How to Evaluate a Fractional CTO

The searches I get pulled into usually start too late. Someone has already decided they need a fractional CTO, generally after reading some version of the post that shows up in my feed about once a week: a rocket emoji, five bullets on why more startups are hiring one, and a closing line about how one might be the missing piece.

The bullets aren’t wrong. They’re also not testable. There’s nothing in them you could carry into a conversation with a candidate and actually use, which makes them fine as marketing and useless as criteria.

This is an area I work in, so weigh the rest of this accordingly. It also picks up where Fractional CTO or Contract Engineer left off: that post covers what the role should and shouldn’t be, and this one covers how to tell whether a given candidate fits it. This is largely a set of questions I would want to be asked, and insight into the conditions that make fractional engagements, in my opinion, a waste of money. It is not an argument that you should hire a fractional CTO.

1. Do You Have a Decision Bottleneck#

The first question isn’t who to hire. It’s which bottleneck you actually have, because there are three and this role only addresses one of them.

You can be blocked on decisions, on capacity, or on presence. A decision bottleneck means there’s a queue of technical calls with long half-lives and nobody with the standing to make them. Capacity means you know what to build and there aren’t enough hands to build it. Presence means the work needs someone in the room daily, in standup, in the middle of the incident, in the hiring loop.

A fractional CTO addresses the first one well. It’s an expensive and slow answer to the other two.

The test is blunt: name the three decisions that are stuck right now. If they come out in thirty seconds, you have a decision bottleneck. They usually look like the data model you’ll still be living with in three years, build versus buy on something core, which vendor you’re about to be married to, or whether what the dev shop delivered is a foundation or a liability.

If you find yourself listing work rather than decisions, this isn’t your problem. Hire a senior contractor or a firm, and skip the rest of this.

2. Judgment Density, Not Discount#

“Senior technical leadership at a fraction of the cost of a full-time executive” states a price. It doesn’t describe what you get, and the two aren’t the same thing.

What you’re buying is someone who has made the decision in front of you several times already, delivered in a small and bounded number of hours. That’s a specific trade: high judgment density, low availability. It’s worth naming out loud during the conversation, because an engagement that was sold as cheap seniority tends to fall apart in month two when someone needs an answer on a Thursday afternoon and nobody’s there.

So ask the boring logistics question. How many hours a month, on what cadence, and what happens to an urgent decision between sessions. A candidate who won’t put a number on their availability is selling you the discount framing rather than the trade.

Then ask a harder one. Give them a real decision you’re facing and ask them to narrow it, out loud, with the reasoning attached. Someone who has seen a lot of companies can collapse twenty plausible options down to two remarkably fast, and that collapse is genuinely most of the value. But they’re also carrying the biases of the last five companies they worked with, and some of those won’t survive contact with your constraints. You want the reasoning visible so you can tell when you’re the exception. A fast answer with no reasoning underneath it isn’t expertise, it’s pattern matching you can’t audit.

3. The AI Question, and How to Score It#

This is the part everyone leads with now, and it’s the part with the lowest signal, because the bar most people set is “has shipped an LLM-powered feature”. That bar meant something in 2023. It doesn’t now.

A fractional CTO, or any technical leader really, making a blanket statement of “you should use AI” in 2026 is about as helpful as saying “you should use computers”. Any candidate you’re considering should be able to identify where AI does, and does not, make sense, and defend that position.

The defense is the whole test. Give them a piece of your product and ask what the non-determinism would actually touch. You’re listening for specifics: what the evaluation loop looks like and who maintains it, cost per request at your volume rather than at demo volume, what tail latency does to the flow it sits inside, where customer data ends up crossing a boundary it didn’t cross before, and what the feature does when the model returns garbage.

Someone who has really shipped this has opinions about the fallback path, because the fallback path is where they got burned. Someone who has read about it talks about capability.

The most useful answer I give clients is often that a model doesn’t belong in a particular spot at all, and that a query, a rules table, or a better-designed form gets the same outcome without the operational surface. That’s an unsatisfying thing to hear from someone you’re paying, and it’s exactly the answer I’d hire for.

4. The Conflict to Name Up Front#

Outside technical due diligence works, and it’s worth being precise about why, because the reason isn’t that the outsider is smarter.

It’s that no career outcome rides on the answer. Asking your staff engineer whether the architecture they built two years ago should be replaced is asking them to devalue their own work in front of their boss. Some people will do it anyway. It’s still a hard thing to ask of anyone, and you shouldn’t be surprised when the answer comes back hedged.

Which means the objectivity is structural rather than personal, and it evaporates the moment the structure changes. A fractional CTO who wants your full-time seat is not a neutral evaluator of whether you need to fill it. Neither is one whose firm would staff the rebuild they just recommended. Ask both questions directly, early, and treat a clean answer as table stakes rather than a selling point.

The version of this I find most telling: ask what would make them tell you not to hire them. Anyone who has done this work honestly has a list. Silence is the answer.

5. Exit Criteria on Day One#

“A bridge, not a placeholder” is the right instinct and it’s missing a mechanism. The mechanism is exit criteria, agreed on day one, while nobody yet has an interest in the engagement continuing.

Ask what ends it. A good answer has a shape to it. Mine is a flat-rate evaluation of the problem set first, ending in a written recommendation, after which you can take the report and do the work in house, hand it to someone else entirely, or extend with me. All three are fine outcomes, and the first engagement is designed to end.

Exits come in two shapes, and it’s worth knowing which one you’re buying. Sometimes what gets handed off is the recommendation, and your team executes it. Sometimes what gets handed off is the role itself, which is the case in the next section.

The failure mode when nobody writes this down is quiet. The engagement doesn’t blow up, it just settles into a permanent part-time hire, which is the worst version of both options. You’re paying for continuity you aren’t getting, and meaningful decisions have started routing through someone who’s around eight hours a week.

6. When the Job Is Review, Not Direction#

There’s a case I used to write off that I’ve come around on. You have a strong senior engineer who’s right most of the time and is growing into the technical leadership role, and what’s actually missing is someone to review the calls and either agree or push back.

That’s a real fit. It just isn’t the same engagement, and it scores differently.

Your engineer holds the role. The fractional CTO isn’t setting direction, they’re checking it, which means the value is the avoidance of bad outcomes rather than the delivery of good ones. That’s a harder thing to point at in a board update and it’s usually the better investment, because the expensive mistakes at this stage are the ones nobody senior looked at.

So what you’re evaluating for changes. You care much less about whether they’ll take charge and much more about whether they’ll defer, and whether they can teach. Someone whose instinct is to reach for the wheel is a poor fit here even if they’d be an excellent fit for the direction-setting version.

And the exit is cleaner than in the general case, because it’s a handoff by design. As the engineer grows into the role, the engagement steps down into something closer to a consultant or a coach: less frequent, more reactive, eventually a phone call a few times a year.

7. Commit to the Answer Before You Ask For It#

Everything above assumes someone is going to act on what comes back, and that assumption is where most of these engagements actually fail.

Two founders who disagree about the technical direction is a legitimate reason to bring someone in. It’s one of the better ones, honestly: an outside read can break a tie or reframe a roadmap in a way that neither of you can do from inside the disagreement. But it only works if both of you commit, up front and out loud, to living with what comes back. Without that, you haven’t bought a tiebreak. You’ve bought a third opinion to relitigate, and the disagreement continues with better slides.

The same is true when the commitment is missing for less interesting reasons. If every recommendation is going to get argued from first principles by whoever liked the old plan, you’re paying a premium for an opinion you’ve already decided to ignore.

So the precondition is worth checking before anything else on this list: has someone agreed to act on the answer, while the answer is still unknown? If not, fix that first. It’s cheaper than the engagement.

That’s really the thread running through all of this.

The questions worth asking a fractional CTO are the ones that could produce a no. If none of yours can, you’re running a sales call rather than an evaluation.

If you work through the list and it does turn out to be decisions, that’s the work I do, mostly with early-stage B2B SaaS teams on backend architecture, reliability, and the operational groundwork that has to exist before scale stops being theoretical. You can book an initial conversation if you want to talk through which side of that line you’re on.


Hopefully this works out for you, or was at least helpful. If you have any questions, don’t hesitate to shoot me an email, or follow me on twitter @nrmitchi.

Read more