TalScan — AI CV screening
Our own product
Problem
Recruiters spent hours reading CVs that didn't fit.
What we built
AI that ranks every CV against the job description, with reasons.
Key numbers
100%of each CV read against the job, scored in seconds TalScan
AI matchingDeep scanInterview questionsContact extraction
How it works
- Upload the job description and a batch of CVs.
- AI reads every CV in full and scores the match to the job, with written reasons.
- Deep scan adds strengths, gaps and tailored interview questions per candidate.
- A ranked shortlist comes out; the recruiter makes the final call.
Claim-call intelligence
US insurer
Problem
Claim filings had errors no one could trace back to the calls.
What we built
AI that reviews calls, flags filing mistakes and guides team coaching.
Key numbers
100%of claim calls reviewed by the AI our build
Speech-to-textRule checksCoaching insights
How it works
- Every claim call is transcribed.
- AI checks each call against the filing rules and flags missing or wrong details before filing.
- Patterns are grouped by agent and team.
- Team leads get coaching points, not just error counts.
Pay benchmarking
Compensation consultancy
Problem
Pay decisions had no reliable market reference.
What we built
A benchmarking report and regression model to price every role.
Key numbers
5budget bands, each role priced by percentile our model
29–42organisations behind every role’s regression our model
Survey + public financialsQuantile regressionOutlier checks
How we did it
- Cleaned the pay survey and matched each organisation to its public financial filings.
- Fitted a log-linear quantile regression for every role, on organisation budget.
- Flagged outliers and suppressed roles with too few data points.
- Delivered percentile pay for each budget band, plus a written report.
Is Copilot moving the needle?
Large IT company
Problem
Copilot was rolled out and people used it — but leaders could not say if it changed results: deals, sales, engagement or work-life balance.
What we built
A before-and-after impact study that tests cause and effect, not just usage: who uses Copilot, and did their numbers move?
Key numbers
4business KPIs tested against Copilot use our study
2comparisons: before vs. after, and heavy vs. light users our method
26 mina day saved per user in the UK government’s 20,000-person trial — our benchmark GOV.UK, 2025 Before / afterCorrelationHypothesis testingTeam-level view
Hypotheses we test
| If Copilot works… | We measure | How we test it |
|---|
| Proposals get faster and better, so more contracts are won | Contract win %, proposals per person | Before vs. after rollout; heavy vs. light users |
| Sales teams spend more time selling | Sales value, days to close | Before vs. after, by team |
| People get time back | After-hours work, work-life balance score | Correlation with Copilot usage |
| Work feels better | Employee engagement score | Before vs. after, controlling for role and tenure |
Each hypothesis ends in a verdict — moved, no change or can't tell yet — and a next step per team.