Account exec, solutions engineer, value engineer, forward deployed engineer. The people getting paid at the top do all four. Most of us do two of them well and one badly.
Twenty questions about what you've actually produced. Not what you know, and not how confident you feel about it. You get four scores out of twenty, the axis you're weakest on, and one thing you could build this week to fix it.
About nine minutes if you answer it honestly. Result's on screen. No email needed to see it.
Not today? Leave an address and I'll send you the four documents below, plus the benchmark when it exists.
Two live postings. Quoted, linked, checked today.
San Francisco and New York. $240,000 to $315,000. Three or more years. One of the listed responsibilities is "help customers develop evaluation frameworks to measure Claude's performance for their specific use cases".
New York, San Francisco and Seattle. $280,000 to $320,000. Four or more years, asking for "advanced prompt engineering, agent development, evaluation frameworks, and deployment at scale".
US sales engineer total comp is about $200k at the median and $308k at the ninetieth percentile. Both those postings start above the median. One starts above the seventy-fifth.
Neither of them is asking for more technical depth than you've got. They're asking for four things at once.
Those are US roles and US numbers. If you're selling from London or Singapore or Sydney, the shape probably holds and the actual figures won't. The diagnostic asks where you are, and where there's no reliable data for your market it says so instead of guessing.
Can you say whose budget it's coming out of, what happens after your champion says yes, and why you actually lost the last one.
Did anything exist in writing before the environment got built. Can you name the input your product handles worst.
What did this cost them before you turned up, in their numbers. And which single assumption is the whole case resting on.
What have you built that other people still run without you in the room. Have you ever built an eval set out of a customer's own examples.
Absolute bands. They don't get recalibrated to whoever happens to have taken it. If eighty per cent land in Exposed, that's what gets reported.
Read the four-axis capability model in fullNot a personality test. There's no single score out of a hundred, on purpose. A number like that hides which axis is empty, and the empty one is the whole point.
Not a comparison. Two hundred people have to take it before any percentile is honest. Until then you get your four absolute scores and a note saying how many people have done it so far.
Not scored to flatter you. Fourteen results are possible. Several of them say plainly that you support deals rather than run them. One of them is called Early.
Not a gate. Your score has nothing to do with whether you can join anything, and nobody sees it but you.
Nothing on this page is for sale.
Not summaries. The whole thing, on this site. If they're useful, take them and use nothing else here.
The evaluation framework you hand a buyer instead of an engineer. Needs your product's real failure modes and thirty of the customer's own examples before it's worth anything.
Read itWhat to say when they tell you it's too slow, across agentic, chat, batch and voice. Needs your own p95, not the one in the deck.
Read itWhat you can actually commit to in writing when the model underneath you changes.
Read itWhat the postings ask for, what gets tested, and where the practical assessment sits.
Read itWhat the role is, what it pays against the wider profession, how people actually get hired into it, and where it differs from a solutions engineer. Every number sourced with its date.
Read itTwenty-three AI-specific questions with model answers. The compliance review is the single most common thing that stalls an AI deal, and buyer question lists are free everywhere while seller answer libraries get sold.
Read itThree more are spreadsheets and they live in the room, because they're meant to be filled in for a live deal. A POC scorecard, a security questionnaire answer library, an eval results sheet.
I'm Lewis Crook. Twenty years in enterprise tech, currently at Parloa working at the frontier of enterprise voice AI. I've closed seven-figure AI deals on my own, taken over a deployment that was already failing across a hundred and ten sites and got it back, and run programmes handling over a million AI-handled minutes a month across APAC and Europe.
None of that came from a framework. It came from getting it wrong first, usually in front of someone senior.
The questions came out of deals I've been in. The four-axis capability model came out of what those postings actually ask for.
More of what I write is at lewiscrook.com, and I'm on LinkedIn at /in/iamlewiscrook.
A private Telegram group for AI sales engineers, solutions architects and forward deployed engineers. Right now it's got a comp database nobody else publishes, an objection lab for what's actually stalling AI deals, and seven working documents. You get in by applying. There's no payment and nothing to buy.
The standard, which is basically the whole point of it. This is a place to do things, not a place to learn about things. Everything in it is built to be used within a week on a live deal, and people will ask you what happened.
If you join and read everything and post nothing and use nothing, you've wasted your time and I've wasted mine. That's not a threat, it's just what would have happened.
The room's built, the documents are in it, and I'm inviting people one at a time by hand at about eight a week. If what you want is a big active community today, this isn't one yet. If what you want is to be early in a small one, apply.
Five questions, about four minutes. The one that matters is the third. A real technical objection from a real deal, in your own words. Not what you reckon about AI selling. What actually happened to you.
I read applications on Fridays and people join on Mondays, so whoever turns up has someone to talk to. Your diagnostic score isn't part of it and I don't look at it.