Last week, Pragma went live.
Pragma is an AI change adviser built on twenty years of my practitioner experience. I painstakingly catalogued the frameworks I actually use, the case studies I’ve learned from, the positions I hold on what good change leadership looks like — based on what I write.
A simple premise: give senior leaders and change practitioners honest, practical guidance on complex change — without the wait, the expensive consulting engagement, or the polished slide decks.
Just short of 50 conversations later, here’s the update.
The data that is my new obsession
Forty-nine conversations. 190 messages. Mean 3.9 messages per conversation, median 2.
That gap between mean and median tells you something immediately: bimodal distribution. Conversations either die at two messages….or cross a threshold and go to six or more. Almost nothing lands in between.
The decision to engage ‘properly’ is made in the first exchange. Which mirrors, pretty closely, how trust tends to work in a consulting engagement…the client forms their read in the first meeting, not the third. Pragma just made that pattern visible in a dataset.
Once someone crossed the ‘trust valley’ and they decided to stay, then 65% went to five messages or more.
Very few asked the initial question...they actually needed answered.
In roughly half of all conversations, the topic shifted between the opening message and the final exchange.
Someone arrived with a question about communication strategy…and ended up working through whether they had genuinely committed to the change themselves.
Someone came in asking about resistant stakeholders and left reconsidering whether the programme had a coherent rationale a sceptical stakeholder could reasonably get behind.
Someone framed it as a resourcing problem and found, three exchanges in, that it was a sequencing problem — they were trying to do too many things at once.
The first thing people present is rarely the problem: it’s the door they walk through to find it. Every good executive coach or psychologist knows this, so it’s not surprising that replicates here.
This is, in my experience, what good change management tends to look like. The Planning Fallacy tells us we’re overconfident in our predictions. The Dunning-Kruger effect tells us we’re working from a reference class that’s too narrow.
What Pragma is showing me in practice is the same dynamic at the problem-definition level — playing out in four messages rather than four weeks.
The change practitioner’s job has always been investigative. Pragma is making that visible in the data.
Specificity is the admission ticket.
Pragma offers suggested conversation starters — three stock prompts people can click rather than blank-paging their own question. I’m considering binning them and would welcome your thoughts in the comments.
About half of conversations started via a prompt. The other half came in with their own opener.
Not surprisingly, self-starters got more out of it: conversations that started with an original, user-generated opener reached deep engagement (more than 5 messages) at 45%. Conversations starting with a template prompt reached it at 16%.
That gap is partly by design— a template opener produces a complete answer on turn one. If the loop closes via the query being answered, there’s no natural reason to continue.
But the more striking finding is what happens within the original openers that people bought themselves. Generic ‘how-to’ questions (e.g.: “how do I get better at change?” “how do I build resilience?”) reached deep engagement at zero percent. Not low. Zero.
(Do you know what wasn’t zero in those conversations? My API and tech stack costs...)
Openers that named a specific person or situation — “my head of delivery keeps blocking every initiative I bring to the exec table,” “I have a staff member underperforming and to be honest he just isn’t the right fit” — reached deep engagement at 58%.
Clarity is binary, apparently.
The practical implication extends well beyond Pragma: if you can’t name a person in your change problem, you’re probably not ready to work on it yet. A generic question is a search query. A specific situation is the beginning of a real conversation.
One more detail worth mentioning: if you tried it last week...it’ll be much better this week!
Unsurprisingly, I also stuffed something up: I loaded Pragma’s research base and content map into the model independently rather than connected, which meant early conversations drew on one or the other….but not both fluidly.
Fixed now. I mention it because it’s representative of something true about building anything complex: the integration is always harder than the individual components…which is almost exactly what I tell clients about their change programmes. (And what I write about in the upcoming Pragmatic Change book thanks to the Flyvbjerg doctrine of modularity)
And somewhere in the first 49 conversations, a user asked: “Can I buy Pragma yet?”
Then, a few exchanges later: “No but seriously, can I pay for this?”
Yes, yes you can. More great questions like that here.
We’re at forty-nine. I want a hundred good ones.
If you’re leading a complex change right now or sitting with a problem you haven’t quite been able to name, then try Pragma. It’s free. Bring it a real question, not a hypothetical.