What I Learned Running SEO at Uber Eats, Zendesk and Williams-Sonoma, and What I Would Do Differently With Agents
Nine years of enterprise SEO, told first person: the five lessons that held across a marketplace, a SaaS company and a retailer, and how I apply them now that we run the work with AI SEO agents.
- The Backlog Was Never the Problem. The Hands Were.
- Templates Beat Pages. Always.
- Attribution Is a Political Problem Before It Is a Measurement Problem
- Technical SEO Is Insurance, and Insurance Is a Hard Sell
- The Tool Was Never the Strategy
- What Changed When I Started With Agents
- Frequently Asked Questions
Three Programs That Could Not Have Been More Different
I ran SEO at three companies that could not have been more different. Uber Eats was a marketplace with millions of pages and no SEO hire in the US until me; by the time I left, SEO was producing $40M a month in revenue. Zendesk was a SaaS company where search sat inside Growth Engineering and reported to the CMO, at $10M a month in SEO revenue. Williams-Sonoma was a retailer with a multi-brand catalog that changed daily, $20M in ecommerce SEO revenue, and the first AI SEO program I stood up. This piece is the five things that held true across all three, and the one thing I would do differently now that we run the work with agents.
The Backlog Was Never the Problem. The Hands Were.
At every one of these companies the SEO team knew what to do. The audits were done, the briefs were written, and the fixes were approved. What we lacked was the engineering, content and analytics capacity to ship them before the next quarter's priorities arrived. At Williams-Sonoma I presented the same approved items more than once because they had not moved.
What I would do differently: treat shipping capacity as the constraint, not ideas. That is the whole reason the platform exists. The agents do not decide the strategy; I still do, and so does the customer's team. They remove the wait between approval and the change going live. On our own site the Technical SEO Agent folded five thin pages in one day on November 5, with the revert kit written before the change went out.
Templates Beat Pages. Always.
At Uber Eats I refined the existing page templates and built new ones across the US, EMEA, Latin America and Asia Pacific: city, region, cuisine, restaurant and "near me". One template change moved more pages than a year of hand-written content. The lesson is not new, but every team I have joined still spent most of its content budget on individual pages.
What I would do differently: put an agent on template hygiene first and content second. On our own site the AI Search Engineer reconciled structured data across 232 pages in one pass, because the fix lived in the build, not in the pages.
Attribution Is a Political Problem Before It Is a Measurement Problem
At Zendesk, search reported to the CMO through Growth Engineering. That reporting line decided what counted as an SEO result long before Search Console did. Finance reviewed our revenue claims, and the cases that survived had the smallest number of assumptions, not the largest number.
What I would do differently: write the measurement protocol before the change ships, with the date the read happens. We now do this for every change on indexableai.com. Each entry in our change log names the target signal, the expected lead time and the validation date before the deploy. The A/B tests on our own home and product pages run the same way, with the decision rule written down on day one.
Technical SEO Is Insurance, and Insurance Is a Hard Sell
I ran three site migrations at RingCentral, each a new site design and a move to HTTPS at the same time, without losing traffic. Nobody celebrated, because nothing happened. That is the point of preventive technical SEO and the reason it loses budget fights to content every year. At Williams-Sonoma I quantified the defensive value in every audit I presented, and it was still the hardest line to fund.
What I would do differently: stop presenting technical SEO as a project and run it as a standing check. An agent that crawls, compares and flags every week does not need a budget fight every quarter. Our Technical SEO Agent runs on our own site on that basis, and the November 5 fold came out of that routine, not out of an audit I commissioned.
The Tool Was Never the Strategy
I have run Conductor at Uber, BrightEdge at RingCentral, Botify and Lumar at Williams-Sonoma, and Semrush, Ahrefs and Screaming Frog everywhere. Every one of them told me what was wrong faster than my team could fix it. The gap between the finding and the fix is where enterprise SEO budgets go to die, and no dashboard closes it.
What I would do differently: buy fewer findings and more fixes. This is the honest version of the pitch for our platform, and it is also why I still keep Ahrefs and Screaming Frog on my own desk. They are where I look. The agents are where the work gets done, with a person approving every change before it reaches the site. The full test I run on a platform before buying it is in how I evaluate an enterprise SEO platform in a week.
What Changed When I Started With Agents
The first thing that changed was my calendar. At Williams-Sonoma a large share of my week went to re-presenting approved work. On our own site that time now goes to deciding what to approve. The second thing was the error profile. People miss things in week six of a project; agents miss things in minute one, when the prompt is wrong. Both need a reviewer, but the second is cheaper to catch. The third thing is that the data got more honest, not less, because every claim on our site now carries a pull record and a date.
I do not think agents replace the SEO director. I think they replace the part of the job I liked least, which was waiting. If you run search at a company like the three I ran it at, the pages below describe how we would set this up for your role: CMO, VP of Marketing, Content Marketing Director, SEO Director, CRO, VP of E-Commerce. And if you want to see the work before you talk to us, customer zero is the running log of what the agents changed on this site and what happened next.
Frequently Asked Questions
What is the biggest lesson from running enterprise SEO at three companies?
That the backlog was never the problem. At Uber Eats, Zendesk and Williams-Sonoma the team knew what to ship; the constraint was the engineering, content and analytics hands to ship it before priorities moved. Everything else on this page follows from that.
Do AI SEO agents replace the SEO director?
No. In my experience they replace the waiting. The director still sets the strategy and approves what ships; the agents remove the gap between approval and the change going live. On our own site a person approves every change before it reaches the site.
Why do you say technical SEO is insurance?
Because when it works, nothing happens. Three site migrations at RingCentral, each a redesign plus a move to HTTPS, went through without losing traffic, and nobody celebrated. That is why it loses budget fights to content, and why I now run it as a standing weekly check rather than a project.
Which enterprise SEO tools have you run first hand?
Conductor at Uber, BrightEdge at RingCentral, Botify and Lumar at Williams-Sonoma, and Semrush, Ahrefs and Screaming Frog at every company. Where a page on this site describes one of those from experience, it says so.
Where can I see the agents' work rather than read about it?
On customer zero, which lists what the agents changed on indexableai.com, who approved it and what happened next, with dates. Every figure on this site carries a pull record.
Key takeaway: across a marketplace, a SaaS company and a retailer the same five lessons held: capacity is the constraint, templates beat pages, attribution is political first, technical SEO is insurance, and the tool is not the strategy. With agents, the director's week moves from re-presenting approved work to approving it.
Written by Vijay Vasu, Founder of Indexable AI, who ran SEO at Uber Eats, Zendesk and Williams-Sonoma. First-person tool experience is limited to the platforms named above. Career figures (Uber Eats $40M a month, Zendesk $10M a month, Williams-Sonoma $20M) are SEO revenue as stated on the author's public resume and LinkedIn profile. Own-site figures (five pages folded November 5, 232 pages reconciled) come from the Indexable AI change log; no market statistics appear on this page.