Selected work

Recent projects

A few of the things we've built. No client names or numbers here, just an honest account of the problem, what we built, and how it landed. The interfaces shown are illustrative, to give a sense of the work.

modelling

model: demand ~ price + season + promo

AI suggestions

  • Consider a one-week lag on ‘promo’, since its effect carries over.
  • ‘price’ and ‘discount’ look collinear; keep one of them.

Draft write-up

Price elasticity is negative and significant; promotions lift demand with a roughly one-week lag, and seasonality explains much of the rest.

Illustrative interface

Build Econometrics consultancy

AI modelling inside an econometrics consultancy

The problem
A consultancy doing serious econometric and statistical modelling, where the heavy lifting ran on scarce expert time and slowed everything down: setting up models, checking specifications, and turning results into write-ups.
What we built
We built AI into the modelling workflow itself: helping draft and document model specifications, flag likely issues, and turn results into clear write-ups, with the analysts keeping full control of the judgement calls.
How it landed
Modelling moves faster, and the experts spend more of their time on the parts that genuinely need an expert rather than the mechanical setup and write-up around them.
experiments

Checkout button copy

Running · day 6
Variant A: “Buy now”
Variant B: “Get started” ahead
The bot: “Variant B is ahead and the gap is holding. Give it a little longer to be sure, then ship B.”

Illustrative interface

Build Experimentation and growth

An A/B testing bot

The problem
Running A/B tests was inconsistent and slow: experiments were set up differently each time, and reading the results properly took attention that was always in short supply.
What we built
We built a bot that helps set up experiments consistently, keeps an eye on them as they run, and reads the results back in plain language, so the test is designed sensibly and the outcome is clear.
How it landed
Experiments run the same way each time and the read at the end is quick and consistent, so more decisions get tested rather than argued about.
knowledge base

Ask me anything from the guides and manuals.

Try a question:

Illustrative interface

Build · Embed Internal knowledge and support

A single, AI-queryable knowledge base

The problem
User guides and training manuals were scattered across different places, so the same questions came up again and again and people struggled to find answers.
What we built
We consolidated all the guides and manuals into one place and made them queryable by AI, so finding an answer or working through a common issue became a conversation rather than a hunt through documents.
How it landed
Common questions get answered on the spot, the same issues stop eating support time, and the knowledge finally lives somewhere people actually use.
your day

Your day

Pulled from Slack, email and Fathom

  • Slack Four people are waiting on your reply in the Q3 pricing thread
  • Email Approve the supplier invoice before it lapses today
  • Fathom Send the follow-up you promised on this morning's call

Illustrative interface

Build · Embed Operations and productivity

Virtual PAs that triage the day

The problem
People were drowning in Slack messages, email and meeting follow-ups, with no clear view of what actually needed doing today.
What we built
We built virtual PAs that read across Slack, email and Fathom call recordings, then pull together what needs attention, what is on today, and what can wait, so the day starts with a clear picture instead of a full inbox.
How it landed
Less time lost to triage and chasing, and a clear sense each morning of the few things that genuinely need doing.

Real projects, described without client names or invented figures. The interfaces above are illustrative.

Got something in mind?

Tell us what you're trying to do and we'll give you an honest view of whether AI can help, and how.

Start a conversation