Tobi Richmond Director, Product Management — Supply · Indeed Flex

Growth product leadership · marketplace supply

Supply is the hard side of a marketplace.

It's where I've spent the last three years at Upwork — getting invisible talent discovered, getting new talent to a first job, and getting a company that measured demand religiously to care about the side that actually fulfills it.

Time to first job — experienced talent, Upwork
5dWarm referral
120dStandard path

Over 80% of experienced talent joining Upwork never earn in their first year. Fewer than 1% earn in their first month. The gap above is the whole problem — and the reason a new Flexer without shift history is a problem I already know the shape of.

Selected work

Four bets, and what they returned.

Growth work is a series of stated wagers. Each of these started as a hypothesis with exit criteria written down before the data came in.

Network effects on the supply side

HypothesisIf talent can find each other and team up on complex projects, warm referral dynamics will carry experienced newcomers past cold start faster than the standard path.

Two compounding problems: experienced talent can't build reputation without a first job, and over half of those who get past it believe reputation is a black box they can't influence. Neither is solvable with more demand.

Over two quarters, 500K freelancers adopted the network feature. More than half invited people from off-platform — genuinely net-new supply, not existing users reshuffling. Roughly 20% of that invited talent went on to fill jobs, at a materially lower CAC than paid acquisition of comparable talent.

Average network size settled at 3.6. The 20% fill rate isn't a like-for-like read against the <1% first-month earn rate — different windows, different denominators — but time-to-hire is a clean comparison, and it moved from 120 days to 5.

Time to first job
120 days · standard
5 days · invited
24× faster
Adoption, two quarters
500K freelancers
>50% invited off-platform talent

Project Aspen — outcome-based acquisition

HypothesisNet-new SMB buyers respond to work outcomes, not freelancer roles. Test it with paid search before building anything.

Organic visibility was down 40% year over year and we couldn't stop it. The deeper problem was cultural — marketing and product ran on metrics that didn't reconcile, so there was no shared definition of a customer worth acquiring.

I ran two tracks. Six outcome-framed landing pages with paid acquisition, deliberately underfunded so SEM acted as cheap market research rather than a scaling channel. And a PQL/PQA framework built with data science, giving marketing, product, and customer success one set of leading signals to build joint plays against.

Conversion came in under hire-page benchmarks — but hire pages draw 84% of traffic from high-intent SEO while these were 95% SEM on three-week-old campaigns. Engagement and user testing agreed on the real gap: people wanted to message a freelancer before committing, and the signup flow didn't allow it. The positioning bet paid off regardless. Outcome framing is now rolling into influencer and affiliate, with localized pages underway for international acquisition.

Hero CTA click-through
3.9% · hire pages
14.6% · outcome pages
3.7×
Median session duration
295s
735s
2.5×
Return on ad spend
Incumbent · $15M spend
Aspen · $200K spend
3× ROAS → budget raised to $600K

Talent enablement across the company

HypothesisTop talent isn't missing. It's invisible until it has already earned — so discovery is a product problem, not a recruiting one.

Reporting under the Chief of Staff to the CEO after a restructuring that removed teams overnight, I partnered with every GM and VP of Product to operationalize one talent strategy. 90% of Upwork's GSV is fulfilled by top talent, and that cohort was declining 7% year over year. For every three clients seeking talent, only one freelancer above an 80% job success score was available and looking.

Human-run talent programs had been cut, so everything had to become product. I ran rapid experimentation across three fronts — acquire, discover, fill — to find what could be validated fast and shipped into the experience.

The hardest stakeholder was the VP of Brand. What moved them wasn't the strategy argument; it was experiment data plus real enthusiasm for the redesign work. Alignment isn't won in the alignment meeting. Getting everyone rowing together still took a full quarter — the part I'd compress if I ran it again.

Top talent supply, YoY
−7% · at start
+1% · three quarters later
Decline reversed
Fill rate, Q4 '25
+4%

Answer engines as a visitor persona

HypothesisDiscovery is moving from search to LLMs. If logged-out pages served structured markdown, answer engines would parse and cite us more readily.

One week to a clear read: AI visibility up 34%, with organic search up nearly 5% against a site-wide baseline of −1.1%. We expanded it to 86% LLM visibility and 29% share of voice on talent-related queries — highest of any competitor in the category.

Mid-experiment, a ChatGPT model update cut outbound links per response by 20% and took our brand visibility down 10%. Rather than let that invalidate the result, I segmented by platform: Claude and Perplexity hadn't made the same change and kept trending up, confirming the approach works wherever link-citing behavior is intact. Industry expectation for that shift was a 30% drop; we absorbed a third of it.

Downstream conversion is still too early to read honestly. Visibility and share of voice are the leading indicators; what they convert into is the next question, not a claim I'll make yet.

AI visibility, week one
+34%
Organic search vs. site baseline
−1.1% · site-wide
+5% · test pages
Share of voice, talent queries
29% · #1 in category

How I lead

Player/coach, tested honestly.

My skills should be the floor of the team's craft, not the ceiling.

I was Upwork's first growth hire for enterprise — no team, no function, no playbook — so early work was almost entirely IC while I built the team. I scaled to four PMs, then moved to a company-wide growth and monetization group built as a team of teams: dedicated and embedded members across product, analytics, marketing, and customer success, so growth work ran in parallel instead of queuing behind one group.

The test I hold myself to is why I'm doing IC work. If I'm finishing something because I don't believe my team can, I owe them coaching. When coaching is working, my IC time moves to the gaps across big problem spaces rather than going deep on one thing.

The clearest case: I was leading homepage experiments to fix a declining visit-to-registration rate, and had just won over executives who wanted to keep optimizing toward a local maximum. Visible work, good momentum. But the team was ready — so I handed them the growth playbook and the executive alignment meetings and moved to air cover. Two people were promoted into verticals I'd built specifically to give away: an analyst into product management, a senior PM to Principal.

Fifteen years

Enterprise, startup, agency.

Different constraints teach different things. Consulting taught me to get to signal fast with someone else's business on the line. Enterprise taught me what scale actually costs.

2023 — now

Upwork · Director of Product Management

Marketplace growth across acquisition, activation, retention, monetization. Launched Business Plus 0→1 — fastest-growing product in company history, 34% QoQ. Ran a 26-concept exploration for agency strategy through factor-rating and engineering scoping; two initiatives shipped as pricing and packaging changes validating a product-led upsell glidepath.

2021 — 2023

Fantasy · Director of Product Management

Built and ran a team of PMs, analysts, designers, and engineers for Amazon, Salesforce, Microsoft, Snowflake, Disney, and Patreon. Patreon mobile redesign moved engagement from 11% to 45% and subscriber retention from under 1% to 5% in a quarter.

2019 — 2021

frog Design · Senior Product Manager

Two-sided matching and personalization. Built a search and recommendation engine pairing financial advisors with clients — 15% engagement lift, 50% cross-sell uplift.

2016 — 2019

Futurefuel.io · Gigatize.io · GE Digital

Head of Product at a fintech/HR marketplace through acquisition of the company I co-founded. Before that, GE's Digital Technology Leadership Program — scaled an operating system from 10 sites to 350 across eight businesses, $13M net profit in year one.

Working range

What I bring on day one.

Growth

  • Marketplace supply & demand
  • Cold start & time-to-value
  • Referral loops & network effects
  • SEM and paid acquisition
  • Activation & lifecycle
  • Pricing & packaging

Analytics

  • SQL · Snowflake · Hex
  • Looker · self-serve dashboards
  • A/B testing & exit criteria
  • PQL / PQA frameworks
  • Funnel & cohort diagnosis
  • Factor-rating prioritization

Building

  • 0→1 product delivery
  • AI prototyping & agent workflows
  • Executive alignment
  • Cross-functional operating models
  • Team of teams design
  • PMP · AI Product Management cert

Why Flex

Here, supply is the mission — not the thing that gets cut.

At Upwork I argued for supply investment inside a demand-constrained marketplace, and watched good work get deprioritized more than once. Indeed Flex is built the other way around. The cold start problem for a Flexer with no shift history is structurally the problem I've spent three years on, and I'd like to keep working on it somewhere it's the point.