The Fractional CTO Test: What Founders Should Actually Check

A disclosure before anything else: I sell this work. Fractional product and technology leadership is part of what I do, which means you should read this the way you would read any vendor's buying guide — as a checklist designed to be verified, not believed. Every check below works against me too. That is rather the point.

The fractional CTO market has a quality problem, and it is a strange one: the polish is inversely correlated with almost nothing. Excellent operators and empty suits produce equally good websites, equally confident calls, equally plausible LinkedIn histories. Polish is cheap now. Evidence is not. So the whole job of vetting one comes down to demanding the thing that cannot be faked in an afternoon.

What four days a month actually buys

Be honest with yourself about capacity first, because the most common failure is a mismatch of expectations, not a bad hire. Four days a month is judgement and cadence — it is not delivery. Realistically it buys: architecture and vendor decisions made and written down, a hiring loop that stops you mis-hiring engineers, a risk register someone actually owns, honest review of what your contractors or agents are shipping, and a technical narrative that survives investor scrutiny.

It does not buy someone building your product. Anyone selling four days a month as "your entire technology function" is mispricing it — and one of you is going to pay for that discovery later.

The test: evidence over polish

Ask for an artefact, not a deck. A real (redacted) decision record, an architecture memo, a code review they wrote — anything produced in the course of actual work for an actual company. People who operate through writing have a pile of it and can share something sanitised within a day. People who operate through meetings have slides about their philosophy. You are hiring for the first kind, because in four days a month, the writing is the work product — it is what your team runs on for the other twenty-six days.

Give them an hour with something real. Not a whiteboard hypothetical — your actual repository, your actual architecture doc, live. You are checking three things: do they find something true and specific rather than generic, do they say "I don't know" about anything (the ones who never do are performing), and can they still read code. That last one matters more than it used to. Most code in most early-stage companies is now written by AI agents or contractors, which means the fractional CTO's real job is verification — and someone who cannot read a diff cannot verify anything, only chair meetings about it.

Ask what they will not do. A good answer names the limits of the format unprompted: they will not be your on-call engineer, they will not personally rebuild the platform, they cannot manage a large team on four days a month. Someone who claims the format has no limits is telling you they have never operated inside it.

Ask how it ends. The correct end state for most fractional engagements is hiring your full-time CTO — and a good fractional CTO says so before you ask, because they have done it before and it did not frighten them. Someone with no exit story is planning to be a permanent line item.

Check references for disagreement, not satisfaction. Every reference call confirms the person was "great". Ask instead: what changed measurably while they were there, and what happened the first time they disagreed with the founder? The second question is the whole hire. You are buying judgement, and judgement only shows up in the moments it costs something.

Red flags, briefly

A rebuild recommendation in week one, before they can name what the current system actually does. No written artefacts from previous engagements. Conversation that gravitates to stack choices rather than risks — the stack is rarely what kills a company at this stage. A rate dramatically under market with an agency behind it. And any variation of "trust me" where "check this" would fit.

The cheap version of due diligence

The whole test costs about two hours: one artefact request, one live hour on a real problem, two pointed reference questions. If a candidate fails it, you have lost an afternoon. If you skip it, you find out the expensive way — around month four, when the polish has worn through and there is nothing written down underneath.

I have been on both ends of this market: I led product and technology across HSBC, Zopa, Smarkets and Shieldpay, took a product from zero to launch in about nine months as a CPTO, and now do the fractional version of the job. The candidates worth having will not resent the test. They will be relieved someone finally applied it.