An origin story, from Sian Rinaldi. February 2026.

The Deficit-First Framework

Everyone wants to see AI implemented. For the most part, there is no clear direction on where the efforts should be concentrated.

At least for the APS, one of the metrics is efficiency targets, but what efficiency looks like for me is very different than what it looks like for someone else in my team, or in another Agency. There is no directive that says that everyone’s AI journey needs to look the same, and I’ll go a step further: it shouldn’t look the same across the board.

Where this came from

Six weeks of explore, try, fail, play, grow

Due to some fortuitous timing, at the end of April, I had the good(!) fortune to have my role made redundant. What I didn’t quite grasp at the time was I was given the gift of runway at the most exciting moment in recent AI times. In fact, the week I found myself fun-employed, Anthropic’s Claude Cowork launched. I didn’t know it at the time, but the next six weeks accelerated so many of my ideas and half formed concepts into reality, and I’ve never been more excited and fulfilled.

Yes, I’m one of those. I know.

The underlying thread of my career has been ‘making change happen’, particularly launching IT initiatives in government. I’ve worn different hats, but ultimately, IT is a great leveller in that no one likes feeling silly with new tech. And AI is no different. In fact it’s almost exactly the same as the challenges we had launching SharePoint for the last decade. We say “I don’t know how to use it”. In response we all get put through mandatory training and alternative savings options are turned off and we are forced to learn because there is no other alternative. But don’t worry, we were met where we were….

The issue is rarely the tech, or even the change management process, or the training, or the communications. The issue is that the majority of people actually want someone or something to sit with them and contextualise how this tech needs to sit within their workflow and their team file structures etc.

And for the most part, SharePoint got over the line (because no other option) and because the file structure can be replicated in an SOP for each team (eventually). This is where AI is fundamentally different, and is far more esoteric. And therefore challenging, because how do I implement something for efficiency when it takes 30 times longer for me to do something and then what I get is actually not helpful and I have to recreate the work anyway?!

The last six weeks (or six weeks post the start of March, whichever makes most sense) has been the most wonderful time and space to explore, try, fail, play, fail, troubleshoot, win, fail and most importantly grow. Along the way, I have helped friends, family, colleagues, random people I have met, to also understand how this new technology can be embraced (perchance enjoyed?!).

Where I have had greatest success is where I have helped an individual, not across a team and definitely not at the organisational level. At the individual level, we have identified the thing they really don’t like, has no value, and is deeply inefficient with no real value for a human to undertake the task. And when we broke that task down, and looked at ways that this could be automated or the process could be enhanced with AI, that is where the true value started to emerge.

I also should clarify, prior to the launch of Cowork, I had had a steep uphill learning curve, trying to wrap my head around why I had really hit a point of inefficacy in what I was able to create with generative AI, and couldn’t reconcile how others were getting these incredible outcomes. Perhaps it was an insatiable need to ‘know’, or a slightly unhealthy competitive streak, but ultimately, I have spent much of the last 12 months understanding how to build tools and workflows for the kinds of problems I deal with in consulting and reform delivery. I built my own GPTs, and I started learning how I could automate, and I was so excited to see, I actually started getting great results. The big thing that I was doing, that not many people I was speaking to were, was starting from the problem (what was I trying to solve) and then thinking about the entire system that sits around it, before even thinking about the tech.

Through that work, I’ve started to notice something. When AI actually becomes part of how someone works, and not just something they experiment with, four conditions are usually present. When it doesn’t stick, one of them is usually missing.

I’ve called it the Deficit-First Framework, and it’s holding up through many use cases. I think it’s one of the better approaches in terms of something that can scale. To clarify, I mean having people engage with the tools usefully for them, not everyone doing the same thing. And most excitingly, the research I’ve found so far supports this approach. I have outlined the framework below, and I would love to hear from people and their experiences, and if this is marrying up to their expectations.

sian@theunordinary.co

The framework

Deficit, Design, Intent, Data

The people I’ve helped get AI working share a set of conditions. They started with a genuine problem, not a technology. The solution was designed into how they actually work, not layered on top. The behaviour change was intentional, not hoped for. And the data feeding the AI was structured and ready.

01 · Deficit

Deficit

A real friction point, a capability gap, or a task someone avoids because it’s not their natural strength (something your brain hisses at). Every role has them, and they’re personal. This is the reason anyone bothers to change how they work.

02 · Design

Design

The AI is designed into how someone actually works, not layered on top. At the individual level: what is the entire workflow, what are the inputs. Once you have that, design where the AI should sit in the sequence, what the person still does, how the output gets used. At scale: co-design, different entry points for different deficits.

03 · Intent

Intent

To actually change behaviour you need to be intentional to break old habits and push through the annoyance of having to do a new thing. It becomes even harder when the tech doesn’t work immediately, so people need to have a conscious decision that this process now includes an AI step. This is the pillar that locks the value in and compounds capability over time.

04 · Data

Data

I definitely underestimated how important data is to success. I knew having data was important, I did not realise how that data was stored, and presented to the model, made a profound difference to the output. Truly, if you only have a short amount of time to make this work, focus on the structure, format, metadata, provenance, and context of your data set, and check with your LLM how it would prefer to receive the data. I have found that tables and matrices over long documents, and taking the time to consolidate duplication, made significant improvements to the output.

How the four hold together

A constraint set, not a menu

These aren’t a menu of nice-to-haves. All four must be present and strong enough for the task at hand. When they are, AI use becomes easier because you’ll be positively reinforced by better and better output and genuine efficiency gains.

Deficit comes first for a reason. Adoption is behaviour change, and behaviour change needs a driver strong enough to push someone through the friction of learning something new. When the deficit is real and felt, people engage with AI because it helps them do their work better. That’s what I’ve seen sustain it.

If the AI vanished tomorrow, would you fight to keep it? Would you notice it was gone because it had become genuinely necessary to how you get your work done?

Full article. Read the long-form version, with the consultant and auditor examples and references to the research, on Substack: The Deficit-First Framework.

Citation. The Deficit-First Framework (Deficit, Design, Intent, Data), Sian Rinaldi, February 2026.

Licence. Creative Commons Attribution 4.0 (CC BY 4.0). Share and adapt with attribution and change-notice.

Sian Rinaldi Founder, The Unordinary. I help people get AI working in their actual jobs. If this matches what you’re seeing, or doesn’t, I’d like to hear.

sian@theunordinary.co · theunordinary.co