The Grunt Work Was Never the Job: What AI Should and Shouldn’t Touch in Analysis
By Glaut
- article
- AI
- Qualitative Research
- Agile Qualitative Research
- Agile Quantitative Research
- Qual-Quant Hybrid
- AI Agents
Summarise with AI
Ask a researcher what they are paid for, and very few will say “moving files between platforms”.
They are paid for judgement. For understanding what a client is really asking. For separating the signal from the noise. For knowing when a clean-looking result is an artefact of how the sample was cut. For turning a pile of evidence into a recommendation that someone can act on with confidence.
That is the work. Everything else is operational noise around the work.
This article covers part of the demo “Glaut Intelligence: Compress Analysis from Weeks to Hours, in One Platform”, which was part of Demo Days Research & Analytics Tools held in June. See Glaut Intelligence in action in the link below.
Glaut Demo: Glaut Intelligence: Compress Analysis from Weeks to Hours, in One Platform
When the Work Around the Thinking Takes Over
The trouble is that the work around the thinking has quietly taken over the hours when the thinking is supposed to happen.
After fieldwork closes, a researcher’s days fill up with tasks like these:
- Moving exports between tools.
- Cleaning data.
- Rebuilding crosstabs every time a filter changes.
- Coding open-ended responses.
- Pulling verbatim.
- Turning numbers into charts.
- Updating slides after a late stakeholder question.
- Rechecking whether a claim still holds after the sample is split differently.
None of this is optional. But at the same time, none of it is the reason the client hired a researcher rather than a spreadsheet. It’s necessary work that crowds out valuable work, and it tends to be heaviest exactly when the deadline is closest and clear thinking matters most.
The Wrong Question Is “How Much Can We Automate?”
When AI enters the picture, the instinct is to ask how much of the process it can take over. Framed that way, the conversation slides toward full automation, and researchers start wondering whether the goal is to remove them entirely from the loop.
But that’s the wrong question. The better one is the opposite: what must stay human?
When you frame it this way, the distinction becomes clearer. The bulk of the work is mechanical and rule-based, precisely what a system can handle efficiently and reliably. The real challenge lies in the judgement, which bears your name.
Draw the Line and Make It Visible
A useful division of labour in analysis looks something like this.
The agent should handle the operational layer:
- Structuring the analysis plan from an agreed brief.
- Coding open-ended responses.
- Generating tables and re-cutting them on demand.
- Surfacing patterns across the dataset.
- Testing whether a claim holds when the sample is split differently.
- Drafting a working version of the report.
The researcher should own the judgement layer:
- Deciding what actually counts as a finding.
- Applying context to the data is not visible.
- Challenging an interpretation that looks neat but isn’t true.
- Shaping the narrative that the client will act on.
The most important word there is visibility. The line must be drawn clearly, or the division of work does not protect the researcher.
If a system quietly makes judgement calls inside a black box, then it hasn’t taken the grunt work off your plate. It has taken the judgement off your plate without telling you, and you are still the one who has to defend the result to your client.
So the line between agent and researcher has to be inspectable.
You should be able to see what was coded, then recode it manually or ask the platform to do it in a specific way. See which responses sit behind a pattern and challenge the framing. See the draft the system produced and rewrite the part where it missed the point. The platform does the assembling, you stay responsible for the meaning.
Introducing Glaut Intelligence
This is the principle behind Glaut Intelligence, the latest agentic analysis suite. It’s not an attempt to remove the researcher from analysis, but rather to give the researcher their time back by compressing the operational work and leaving the judgement where it belongs.
Glaut Intelligence structures the plan, codes the open ends, builds the tables, surfaces the patterns, tests the hypotheses, and drafts the report. Then it gets out of the way. The researcher examines the evidence, considers context, determines what is real, and writes the recommendation. Every step the agent takes is open to review, so nothing is decided on your behalf without your sign-off.
The result is no less human research. It’s research in which the human part, what clients actually pay for, is protected rather than squeezed to the margins.
The best researchers were never valuable because they could copy tables into slides faster than anyone else. They were, and are, valuable for what they do once the tables are built. Better tooling should give them more of their day back so they can do more of that work.
Glaut Intelligence is now available to research teams who want to test a workflow that takes on the grunt work and leaves the judgement to you.
Request one month of free access to Glaut Intelligence here.
This article covers part of the The Founders & Leaders Series podcast episode 2. Listen to the full episode here:
This article covers part of the demo “Glaut Intelligence: Compress Analysis from Weeks to Hours, in One Platform”, which was part of Demo Days Research & Analytics Tools held in June. See Glaut Intelligence in action in the link below.
