A concierge MVP means delivering the outcome your eventual product will automate, but delivering it manually, with human expertise doing the work software will one day do. The specialist consultancy most ready to build a SaaS product is often the one that has never thought of itself as a software business at all, because it has been running a concierge MVP for years and calling it the service.

Part of our guide to productisation.

  • You are probably running one already: a consultancy delivering a repeatable method for outcomes is running a concierge MVP, and the manual months are what make the product hard to copy.
  • Test willingness to pay, not interest: a defined input, a defined output, a set delivery window and a price nobody negotiates.
  • Existing clients flatter the result: start with them, then add two or three clients who did not come through a prior relationship.
  • Enough runs: three to five paid engagements at a consistent price, with at least two clients returning or asking what comes next, justify a serious product conversation.
  • A weak or mixed signal points to scope, price or segment, and each can be fixed without writing software.

That reframing changes the timetable. Treated as a startup technique, the concierge phase is a hurdle to clear before the real work of building. Treated accurately, it is the most valuable asset in the business and the one part a competitor cannot acquire quickly. The months you spend delivering by hand are not the delay before the product. They are what makes the product difficult to copy.

Are you already running one without realising it?

If your business delivers a repeatable process for clients, charges for outcomes rather than hours, and applies an internal method to produce those outcomes, there is a reasonable chance you are running a concierge MVP and filing it under normal trading.

Consider what that looks like in practice. A growth advisory analysing a client's acquisition funnel and returning a prioritised action plan. A due diligence practice synthesising data from several sources into an investment memo. A regulatory consultancy reviewing documentation and returning a structured gap analysis. In each case there is a method, applied consistently, producing an output somebody pays for. The method is the proto-product, and the clients paying for it are validating demand before any software exists.

The clients who keep coming back are not buying your time as such; they are buying the outcome your method reliably produces, which is precisely what a subscription sells. The signals are easy to miss because they sound like scope creep. When somebody asks whether there is a portal, or whether they could run their own queries between engagements, they are describing a product they would pay for: the method, packaged separately from the person who delivers it.

The inverse signal is worth as much. If delivery is consuming more capacity than you can comfortably scale, and the constraint is the hours required rather than the quality of the thinking, the bottleneck is operational rather than intellectual. Software tends to be a better answer to an operational bottleneck than hiring is.

The slow part is the point

Running your method as a service generates something no competitor can replicate on a schedule: the actual patterns of how your specific client type deals with your specific kind of output. Not market research and not synthetic test data, but the granular record of what clients flag as wrong, what they ask to have reformatted, which sections they read first, what they print and what they forward. That accumulated understanding becomes the specification for the software, and somebody starting two years later cannot buy their way to it.

A business that builds without this foundation is guessing about interface, edge cases and how much the output needs to explain itself before a professional will rely on it. A business that has run the approach with real clients knows, and that gap does not close with funding, because what produced it was time and proximity to a particular market.

In specialist markets the effect compounds, because clients know each other and referrals travel quickly. Early clients who helped shape the tool become visible advocates, and social advantages are the slowest of all to copy.

One honest limit: this works because you already have clients willing to engage with something half-built and tell you the truth about it. If you are entering a market where you have no relationships, validation looks quite different. For an established specialist business, the existing client base is the most valuable asset in the exercise. It also flatters the result, which is why the test needs some structure.

How do you test willingness to pay during a concierge MVP?

Interest is not demand. A client who says "yes, we'd use that" is imagining a frictionless future tool, not the onboarding call or the invoice approval. A transaction collapses that into one honest signal. If they pay, you have demand. If they pay again, retention. If they refer someone, growth.

So a concierge MVP is not your normal consulting work relabelled as validation. Four things have to hold constant across every client run:

  • A defined input: what the client provides, in what format, by when. Accept whatever arrives and you are doing bespoke work.
  • A defined output: the same type of deliverable every time, whether a report, a score or a set of recommendations, even when the content varies.
  • A set delivery window: a product has a service level, and the concierge must too.
  • A price that is not negotiated: if every client pays a different amount you cannot compare willingness to pay across the cohort. Charge a real price from the first run: a discounted pilot says nothing about whether a subscription price would hold.

Holding those four constant is the point: you find the step that takes three times as long as expected, the input always sent in the wrong format, the section nobody reads. Those are your product requirements.

Run a pricing experiment at your next renewal

You do not need a product to test a product price. You need a proposal. Structure your next retainer renewal as if you were selling software: a named, bounded scope (not "we will advise you on X" but "you receive the Y report and the Z data summary, delivered monthly"), a set monthly fee and a defined term. Clients who accept without negotiation are telling you the value is clear and the price is right. Clients who negotiate are telling you the value is real but the calibration needs work. Clients who decline because they want the bespoke relationship are not your SaaS market, at least not yet. Silence usually means the scope was unclear.

Writing that proposal is useful in itself. Most methods turn out to contain three or four genuinely repeatable components and one judgement layer that varies by client. The repeatable components become the software. The judgement layer may stay a service, or become the onboarding, configuration and support around the product.

Discount the goodwill in your existing clients

Existing clients carry goodwill debt. They want to support you, and they may pay for a concierge run because they like working with you, which says a lot about loyalty and little about whether a stranger would subscribe to the same output. Treat their runs as the development environment, and recruit at least two or three clients who found you through some other channel before drawing conclusions about demand.

How many runs are enough?

There is no universal number, but in our experience three to five paying engagements at a consistent price, with at least two clients returning or asking what comes next without being chased, is enough to justify a serious product conversation. A pricing experiment across your client base needs six to twelve clients over one or two renewal cycles, typically two to four months; waiting longer rarely changes the conclusion, and only delays the decision.

Scope follows signal. If five clients complete a paid assessment delivered within ten working days and three return for a second one six months later, that justifies software that automates producing the assessment. It does not justify a full platform with user management, integrations and a marketplace.

How do you read a weak or mixed concierge MVP signal?

A weak signal is information, and far better had before a build than after. It usually means one of three things: the scope is not clear enough for clients to buy confidently, the price is out of line with the value they perceive, or the right buyer is not the client you currently serve. Each has a fix that needs no software. If clients will only pay with a heavy discount, the market may exist at a price your product economics cannot support.

A mixed signal, where some clients accept and others decline for coherent reasons, usually points to segmentation: two or three client types with different needs, and a product opportunity for one of them. If a rerun with a refined scope and price still reads weak, either the method is not yet product-shaped or your client base does not map to the market you would need. Both are far cheaper to learn this way than through a build.

Sometimes the runs show that clients are paying for who delivers the work, and the output is secondary. That points to an internal efficiency tool rather than a client-facing product, a different business case with different economics. Keep this idea risk separate from execution risk: no build partner or technical co-founder fixes a weak signal, and the execution question is covered in building a SaaS product without a technical co-founder.

Which parts of your service are worth productising?

Not everything a consultancy does is a sensible candidate. The parts worth examining share a few characteristics: the work is broadly repeatable between clients, it involves processing or structuring information rather than exercising creative judgement, and it consumes more time than the value it produces really justifies. Those are the places where software can absorb the volume so that your expertise concentrates on the decisions that genuinely need it.

Within those, automate the purely mechanical steps first: data ingestion, formatting, cross-referencing inputs against a fixed framework, generating a structured output from a template. That does not change what the product is; it changes how much senior time each delivery consumes. Where the expert judgement is too contextual to encode, the software should support the expert rather than replace them, and you only see where that line falls after watching the thinking happen across enough manual runs.

What clients ask for that the manual process cannot easily give them is your feature list. If five in a row ask to see their results updated monthly rather than annually, you have found a recurring-revenue product. If nobody asks anything like it, the product thesis is probably weak.

Jason was delivering it by hand for years

Jason Nguyen spent eleven years as a professional poker player, making decisions on incomplete information and reading behaviour under pressure, then formalised that instinct through behavioural science and built Bloom Story, his behavioural marketing consultancy. The thesis was sharp: most companies profile an audience by demographics, when what matters is understanding why people buy, leave, or never considered you at all.

For years that thesis was delivered as consultancy, client by client. Which is to say the concierge phase had already run, thoroughly, before any of the software questions came up. When Jason did build a prototype with Claude Code it proved the concept and had every problem an early build has: brittle at the edges, no multi-user support or proper data layer, no structure anybody could actually purchase, and not something a paying customer could use without him. We took three months to turn that into a production platform with four core workflows, and it went live with paying clients.

One lesson from that build applies to most expert businesses: surfacing the reasoning behind outputs mattered as much as the outputs themselves. Much of an expert's perceived value rests on visible senior judgement, so the software should make the expert's role more visible, not less.

The part I find most instructive is what AskPhi looks like now, because it answers the question this whole approach usually leaves hanging.

So when do you stop doing it by hand?

The usual answer is to keep going until you have seen the full range of what clients do with the work, not the range you expected. A product built before then is built for your first client rather than your tenth.

But AskPhi suggests a better answer, which is that you may not stop at all. It runs two revenue streams: a done-for-you sprint alongside the self-serve platform. The concierge service did not end when the software shipped. It became a tier, and a premium one, for clients who want the outcome without operating anything themselves.

The service and the product are not competing for the same customer, and a consultancy that assumes it must choose usually delays the product for years on a false premise. The software serves clients who want to run it themselves. You keep serving the ones who would rather you did it, at a rate that reflects the fact that you built the thing.

Start by renaming what you already do

The practical first move costs nothing, which is unusual for anything worth doing. Go back over your last two years and mark which engagements were substantially the same job under different client names. Then look at that list and stop calling it delivery. It is a product trial that clients have been funding, and the only thing missing is that nobody wrote down what it was teaching you.

Start writing it down now: which parts of the output clients query, what they always ask to have changed, where the work takes longest relative to what it is worth. In our experience three months of that produces a better requirements document than any workshop, and unlike a workshop it pays for itself while it runs.

What you end up with is more useful than a validated idea: clients already invested in the outcome, requirements drawn from real use, and pricing anchored to what people have demonstrably paid rather than to a competitor's website. Most businesses trying to build a product would trade a great deal for that position, and you are already in it.

Ready to turn what you know into a product?

Our AI Product Build track builds your first recurring-revenue product with you, and it is yours to own, without hiring a product and AI team.

See AI Product Build Talk to a co-founder →

Already running one?

Our AI Product Build track starts exactly here: working out where the software sits inside the service you already run, then co-building it with you and sharing the upside.

Productise your consultancy