Growth consultants are expensive. Their value is real but it lives almost entirely in their heads — pattern recognition built over decades, applied to client situations through instinct, experience, and frameworks that are never fully documented. The question was whether that intuition could be systematised. Captured, trained, and deployed at a fraction of the cost.

It was a genuinely interesting question. Interesting enough to take seriously, which meant working with people who actually knew what they were talking about — university AI research units, business school academics, and psychology departments who studied how experts make decisions.

Spent significant time across those university departments mapping the problem properly. The AI researchers were honest about what the models could and could not do. The psychologists helped deconstruct what consultant intuition actually consisted of at a cognitive level. The business academics understood the methodology being encoded.

The conclusion, reached collaboratively and without drama, was that replacing a consultant's full intuitive process was not a near-term possibility without an investment that no business case could justify. The complexity was real. The cutting-edge research was genuinely cutting edge — meaning nobody had solved it yet, and the cost of being the first to do so was prohibitive. The grant application that might have changed that equation was unsuccessful.

So the scope was reduced. Significantly.

Rather than replacing consultant intuition, built a Small Language Model focused on a narrower, more tractable problem — organisational diagnostics. Not replacing the consultant. Not replicating their full judgement. Just compressing the diagnostic phase — the structured information gathering, pattern identification, and initial hypothesis generation — into something faster and more consistent than doing it manually every time.

It saves weeks. It does not replace humans. It was never going to, and being honest about that earlier would have saved some time. The research conversations with the university units were valuable regardless — they gave a clearer picture of where AI genuinely is versus where the marketing says it is, which turns out to be a useful thing to know.

The data existed. Enormous amounts of it. But it was unorganised, inconsistent, and sitting in a lake that nobody had properly mapped. The ambition was a live BI dashboard that updated in real time and surfaced the insights that previously required a analyst to go looking for them. The problem was everything that had to happen before the machine learning could start — understanding the data, cleaning it, orchestrating it, and building the pipeline that would make it usable at all.

This was before the tooling existed to make this straightforward. AWS Lambda was the compute layer. Everything else had to be figured out from first principles, at a time when most organisations were still debating whether real-time BI was even worth attempting.

Started with the data itself — no assumptions, no shortcuts. Mapped the lake, understood what was actually in it, and built the orchestration layer that turned unstructured data into something a model could be trained on. Once the pipeline was stable, trained the algorithms that powered the automation — moving from static reporting to live dashboards that surfaced patterns and anomalies without a human having to go looking for them. The machine learning was not the hard part. The data engineering that made it possible was.

A working real-time BI platform powered by ML automation, built at a time when most SaaS businesses were still running on static reports and manual analysis. The dashboard surfaced the kind of insight that had previously required significant analyst time — automatically, continuously, and in real time.

A major real estate group was consolidating its enterprise project portfolio into a dedicated subsidiary — a significant structural shift that required not just moving work across, but changing how that work was done. The existing processes were fragmented, inconsistent, and built around individual project heroics rather than repeatable systems. BIM adoption was patchy at best. The people doing the work understood construction and property deeply. They did not necessarily understand why their processes needed to change, or what a modernised operating model would actually look like in practice.

Designed and embedded the delivery processes and standards that would make the subsidiary function as a coherent operating unit rather than a collection of projects. Drove BIM adoption as the backbone of how projects were planned, coordinated, and delivered — establishing the digital thread that connected design intent to construction reality. Worked with teams who had been doing things a certain way for a long time, which required as much change management as it did process design. There were strong opinions in the room, held by people with the seniority to act on them. Some of those conversations required patience and careful positioning. The work got done.

Enterprise project processes consolidated and standardised across the subsidiary. BIM adoption embedded as standard practice. Delivery consistency improved across a portfolio that had previously operated as independent silos.

Fraud detection at scale is a product problem as much as a technical one. The existing system had high false positive rates — creating operational overhead, damaging customer experience, and undermining trust in the product. The KYC and identity verification platform needed to serve B2C, B2B, and B2G use cases simultaneously, each with different risk tolerances, regulatory requirements, and user expectations.

Worked across product, engineering, and data science to redesign the fraud detection logic with a focus on explainability and operational adoption — not just model accuracy. Built the AI-linked KYC and identity verification platform with clear service boundaries for each use case. Used Amplitude and Looker to drive data-informed prioritisation throughout, ensuring decisions were grounded in evidence rather than intuition.

Reduced false positives materially, cutting manual review overhead and improving customer experience. Delivered a KYC platform serving multiple customer segments with a single underlying architecture. The explainability work — making clear why a decision was made, not just what the decision was — was the detail that made operational adoption stick.

The product worked. The growth did not. Revenue was growing but not at the rate the business needed, and the reasons were structural rather than market-related. Onboarding was leaky. Time-to-value was too long. The path from self-serve to enterprise was undefined. The team was working hard but optimising locally rather than thinking about the full customer journey as a system.

Mapped the entire growth loop — acquisition, onboarding, activation, monetisation, expansion — and identified the highest-leverage intervention points. Built an experimentation framework using A/B testing and cohort analysis to test hypotheses quickly and retire the ones that did not hold. Redesigned the self-serve onboarding flow to reduce time-to-value. Aligned the PLG and sales-led motions so that structured upgrades from self-serve to enterprise contracts had a clear, repeatable path.

150% MRR growth within 11 months. Improved activation and retention metrics across the board. A sales and product motion that could scale without friction.

A major PE transaction was on the table and the product and engineering organisation — 750 people across global distributed teams — needed to be audit-ready. The gaps between how the organisation presented itself and how it actually operated were significant. Architecture was sprawling, delivery metrics were weak, and the operating model had grown organically without design. None of this was unusual. All of it was visible to anyone who looked carefully.

Worked directly with the CTO and regional VPs to conduct an honest assessment of the organisation's real state across product structure, engineering capability, architecture, delivery systems, and governance. Led integration readiness planning, identified and prioritised risk, and drove operational remediation in the areas most likely to surface in due diligence. Simplified the application estate from 130 services down to 45, cut delivery lead time by approximately 40%, and introduced the operating standards and metrics that gave the acquiring party confidence in what they were buying.

The acquisition completed at $2.5bn. The organisation entered the post-acquisition integration phase with a cleaner architecture, a documented operating model, and delivery systems that could withstand scrutiny. The remediation work done pre-acquisition materially reduced integration risk on the other side.

A Tier-1 bank was off track. Major incidents were happening regularly, integration issues were stacking up — 33 in total — and the relationship was deteriorating fast. The account was at serious risk of being lost entirely. The internal response had been reactive, fragmented, and increasingly focused on managing perception rather than solving the underlying problems.

Stopped the noise and started with the truth. Worked through every one of the 33 integration issues systematically, rebuilt the delivery cadence, and changed how the team engaged with stakeholders — from defensive to transparent. Coached the product managers and owners on how to talk to customers about timelines, shifting the narrative from missed release dates to a Now, Next and Future model that gave the bank confidence without creating false expectations. Stabilised the programme, established proper incident follow-up, and rebuilt trust from the ground up.

The relationship was recovered. The bank increased their spend by £3m per year. The account went from churn risk to expansion. The Now, Next and Future approach became standard practice across the wider team — giving product managers room to work without fear, and giving customers genuine confidence for the first time.