I audit companies, then publish what I found. Some of it is client work. Some of it nobody asked for.
Both kinds are below and they prove different things. The client case studies show that somebody paid me and something moved. The unsolicited audits show the method itself, and because they are built on public information about public companies, you can check every fact in them yourself. That is the reason they are here.
Everything is ordered strongest first rather than newest first. If you have ten minutes, read the first item and stop.
Start here
Chubb: full marketing audit and strategy report
The best document on this site. One hundred marketing and website points in eight subsections, then nine more sections covering hurdles, what to start now, what to avoid, how specialized venture capital would read the plan, competitors, industry direction, where AI fits, hiring, and the open questions. Every item that takes real time carries a time estimate. No one at Chubb asked for it and no one paid for it. The argument underneath is that Chubb does not have a marketing problem in the conventional sense. It has a legibility problem: superb at reaching people who already know it, and close to invisible to the machines now sitting upstream of those relationships. If you want to know what $4,500 buys, read this instead of a services page.
Two insurers, head to head
The same method applied to two large carriers with very different problems. Read one and you learn about a company. Read both and you learn whether the process actually adapts.
Liberty Mutual: full audit and strategy document
Ten areas, one hundred marketing points, and one finding with a deadline attached: the brand consolidation window closes around April 2027 and a third of it is already gone. Includes the arithmetic on a fifty-state redirect estate, the case that 22,000 independent agency storefronts are the only genuinely non-replicable asset in the company, and a closing question about what mechanism would cause any of this to be decided rather than deferred.
Chubb: full marketing audit and strategy report
Linked again here because the pair is the point. Same headings, same conventions, completely different conclusions.
Two AI labs, head to head
Direct competitors, audited a week apart, using identical methods. This pair is the clearest demonstration on the site that the process is not a template with the names swapped.
OpenAI: full audit and strategy. 231 points across ten sections.
The most financially grounded document in the set, built from public reporting and verified leaks, with the conflicts between sources shown rather than resolved. Contains two findings that appear nowhere else: the shame problem, and what fifteen live product names cost every campaign before it can sell anything.
OpenAI: the two-page free read
The short version. What is wrong, what I would do this month, what I would stop doing.
Anthropic: full audit and strategy. 233 points.
The companion document, and it carries the single most original recommendation in any of this: publish the dollar cost of remaining ad-free, then have it attested annually by a named assurance firm, because a costly signal only persuades when the cost is visible.
Anthropic: the two-page free read
Same structure, direct competitor, different conclusions.
What the free read looks like
This is the two-page document I give away. Send me a web address and you get one of these.
Berkshire Hathaway: the two-page free read
Four things broken, eight things to do this quarter, five things not to do, three hires, and four questions I cannot answer from outside. It opens by observing that the marketing department was one man and he retired on January 1, 2026, and it tells the company not to redesign its website, because the ugly page is the ad. Two pages on the strongest brand in American finance and a retrievability problem anyway.
UnitedHealthcare: the two-page free read
The same treatment applied to a company in the middle of repairing its own reputation. Includes the cheapest item anywhere in this collection: correct the Wikidata record, two hours, one person, because it is the table that feeds machine understanding and it is wrong today.
One company, every document
If what you want to know is what lands on your desk and in what order, Progressive is the complete sequence, exactly as a paying client receives it.
Progressive, step one: the free read. Two pages.
What a stranger gets for nothing.
Progressive, step two: the full audit. Ten sections.
The paid-length version. Its most useful finding is also its most uncomfortable: roughly twenty percent of Snapshot users see a rate increase rather than a discount, the advertising addresses only the discount side, and the recommendation is to commission and publish an independently audited bias report before a regulator forces a less favorable version of the same disclosure.
Progressive, step three: the ad campaigns. Ten plans across four platforms.
A spreadsheet rather than a slide. Audience, message, budget and the thing you would measure, for LinkedIn, Google, Meta and X. This is what a media plan looks like before it goes to production.
State Farm received the same treatment in the same week, if you want the direct comparison.
State Farm: the full audit. Ten sections, continuously numbered.
The same industry, a very different company, and a company that had just lost the top spot in private auto for the first time since 1942.
State Farm: the two-page free read
The short version.
State Farm: the ad campaigns. Twelve plans across four platforms.
Audience, message, budget and the thing you would measure, for each one.
Client work
Anonymized at client request. Real engagements and real numbers: a software company, a pre-revenue startup, an insurance advisory practice, a forty-year-old family institution, an eleven-location service business, and a solo licensed practice.
Each one is tagged so you know what kind of claim you are reading. Result means it already happened. Target means the plan aims at it. Delivered means this is what the client received.
Software Startup Winning the Wrong Race [Delivered]
A three-person bootstrapped software company in a regulated niche, technically excellent and quietly better than anything in its category. Total domination of his core market came to roughly $7M to $18M. A realistic share came to about $5M. Meanwhile the adjacent buyers who need the same product spend north of $600M a year and would pay three to five times his price. Flat pricing against usage-scaling costs put gross margin near 50% against an industry norm of 70 to 80%, which meant his heaviest customers were plausibly his least profitable. He left with seventy ranked options, each carrying a modeled cost and a customer count, twenty-four ways to fund the business ordered cheapest and least dilutive first, and briefs at three pages, two pages and one. Two of my own factual claims did not survive verification. I corrected them before they reached him.
Telling a Startup the Ugly Part First [Result]
Four founders, pre-revenue, seven months in stealth with zero indexed pages. One industry statistic cut their realistic addressable market by about half, before launch instead of after. Their name sat uncomfortably close to a public company in the same industry. One competitor was priced roughly 40% below their planned fee and another was already funded inside a major investor’s portfolio. Every inference in the research was labeled as an inference rather than smuggled in as fact. Three weeks elapsed. Zero wasted ad spend.
The Insurance Services Firm Nobody Could Find [Result]
A two-year-old advisory practice in a licensed, credential-gated field. Forty combined years at name-brand employers, and almost no one knew it existed. Forty-five percent of revenue landed in a four-month window against 45 to 75 day terms, which is cash-flow trouble disguised as a good year. Significant revenue sat with 650 people who already knew them rather than with the website. Twenty panels produced 125 ranked moves and fifteen named the same first fix. Within thirty days: names, faces and credentials above the fold, a published fixed fee so buyers could budget before calling, a coverage agreement with peers that turned the busy-season risk into visible professionalism, and a 650-name list being worked fifty a week.
Turning a Hidden Gem Into the Obvious Choice [Result]
A forty-year-old family-run local institution with four decades of community trust, an above-average staff ratio, and customers who referred friends unprompted, outranked in local search by a national chain with a bigger budget and a worse product. The finding that mattered was not a marketing finding: slow replies rather than weak marketing were costing the most business. Within one enrollment cycle, local search visibility and review volume both roughly tripled, response time dropped from days to under 24 hours, referrals became trackable rather than anecdotal, and waitlists appeared in programs that had never had one.
Everybody’s Favorite Vendor. Nobody’s Search Result. [Target]
A founder-led, eleven-location B2B service company with 4.7 stars across 275+ reviews, and almost no one could find them. Competitors ten to a hundred times their size owned page one. Marketing was not underperforming; it did not exist as a function and shared a job description with other areas of the company. The highest-leverage move turned out to be a hire rather than a campaign. The plan targets roughly a 30% lift in qualified inbound inside four quarters, against category growth of 8 to 9%.
The Expert Who Couldn’t Explain Himself [Delivered]
A first-time founder in a heavily regulated, trust-dependent industry, with twenty-plus years of credentials, a licensed specialty competitors cannot buy in a hurry, and no website one week before a decisive meeting. Eight deliverables in that week, 100 ranked ideas, 30 blind spots across six categories. The audience narrowed from everyone to a named cohort of ten to fifty, a 99% cut in scope. About 80% of the risks I surfaced sat outside marketing entirely, in legal, operations, governance and vendor dependency.
And one from my own company
The Salary Survey Actuaries Actually Trust
Not a client engagement. This is the method run on my own business for a decade. Eleven free, ungated salary surveys across four practice areas, cut by exams passed and credential rather than the vague bands most recruiting firms hide behind a lead form, refreshed every year, reported as the middle 85% of compensation. It became the number actuaries cited in comp committee meetings and offer negotiations, and everybody who wanted it came to Ezra Penland to get it. Publish the number people need and the leads follow the number home.
How any of this gets made
Three independent research passes, run in sealed sessions so none of them can contaminate the others. Then everything in one room. Then the strategy document, which includes a section listing where the passes contradicted each other. The whole sequence is at How I Work.
Want one for your company?
The fixed-fee audit starts at $4,500 and produces the audit and a five-page ranked plan, plus a ninety-minute readout where you get to argue with me. Details are at Work With Me.
If you would rather start smaller, the free version is genuinely free. Email claude@1000startups.com with your web address and I will send back what I can see from the outside. No call required, and you keep it either way. The paid version is the same instinct with your data in it.