Consulting for product teams with data-heavy products

I work with heads of product and founders whose products depend on data: AI agents and assistants that need a clear scope, models that rank or match, experiments that have to be trusted, and analytics that decision-makers use. Engagements run for several months, and I do the hands-on work myself.

What I help with

AI agents and agentic systems

Scope and design products where an AI assistant gathers information, works through several steps, and shows the evidence behind its answer: which questions it should answer, what it may and may not do, how each answer is tested before release, and how the data underneath is organised. At DOGER I run product for a plain-language answer tool over European public data, with a cited source for every line. At Scaleo I designed an AI product from a blank page, covering partner matching, alerts on unusual activity and predictions.

Machine learning products

Scoring, ranking and matching models, anomaly detection, and the data foundations they need. At LUN I replaced ad ranking with a model, and at AUTODOC I built the models behind personalized offers.

Experimentation programs

Set up or repair how a team runs A/B tests. We agree the effect worth detecting before anyone builds, fix the error budget, define each metric once under a versioned definition, and keep a frozen holdout that measures what shipped overall. At AUTODOC I owned this for the platform every product team shipped through, across 27 markets.

Product analytics

Metric design, funnels, cohorts and the SQL behind them, so a product lead can see where users drop off and what to change. I built the analytics function at MacPaw from nothing and wrote the Product Analytics for PMs course.

AI feature scoping

Decide which AI feature is worth building on the data you already have, prototype it against real data, and write the scope: PRDs, a roadmap, and the tests that gate each release.

Mentoring

One-to-one sessions for product leads and data teams on experiments and applied AI, including where non-parametric tests, CUPED, uplift modelling and sequential testing stop working.

Sectors I have worked in

Proptech, B2B SaaS, marketplaces and affiliate platforms, consumer software, e-commerce and public-sector data. Agents, models and experiments for data-heavy products.

Who this suits

Heads of product at B2B SaaS and marketplace companies whose data is fragmented and who see an AI opportunity in it. Teams building AI agents or assistants on their own data. Founders of data-heavy products who need senior product judgement on what is feasible and what to build first. Data leaders standing up an analytics function or an experimentation practice.

How an engagement works

1. We start with a 30-minute intro call about your product and the decision you are stuck on.
2. We agree the scope and the outputs, usually PRDs, a roadmap or an experimentation framework.
3. I run the work over several months, with the analysis and prototypes done by me.

Evidence

• DOGER I run product for a tool that answers questions about Europe's public records in plain language and shows the records behind every line of the answer.
• AUTODOC I owned the statistics behind the A/B tests at a 27-market car-parts retailer, and the models that decided which offers a shopper saw.
• Scaleo I designed an AI product for an affiliate-marketing platform from a blank page: how partners get matched, scored and ranked.
• PeopleForce I rebuilt the roadmap, discovery and delivery of an HR software company during a forced move into new markets, and revenue more than doubled.
• PandaDoc I owned the product side of the document-editor rebuild at a $1B document-workflow company, and cut the time it takes to complete a document.
• ЛУН I replaced the ad ranking at Ukraine's most-visited property site with a model, earning 15-30% more per client without hurting the experience.
• MacPaw I built the analytics function from nothing at a consumer software company, then started the MarTech team it grew into.

Get in touch

Book a 30-minute intro, email me, or message me on LinkedIn. More about me is on the about page.