Corporate B2B thought leadership reports rely almost entirely on quantitative surveys as a primary data input. This approach provides a useful sense of where trends are headed, showing whether respondents are veering left, right, up or down on any given topic. That trend data is usually complemented with qualitative interviews that can help provide some context and colour as to why more folks are doing what they’re doing. 

This formula is well-trodden and can deliver decent outcomes. But both quantitative and qualitative research approaches are problematic in various ways. 

In the B2B realm, quantitative surveys take a long time to deploy and cost a lot; more importantly, they are very rigid, relying on closed-ended questions. While surveys often encourage respondents to input some kind of open-ended text to try and explain why they’ve answered in a given way, any researcher worth their salt will quickly tell you how poor those inputs usually are; more often, they’re simply used as one of the ways to weed out poor quality or fraudulent responses. You can safely forget about discovering any interesting nuances or quotable quotes. 

Qualitative research opens the door to real nuance, lively quotes, actual case studies, and more. But you have to compromise on scale. I’ve seen attempts to secure 50+ qualitative interviews (and I send my deepest sympathies to those teams), but unless you’ve got a rare combination of deep pockets, a sizeable contact book, and no deadlines, this is unlikely to work.  

All of which makes the notion of AI-moderated interviews a fascinating new entrant in the field. Spurred on by a wave of venture capital money, a series of start-ups are competing to disrupt the traditional field of market research. For B2B thought leadership, these new options are not a silver bullet, but they do present a compelling third way alternative to the traditional quant and qual approaches.  

Qual at scale: new options, new insights 

A quick side note, before we continue: companies should still start by keeping their eyes open for opportunities to tap into other, non-survey based datasets and inputs. Novel or innovative data inputs are an outstanding way to stand out from the crowd. But of course, this isn’t always feasible. Many topics simply don’t lend themselves to secondary data. For example, a study exploring how banks are responding to new rules around how they communicate with customers isn’t likely to find a lot of useful secondary data to tap into; in instances like that, a survey is probably the best way to get a handle of what companies are doing. 

For these types of issues, the emerging field of AI-moderated surveying provides some compelling options. The actual approach varies widely from one platform to another: at one end of the spectrum, it’s akin to a back-and-forth WhatsApp chat; at the other extreme, it’s a bit like a conversation with HAL from 2001: A space odyssey. Accordingly, formats vary: some discussions are text based, others are audio, and are done via video. This means differing levels of detail are captured, but a common theme is that all of these options collect far more nuance and input than a traditional quantitative survey, but still less than an in-depth one-to-one qualitative interview. (All platforms include options to allow AI to ‘probe’ respondents in more detail on something, depending on how they respond to a defined question.) Here’s an example from a recent survey asking about the impact of AI on the finance function: 

I think one of the biggest concerns is around quality control and measuring the output and making sure that, uh, we're— things are still being reviewed, and there's still human intervention on some of the tasks to make sure that the output's in accordance with GAAP and accounting principles and that there's a quality control check on the output from AI. There's a concern obviously about the headcount as well, and that's more one for the owners of the business. But quality control is, um, probably the biggest concern right now.

Three observations from this: 

• First, the transcript and related video quickly verifies that this is a real respondent, and is speaking from within the finance function. Other answers share details of their accounting systems and approach taken, for example. 
• Second, while a multi-choice list of survey options could have captured “quality” and “risk of job losses” as the twin fears, this quote gives a much more specific insight into what’s actually going on here: quality isn’t just about hallucinations, but is also about ensuring compliance with accounting principles. 
• Third, given this added depth, you can realistically consider working with a smaller sample size, without compromising on depth. Sampling 100 finance execs in this manner will easily provide far more depth than a traditional survey of 500 respondents.  

AI-moderated interviews can often be conducted more quickly: there’s infinite capacity, and the AI is always available. Plus most platforms can switch into whatever language the respondent prefers, while also translating back their responses for you. So the administrative burden and time for setting up and running a survey is lower. Also, many platforms offer the ability to query respondents when they provide a contradictory response later in an interview. 

Limits and considerations 

So far, so good. But before you ditch the alternatives and go all-in, there are some considerations to be aware of. 

• Cost: given the added depth, plus the frequent use of audio and/or video for respondents, the cost per respondent in B2B can be high. In some instances, the investment can be analogous to doing one-to-one qualitative interviews. This comes on top of the costs required for a dedicated platform, which will be additive to any other platforms you’re still wanting to use. 
• Surface-level probing: Don’t expect AI to handle complex, niche, or technical topics as well as a human moderator does (or should). It typically works well on the basics and simple follow-ups, but don’t expect a deep dive into a specific domain that a journalistic interviewer would typically offer. 
• The trap of the average: Most platforms offer AI-driven synthesis of the data, providing a summary of your core findings even while the fieldwork is underway. This is helpful, but these summaries tend to emphasise the most common responses, rather than having a sense of what might be the most interesting or unexpected ideas. 
• Missing non-verbal cues: A respondent might offer a guarded reply on a sensitive point, while smiling or otherwise indicating that they’re not sharing the full story. These gems are usually missed by AI, whereas human interviewing would take that as an ideal chance to explore an editorially interesting tangent. 

All told, AI moderated interviews provide an interesting additional option to add into the mix, alongside your traditional research methods. This is still an emerging field, with a lot of new entrants, so it’ll be fascinating to see how it looks in a couple of years from now.