When the Research Failed, the Behaviour Didn't

Finding a demand signal at Detect Technologies by triangulating a failed survey against behavioural data

Company

Detect Technologies (B2B industrial AI SaaS)

Role

Software Engineer, functioning as product lead for UX & web experience

What I Owned

Discovery & funnel design

Context

Detect Technologies sells industrial AI products to large industrial buyers. I own the UX and web-experience charter — the corporate website is the top of the B2B lead funnel, so how we segment, speak to, and route buyers on the site directly shapes what reaches sales.

We had already built and shipped a real-time industrial surveillance security product. The question I set out to answer was not should we build this — that was decided — but who is it actually for, and how should we position it?

What I did first, and why it didn't work

I ran a structured buyer survey, built in coordination with the sales team and targeted at an industrial stakeholder list assembled through ZoomInfo (ABM). The intent was primary discovery: get industrial buyers to tell us directly what they needed from a surveillance product and how they evaluated it.

It underperformed badly. We received roughly 10–15 responses — under 1% of the targeted list. That is not a sample you can steer a product or a positioning decision on, and I treated it as such rather than over-reading a handful of replies into a narrative.

This is the part of the project I’d normally be tempted to leave out. I’ve kept it in because what happened next only makes sense because the survey failed.

Triangulating instead of forcing it

With primary research thin, I went to the behavioural data we already had and cross-read several sources rather than trusting any one of them:

  • GA4 — what visitors actually came for, and which product journeys they entered
  • Zoho PageSense — on-page behaviour and conversion patterns
  • Heatmaps — where attention concentrated on product and category pages
  • SEMrush — what demand looked like upstream of our site, in search

Stated intent is what a survey captures. Revealed preference is what this stack captures. When the first was unavailable at any useful volume, the second became the primary instrument.

What the data actually said

Two findings came out of it, at different altitudes.

1. The security traffic wasn’t all one audience. A meaningful share of people arriving on security-related pages were not industrial buyers at all — they were generic security interest: home security, individuals looking for consumer-grade solutions. They looked like demand in the aggregate numbers and behaved nothing like our ICP. Treating them as one pool was quietly distorting how the funnel read.

2. The market was pulling on a different product. Around 17% of total leads were oriented toward the workplace safety product; roughly 3% toward the security product — close to a 6x gap. We were investing in positioning a product that demonstrated demand was not asking for at anything like the same intensity.

The first finding was a segmentation problem I could act on. The second was a strategic signal above my charter.

What I decided, and what I recommended

Within my charter — the website and funnel — I acted on the segmentation. I separated industrial security buyers from generic and consumer security interest, and rebuilt around that split: the content pipeline, the outreach funnels, and the lead-generation journey were designed to qualify industrial intent and stop treating consumer traffic as ICP volume. The deliberate choice here was to not chase the consumer security traffic even though it would have inflated top-of-funnel numbers. Volume that sales can’t sell to isn’t a win; it’s noise that makes the funnel look healthier than it is.

Above my charter, I surfaced the demand signal and made a recommendation: weight our content and positioning investment toward the demonstrated demand — safety — rather than toward the product we had most recently built.

Leadership decided differently. They chose to weight the security content pipeline more heavily, to strengthen positioning for that product. That was their call to make, with commercial and positioning context I did not fully own.

Where it stands

Safety remains Detect’s most sought-after product. The gap I flagged has not closed.

I’m not presenting that as vindication — leadership was optimising for something real, and positioning investment can create demand rather than just follow it. But the signal I identified is still live, and it’s still measurable, which means it’s still actionable.

What I'd do differently

  • I over-invested in one instrument. I designed a survey as the primary method and only reached for behavioural data once it failed. Triangulation should have been the plan, not the recovery.
  • The ABM list was the wrong channel for that ask. A cold industrial stakeholder list has almost no incentive to complete a survey. Fewer, deeper conversations sourced through sales relationships would have beaten a wide cold send — that’s the discovery muscle I’m most deliberately building now.
  • I’d instrument the recommendation. I surfaced the demand gap qualitatively. If I ran it again, I’d attach a proposed test and a success metric to the recommendation — a defined content-weighting experiment with lead-quality as the read — so the decision had a measurable path rather than resting on a directional argument.