FIELD WORK / JOSHUA AJIEH
Selected stories about finding the real constraint, testing the decision and turning evidence into measurable outcomes.
SELECTED CASE STUDIES
Each story follows the same discipline: understand the problem, examine the evidence, make the trade-off and learn from the outcome.
FIELD WORK / 01
A voice AI product worked technically, but users waited long enough for the experience to feel broken. The long-term build would take six months. Users could not wait.
FIELD WORK / 02
Low retention looked like a feature problem. The evidence pointed to a more fundamental question: which users had a recurring reason to return?
FIELD WORK / 03
The challenge was not collecting more information. It was turning thousands of cases into interpretable signals that could support earlier action.
* The retention definitions and cohort windows should be normalised before this is treated as a like-for-like comparison.
CASE STUDY / 01
A technically working product can still feel broken.
A spoken user turn passed through speech-to-text, LLM inference, text-to-speech and audio playback. Together, those stages created waits of up to 35 seconds. Support tickets, discovery interviews and observed behaviour showed that users were frustrated, and some abandoned the experience.
The useful question was not simply “How do we make the AI faster?” We decomposed the pipeline to understand where time accumulated. Streaming was excluded because the therapeutic experience required the model to reason over a complete user turn before responding.
Building the proprietary voice capability to the required standard was estimated to take at least six months. I recommended integrating ElevenLabs for text-to-speech as an interim infrastructure decision while preserving the longer-term internal voice strategy.
What I learned Don't optimise architecture before identifying the constraint users actually feel. A build-versus-buy decision can buy speed today without surrendering tomorrow's differentiation.
CASE STUDY / 02
More features were not going to answer the question.
Retention was approximately 2.5% on the measure being tracked. Onboarding, activation, notifications, acquisition quality and new features were all plausible explanations. Each could have generated work without establishing why users left.
I examined the journey from acquisition through first conversation and return behaviour, then paired the quantitative funnel with interviews across retained and non-retained users. The comparison helped separate temporary interest from recurring value.
The evidence suggested that the ICP was too broad and the recurring problem was not consistently painful across that audience. We redirected the work toward the strongest problem-user combination, clearer positioning and a shared product-and-marketing hypothesis.
* Cohort windows and retention definitions must be made comparable before presenting this as a direct causal result.
What I learned Retention is a symptom before it is a metric to optimise. The stronger question was not “How do we make more users stay?” but “Who already has a reason to stay, and what does that teach us?”
CASE STUDY / 03
Data becomes useful when it changes what people do next.
Across more than 2,000 active mortgage cases, customer characteristics, exposure and repayment behaviour created many possible signals. Case-by-case review remained useful, but it was difficult to apply consistently at portfolio scale.
I analysed historical and active case data to identify observable patterns associated with elevated default risk. The aim was to turn raw information into signals that could prioritise review before poor outcomes became obvious.
I helped shape risk-scoring and signal frameworks that segmented cases by relative risk. In a regulated environment, the outputs also needed to remain interpretable, operationally useful and explainable to the people acting on them.
What I learned Don't collect more data because you can. Identify the uncertainty that matters, then gather evidence capable of changing the decision.
MORE THAN THREE STORIES
There are many more case studies from working with founders and product teams across the USA, Canada, the UK, the UAE and Australia, spanning different industries and stages of growth. Listing every engagement here would hide the thinking that matters.
If you're still curious—or facing a decision that resembles one of these stories—contacting me directly can lead to a much more useful conversation.