Generative AI for retailPeter DohertySenior Product Managerdunnhumby

I put generative AI to work in enterprise retail software.

I led my company's first client-facing generative AI, from what it does and how it's tested to pricing and running costs. I came to it from the customer's side, and I launched the platform that replaced the old one.

See my experienceRochester, New York
Share of my platforms' ~1,000 monthly active users who use the AI
~40%
Agent documentation lead time, down from months
1 week
Client-facing generative AI launch at my company, which I led
1st

Focus areas

  • AI agents in production
  • LLM evaluation
  • AI unit economics
  • Category management & assortment
  • Prototyping in Claude Code
  • Legacy modernization

About

Where I'm coming from.

Customers taught me what matters in product. Formal training and practice taught me the craft.

I've worked in enterprise SaaS since 2017, and four-plus of those years have been in product management. I started at a startup, working in support, and came to dunnhumby when the startup was acquired. I came to product from the customer's side of the table: I led support teams, and later looked after enterprise client accounts. That's where I saw our older products, a collaborative vendor portal and promotion planning software, needed replacing, so I moved into product to help build what replaced them.

In my current role I led the company's first generative AI features in a client-facing product: an agent that works across our platform, answering questions about performance, generating insights and helping people find their way around. It's tested before release, and experts and users keep an eye on it after. I shaped how it's packaged and track what it costs to run.

I know retail merchandising hands-on: category management, assortment, space and planograms, pricing and promotions. I also launched the multi-module platform that replaced our legacy software, and moved most of our clients across. Day to day, I work with squads of 20 to 35 across product, platform engineering and data science, and I author executive playbooks on taking AI from research into production.

Habits

Close to customers, close to the work.

  1. Hear it first-hand

    I run discovery sessions with customers every month. Roadmap decisions start from what we heard there.

  2. Prototype before we build

    I build clickable prototypes in Claude Code and Figma Make, then go through them with design and engineering so we find the awkward parts early. I also build internal tools in Claude Code that people use.

  3. Test AI before it ships

    Before release, our AI has to pass a scored test set and clear evaluation pipelines that act as release gates. After launch, experts review it and users tell us what's working, and I use A/B testing to check that a change actually helps.

  4. Ship with sign-off

    I run UAT for market launches and for regular releases: scripts, release notes, acceptance criteria and sign-off. That includes enterprise clients with change-control gates and formal governance.

  5. Set data expectations up front

    I scope the data inputs behind a feature so quality expectations are clear from the start, and I've worked out inbound and outbound data ingestion and pipelines to meet freshness requirements.

  6. Keep watching after launch

    Telemetry and session replay show me where people get stuck. When something frustrates them, we rebuild it.

Experience

How I got here.

I started in vendor portal software, working in support, ran service operations for a time, then moved into client success and product. Most of those years were at dunnhumby, working with enterprise retail clients.

  1. 2025
    March 2025 – Present

    Senior Product Manager

    Current

    dunnhumby

    I own two platforms used by several enterprise retailers, and lead AI and platform modernization with squads of 20 to 35 across product, platform engineering and data science.

    • Led my company's first client-facing generative AI features: an agent for performance questions, insights and platform help, now used by about 40% of the roughly 1,000 monthly active users on my platforms.
    • Set the quality bar for AI releases: evaluation pipelines and a scored test set that gate release, then expert review and live user feedback after, with A/B testing to confirm a change helps.
    • Ran UAT for market launches and releases across large enterprise clients, including those with change-control gates and governance: scripts, release notes, acceptance criteria and sign-off.
    • Worked out inbound and outbound data ingestion and pipelines to meet freshness requirements, and scoped the data inputs that feed product features, for data quality.
    • Shaped how the AI is priced and packaged (it's included in the subscription), and tracked its running costs, including cost per query and the effect of model choice.
    • Cut the time to document a new agent from months to one week. A fleet of agents and supporting skills has been built on my guidelines since.
    • Launched a multi-module cloud platform that replaced legacy software at full feature parity, with better performance. Moved most clients across, then retired the old system.
    • Automated pricing and privacy rules across NAM and EMEA, including the UK and Netherlands, with regional Heads of Market and legal counsel.
  2. 2022
    March 2022 – March 20253 yr 1 mo

    Product Manager

    dunnhumby

    I moved into product to help deliver the modernization program I'd flagged while working with clients.

    • Turned the legacy consolidation strategy into a roadmap and delivery plan, sequenced around live client commitments.
    • Served as the main product contact for bespoke data science work, connecting data science engineers directly with enterprise client needs.
    • Built data privacy, compliance and local market rules into roadmaps for capabilities used across international regions.
  3. 2021
    June 2021 – March 202210 mo

    dunnhumby

    I came back to dunnhumby to look after a portfolio of complex enterprise accounts.

    • Turned clients' pain points into a case for consolidating our legacy products. It became the company's multi-year modernization program.
    • Drove retention, expansion and adoption across pricing and optimization platforms, and negotiated contracts, service agreements and scope amendments.
  4. 2020
    January 2020 – July 20211 yr 7 mo

    Calero-MDSL

    I ran service operations for enterprise telecom expense management software through a period of major organizational change.

    • Turned business strategy into OKRs, KPIs and monitoring dashboards that consistently beat performance benchmarks.
    • Rebuilt operational processes and escalation paths during the change, keeping service steady for enterprise accounts.
  5. 2018
    May 2018 – January 20201 yr 9 mo

    dunnhumby · Rochester, New York Area

    My first role at dunnhumby. A lot of how I make roadmap calls today started here.

    • Led and mentored support teams of 10 to 20 people, with clear paths for career progression.
    • Built the escalation, quality and monitoring practices behind North American product support.

Skills

What I know best.

Most of it I learned by doing the work: shipping generative AI, retiring legacy software, and working through pricing rules across NAM and EMEA.

Product strategy

  • New platform launches
  • Legacy modernization & retirement
  • Roadmap & delivery planning

AI & data science

  • LLM evaluation pipelines & test sets
  • A/B testing to confirm a change helps
  • Data ingestion & pipelines, inbound and outbound
  • Data quality & freshness requirements
  • RAG & agent corpus design
  • AI pricing, packaging & cost per query

Tools

  • Claude Code prototypes & internal tools
  • Figma Make prototyping
  • FullStory telemetry & session replay
  • Azure & GCP

Ways of working

  • SVPG continuous discovery
  • Three Amigos collaboration
  • UAT, acceptance criteria & sign-off
  • Backlog, epics & user stories
  • OKR strategy & execution

Compliance

  • Pricing & data privacy compliance
  • NAM & EMEA, incl. UK & Netherlands
  • Automated rules frameworks

Accounts & commercial

  • Enterprise account strategy
  • Contract & scope negotiation
  • Adoption, retention & expansion

Domain knowledge

The retail areas I know hands-on, and the industries I've worked in.

  • Retail & grocery
  • Category management & assortment
  • Space & planograms
  • Pricing & promotions
  • Retail data science
  • Enterprise SaaS
  • Telecom expense management

Portfolio

Work I'm proud of.

Here's the short version of each, and I'm glad to tell the long one.

Case study 01

Getting AI out of the demo

How my company's first client-facing generative AI became a tested agent for performance questions, insights and platform help, now part of users' regular work.

Case study 02

Replacing a platform people relied on

Spotting the need on the client side, then building a multi-module replacement at full feature parity, moving most clients across, and retiring the old system.

Case study 03

The cost of an answer

Shaping how our first client-facing generative AI is priced and packaged (it's included in the subscription), and tracking what it costs to run, down to cost per query and model choice.

Contact

Let's swap notes on AI in retail.

I'm always happy to talk shop: generative AI, retail merchandising, product discovery, data and analytics products, or whatever you're prototyping. Email is the quickest way to reach me.