Why ‘failing fast’ is the best way to get AI to stick in your business

Why ‘failing fast’ is the best way to get AI to stick in your business

Plus: The non-technical founder who automated his commercial thinking

From the aibl team

Every mid-market firm feels the pressure to move on AI and most of us reach for the most visible thing first. Pick a platform, roll it out, measure who’s using it, and it feels like progress. But the organisations that get somewhere tend to start with a specific problem and build a structure around solving it.

Sara Maldon at Make spent weeks in one-on-one conversations with over 160 employees before launching the formal programme. Then she embedded full-time AI specialists inside every department. They call them samurai, which gives you a sense of how seriously Make takes the role.

She set targets high enough that early projects failed on purpose. Teams would overreach and discover the real blocker was broken data or a missing process. Those failures told her more than the successes did. By January 2026, 96% of employees had built AI automations. But that was only possible after months of learning where work was actually falling apart.

Terry, aibl’s co-founder, came at this from a different angle. He was losing 20 to 30 minutes a day to processing voice notes from calls, and most of the time it just didn’t happen. So he built a pipeline on Make.com and Gemini. Terry has no engineering background, but raw voice dumps go in, structured actions and commercial signals come out. He set out to stop losing information. He’s ended up with a searchable archive of thinking that gets more valuable the longer he runs it.

Sarah and Terry both worked backwards from where things were stuck or breaking, and built from there. There’s enough detail in both of these playbooks to save you months of missteps.

Why 96% AI adoption at Make didn’t start with tools or training

When Sara Maldon joined Make two years ago, there was no approved AI tool. Nobody could use ChatGPT or any other model for work – there was nothing to adopt.

Make already had a culture built around automation. As a platform, automation was the product. “Do it once, do it twice, automate it, never do it again” was already how teams operated. But AI didn’t have the same structure around it yet.

Leadership knew AI would be critical but didn’t have a vision for where to take it. Sara was hired before they had a specific problem to solve. Her brief was to find the vision and build it.

Part of the logic was commercial. Make needed to understand what the market would expect from an AI-powered automation platform. Leadership believed the fastest route was to experience it internally first.

Miriwa – a Korean expression meaning “going into the future” – became the programme to make that happen. By the time it launched, organic adoption had already taken hold. People were using tools like Gemini day-to-day, but only 16% had built AI automations and 5% were using agentic automation. Make tackled this as a company-wide priority – aiming for everyone to build with AI and agents.

Most mid-market firms try to get there by approving a tool and tracking logins, but Make’s route looked nothing like that.

Watch the full video interview:

NEWS

AI spend has become routine, but the governance hasn’t caught up

According to Spendesk’s analysis of UK payment data across firms ranging from under 20 to over 250 employees, 72% of organisations now spend on AI tools, up from 61% in 2024. Subscriptions account for 70% of AI purchases, up from 43% in early 2023, turning what used to be one-off experiments into recurring line items.

Until recently, most AI spend was a pilot someone expensed on a card. Now it’s recurring software, and that raises obvious questions around renewal tracking, usage monitoring, and vendor consolidation. The 28% of organisations with no AI spend at all have time to get this right from the start, but the 72% already spending may not have.

The firms on firmer ground aren’t necessarily spending more, they’re just treating AI spend like any other operational software line.

From voice dump to action list

Voice notes from calls, meeting transcripts, half-formed ideas recorded on the move. They contain commercial signals, decisions, and action items, but the format makes them hard to revisit and almost impossible to act on quickly.

aibl co-founder Terry O’Dwyer was spending 20 to 30 minutes processing each one manually. Often it didn’t get done at all.

So he built a pipeline in Notion, Make.com, and an OpenAI Assistant. He records a voice note, pastes the raw transcript into a Notion database through a custom quick-capture mobile interface, and moves on. Make.com triggers automatically and sends the text to an OpenAI Assistant trained on his specific executive profile.

He loaded his Hogan Assessment data, known ADHD traits, and commercial goals into the assistant. The AI processes what he says through the lens of how he actually thinks, his known behavioural patterns, and what he’s trying to achieve commercially.

Product spotlight of the week

Most mid-market AI firms are still stitching spreadsheets to figure out what each customer actually costs to serve. It’s one of the messier scaling problems aibl sees come up again and again. Stripe’s new AI cost tracking tool, previewed this week, takes a direct run at it. It meters inference spend across tokens, models and compute, then auto-applies your markup to customer bills without custom dev.

Stripe is betting usage-based AI billing becomes as standard as subscriptions. For the AI-native firms in aibl’s orbit, that’s probably right. Standard processing rates apply, plus 0.7% on the billing layer. Worth a look if your margins depend on getting the inference math right.

Copyright (C) 2026 aibl. All rights reserved.