it seems encouraging - if still limited- results.
Over to You
I’m no AI expert so these are simply the musings of someone who’s been doing qual and quant fieldwork for over a quarter of a century. But we’d love to hear from you. Have you used any of the new AI-based data providers in your projects? Did you find them more accurate? How did you justify the budget- and was the investment worthwhile?
We’d love to hear about your experiences. Drop us a line on this or anything fieldwork related!
.svg)
