TDRA Consulting is being built and isn’t formally taking engagements yet. The services and projects described here reflect completed work and current capability. If something here matches a problem you have, get in touch anyway. Early conversations are welcome.
WorkFive projects
The question, the analysis, and what it changed.
The work behind the thinking: how a business question gets investigated through data, analytics, machine learning and applied AI, from the question to the analysis to what it changed. Each engagement is labelled by what it actually was. Client work and portfolio projects are marked separately, and where a finding is a hypothesis rather than a measured result, it says so.
Client engagementPortfolio projectPortfolio work uses public, public-style or self-collected data and was not deployed for a client.
Showing 5 of 5
No published work matches that combination yet.
01 / 05Client engagement
Astrum Digital
12+Competitors mapped
$40–80Monthly pricing gap vs leaders
3Positioning pillars held up
FinTech go-to-market strategy · via Riipen Level UP · Feb–Mar 2026
Problem
How should an early-stage payments platform differentiate against 12+ established competitors, and which customer segments are most viable?
Approach
Mapped competitive pricing, compliance and settlement positioning across the market; segmented buyers into three types; prototyped onboarding and reconciliation concepts.
Insight
A $40–$80/month pricing gap existed against market leaders, and three positioning pillars held up under scrutiny: compliance automation, real-time settlement, and SMB-focused pricing.
Decision
Which segment to lead with, what to say to it, and which partnerships to pursue first.
Outcome / next step
A sequenced go-to-market plan was delivered. Commercial results of executing it sit with the client and are not reported here.
02 / 05Client engagement
FanRyng Inc.
7NCAA Division I programs
30/60/90Day pilot framework
Customer engagement research · via Startup Tech Pros / Riipen Labs · Aug 2026
Problem
Is there room for a new digital engagement product inside the live college basketball experience?
Approach
Synthesized attendance, loyalty and competitive data across seven NCAA Division I programs, scoring candidate concepts against relevance, evidence strength and feasibility.
Insight
A gap appeared in continuous in-game digital engagement. Live trivia and crowd-participation concepts scored strongest on the evaluation criteria.
Decision
Whether to pilot, which concept to pilot first, and what would need to be true for it to work.
Outcome / next step
A 30/60/90-day pilot framework covering university coordination, rollout and KPIs. Attendance and revenue impact were explicitly flagged as hypotheses requiring validation, not as results.
Given raw session-level behavioural data, which visitors are actually worth prioritizing?
Approach
A two-stage model combining conversion propensity with predicted customer value, validated on held-out data. The build also covered session-level feature engineering, calibration, leakage checks, SHAP-based interpretation and revenue estimation, delivered as a Streamlit prototype.
Insight
Roughly 0.80 ROC-AUC on held-out data. The top decile of ranked sessions converted at about three times the base rate.
Decision
Where a marketing or sales team should concentrate effort, and which sessions are not worth chasing.
Outcome / next step
Model output converted into practical prioritization segments a team could test. This is a portfolio project on a public-style dataset, not a client deployment.
04 / 05Portfolio project
Downtown Winnipeg
2Signal families that disagree
Business environment analysis · publicly available datasets
Problem
A downtown core where investment, vacancy, activity and development indicators did not tell one consistent story. What is actually happening?
Approach
Combined multiple public datasets to compare long-term investment signals against shorter-term vacancy and street-level activity indicators.
Insight
A genuine divergence: long-term investment indicators were stronger than short-term activity indicators over the same period.
Decision
For an operator or investor, whether to read the area as recovering or as still declining, and which indicator to watch.
Outcome / next step
Framed as an executive briefing on what the divergence means for decisions rather than on the mechanics of the analysis.
05 / 05Portfolio project
Application Funnel Analysis
86Applications coded
21 → 0Highest-priority targets → interviews
0%Interview rate from one job board
Self-collected dataset · 86 applications coded by source, sector, role track, priority tier, response time and funnel stage
Problem
A high-volume pipeline with no visibility into which inputs were actually producing outcomes. It’s the same problem most businesses have with a sales funnel. Plenty of activity at the top, no reliable read on which channel or segment justifies the effort.
Approach
Built a structured tracker rather than a list. Every application coded by source, sector, role track, a self-assigned priority tier, days to first response and the furthest stage reached. Conversion then measured by segment instead of by impression.
Insight
Overall: 37% produced a response of any kind, 12% reached an interview, 63% were never answered at all. The channel split was the finding. 62 direct applications converted at 9.7% to interview; 14 through one job board converted at zero, and 9 of those 14 had closed or expired before anyone reviewed them. Two channels with the same effort per application, one structurally unable to return a result.
Decision
Stop spending time on the channel with the broken feedback loop. Every interview in the dataset came from a direct application, a structured program or a referral.
What the data contradicted
Two prior beliefs failed. The self-assigned priority rating did not predict anything: the 21 highest-priority targets produced no interviews, including the 12 that had been open long enough to resolve, while the two middle tiers accounted for all 10. And fast replies were expected to signal interest: applications answered within a week advanced 40% of the time against 36% for slower ones, which is no difference at this sample size.
Next step
An ongoing single-subject dataset, so findings are descriptive rather than generalisable, and several segments have fewer than ten observations. The method is the transferable part. It applies directly to a sales or recruitment funnel.
Have a problem shaped like one of these?
The first conversation is about understanding what you’re trying to figure out and what data you already have.
