LinkedIn invited me to try their new lead generation tool. I asked it to look at this blog, and it came up with a number of possible future posts I could use to gather new readers, all generated by AI (artificial intelligence) from some key prompts and audience identifies that I provided.
It is an interesting piece, although, I have altered (added in italics) some additional text. I would be fascinated to hear what readers think.
AI starts here
A dashboard showing applications up fifteen percent across the board is one of the easier things in education data to misread, and can be one of the more expensive in terms of outcomes.
If you consider recruitment to Initial Teacher Training (ITT) as an example. The structural reason that a dashboard can be misleading is that an ITT recruitment dashboard aggregates things that behave nothing like each other. Domestic and international applicants, candidates with and without visa routes, subjects with entirely different seasonal curves, and applications that will convert at wildly different rates all sit inside the same headline number. The dashboard is honest. The aggregation is what misleads.
Three patterns come up repeatedly:
– Treating a total as a common trend, when a surge in one subject or one applicant group is carrying the whole figure
– Reading application volume as recruitment, when the gap between offer, acceptance and enrolment is where subjects like physics may be very different to that in Classics.
– Comparing the current month to the same month last year without checking whether the policy environment, bursary levels or visa rules changed in between
However, the second-order effect is the one that causes real damage. When a headline figure looks healthy in December, the case for intervention weakens, and by the time the subject-level picture is unambiguous in March the levers that would have mattered, bursary announcements, provider allocations, targeted campaigns, may have already closed for the cycle.
Intervention decisions are can risk being made against an aggregated number at exactly the point in the year when disaggregation would tell you, the reader, something much more useful.
The key takeaway: the dashboard rarely gets the arithmetic wrong. However, all too often it might be read as a single story when, in reality, it is holding four or five different stories, and the one that matters is usually the quietest.
AI ends here.
My message that isn’t in the AI piece, is, if designing a dashboard, first know what is its purpose. If you don’t know what it might reveal, check regularly for trends. As an example: when I first started counting headteacher vacancies in the 1980s, I didn’t know that church schools will be more likely to re-advertise their vacancy than non-faith schools. When looking at teacher vacancies; I didn’t know the rhythm for temporary and maternity leave vacancies being advertised, but I had a hunch, and added the field to the database. As a result, I would also say, collect as much data as you can, when constructing a database, as you might not know what it will tell you.
So, thanks to AI for simulating this post, and quickly offering me some thoughts about many posts I had already written. Does this make this a genuine post or an artificial one?
