What Counts as a Real Commitment in Customer Discovery

Two questions separate a real commitment from a polite one. How expensive is it for the customer? And is it unconditional?

Cost comes in three currencies. Time is cheapest and easiest to fake, so it only counts when it is scheduled and involves the right person. Money is the strongest signal in most markets and the easiest to verify. Reputation is the most underused: a customer who introduces you to a peer or posts publicly on your behalf spends social capital they would not spend lightly.

The condition axis matters just as much. "I will pilot this in Q3 if it integrates with our warehouse system" and "I will pilot this in Q3" describe different levels of belief. The first is a real signal wrapped in a real risk. It does not count until the condition is met or converted into a deposit.

A workable scoring rule:

The scoring rewards deposits above everything, because money is the only currency that cannot be given by accident. Referrals sit close behind, because a referral to a named peer both proves belief and generates the next conversation.

It also gives you a decision rule. If a segment produces no commitment scoring 4 or above across roughly ten conversations, the problem is probably not urgent enough for that segment, however much they liked the demo. If you see three commitments at 5 or above, you have enough to justify a bounded build. Adjust the thresholds to how expensive your product is.

Generic advice fails because what counts as expensive varies by product. A consumer who gives an email address has given you almost nothing. An enterprise buyer who books a second meeting with their CFO has given you a great deal.

B2B software

B2B has the richest menu because the buying process is itself a series of commitments. In rough order of strength:

1. A scheduled session with the economic buyer, not just the end user. Insist on a calendar invite with a named attendee.
2. Access to real data or a real workflow, even read-only. This is a time and reputation commitment, because someone approves it internally.
3. A letter of intent naming a price range and a target start date.
4. A paid pilot. 100tasks treats paid pilots as standard validation evidence, and notes that a paid pilot validates willingness to test, not willingness to renew (100tasks). Price it small enough to clear a discretionary budget, large enough that someone signs.
5. A reference call with another of their customers, or an intro to a peer at a comparable company.

Avoid the pilot with no money and no success criteria. That is free consulting for the buyer and a signal of roughly zero. If they will not pay, ask what internal approval payment would require, then ask for that introduction.

Consumer apps and prosumer tools

Consumer commitments are hard to make expensive, so manufacture cost. The staged approach of two hours on desk research, twenty hours on conversations or a landing page, and only then two hundred hours on a build is a useful discipline here (Isabella Pennington). Commitments worth counting:

• Joining a waitlist that requires a qualifying form, not just an email. The qualification is the cost.
• A refundable pre-order or deposit with a clear refund date.
• Referring a named friend who also signs up. KickoffLabs treats referrals as real validation because they move a signup from passive interest to advocacy (KickoffLabs).
• Booking a call. For consumer products this is a surprisingly high-cost ask and a good filter.

Be careful with waitlists. KickoffLabs says plainly that a signup is not a paying customer and warns against over-reading a waitlist (KickoffLabs). A thousand signups is a channel test. A deposit page that takes real cards is closer to proof.

Marketplaces

Marketplaces need two commitments, one per side, and supply is usually easier to test. A seller who lists an item commits time and, in many categories, reputation. Commitments to ask for:

• A real listing with real inventory and a real price, and whether they keep it live.
• A refundable booking fee from buyers. A fee converts curiosity into intent in a way a "notify me" button never will.
• A manual, concierge run for a small number of buyers before matching at scale. Harvard Innovation Labs describes this approach as validating problem and solution before automating (Harvard Innovation Labs).

Supply commitment predicts viability better than demand surveys, because a seller who lists inventory has thought about shipping, returns, and pricing. A buyer who says "I would use this" has thought about nothing.

Hardware and physical products

Hardware commitments are the most legible, because crowdfunding normalized the pre-order. Ask for:

• A refundable pre-order at a stated price with a delivery window you can miss gracefully.
• A paid prototype evaluation for a business buyer, invoiced at cost plus a small margin.
• A design partner agreement with a purchase order or committed quantity.
• For industrial products, a site visit where the customer shares measured operational data: downtime, throughput, cost per unit. That is a time and reputation commitment and often more valuable than a signature.

One caveat: pre-orders validate willingness to pay at a price, not at a cost. Run unit economics separately, because the classic hardware failure is a popular product with a negative margin.

Services and consulting

For services the commitment is the engagement. Stage it:

1. A paid discovery session or diagnostic. This is the cleanest test: will they pay for your thinking before you have built anything?
2. A signed scope for a small first engagement, with a deposit invoice.
3. A public testimonial or a named referral, the highest-value commitment for a business whose whole acquisition channel is reputation.

Services have an advantage worth exploiting: deliver the first engagement manually and you learn more in two weeks than six months of research would teach you. Automate once you know which parts the customer values (Harvard Innovation Labs).

The ask should be a natural next step, not a test the customer can fail. Three practices make it work.

First, reconstruct a recent, specific event before asking for anything. Alloy's question guide recommends covering what triggered action, what the current approach costs, and what evidence a buyer needs before changing behavior (Alloy). By the time you ask for a deposit, you should know what they spend today.

Second, name the ask and the reason. "We are building this for six design partners, and participation means a small deposit and two hours a month. Want one of the six slots?" This is honest and makes declining easy, which is what you want: an easy no beats a difficult maybe.

Third, offer a smaller version if they hesitate. If they will not pre-pay, ask for the decision-maker meeting. If they will not book that, ask for a referral. Log which ask they accepted, because the level is your score.

One ethical note. Fake door testing, where you show a button for a feature that does not exist and measure clicks, is a legitimate demand test with a reputation cost if handled badly. ConfigCat's guidance is to lead with transparency, say the feature is under consideration, and offer a notification when it ships (ConfigCat). Userpilot warns that in a young product a poorly handled "coming soon" page can damage trust before it forms, so space repeated tests far apart for the same segment (Userpilot). A click is also a weak commitment, closer to 1 or 2 on the scale. Use fake doors to recruit people for the real ask, not as final evidence.

Most early data is partial. Treat it as an instruction to make a follow-up ask, not as proof.

A partial commitment is a commitment on one axis only: two hours but no budget holder, or a referral but no willingness to pilot. Score it for what it is, then ask for the missing axis: "Would you introduce me to whoever owns this budget?" If they will not, that tells you something about their authority or belief.

A conditional commitment is more interesting, because the condition is the deal talking. Sort conditions into four buckets:

The decision rule: a conditional commitment scores at most 3 until the condition is met or converted into something unconditional. Convert by asking for something that survives the condition. For a timing condition, ask for an LOI with the date on it. For an approval condition, ask for the meeting. For a price condition, ask for a refundable deposit at the target price.

Discount one variant entirely: the condition that is vague and unmeasurable. "If it is more robust" and "if it saves us time" are not conditions, they are politeness. Ask what number would count as saving time. If they cannot give one, the conversation is over for now.

Patterns beat single data points after tagging. Kromatic's guidance is to tag notes with recurring themes and look for signals like seven of ten interviewees describing the same behavior (Kromatic). Do the same with commitments: a spreadsheet with one column per commitment, its score, and its condition type. If the highest scores are time commitments from users and there is nothing from buyers, you have a channel problem no amount of conversation will fix.

Suppose you sell scheduling software to regional clinics and have run 12 discovery calls. Here is a scored log.

Read the numbers, not the enthusiasm. Two commitments (C and D) are unconditional and strong. One (B) is a strong conditional with a date. One (E) is a condition you may or may not satisfy, and it does not count until you hold a technical session with the EHR owner.

The decision: this is enough to run a bounded pilot with C and B if B's date can be met. It is not enough to justify building the EHR integration, because one integration condition in twelve calls is a segment question, not a product question. Ask every clinic in the next round about their EHR setup. If five of ten raise it, the integration becomes the product.

The same logic scales across product types, and it scales better with a system of record than with memory. If you are building AI into the product and want to test a pricing or workflow hypothesis quickly, a single API for text, image, audio, and embedding models lets you stand up a working test rather than a mockup (Quanta Inference). If the real question is whether you can build fast enough to test at all, that is a different problem, and one a specialist can take on with you (Quanta).

The next step is yours: rewrite one discovery script so it ends with a specific, sized ask, add a commitment score column to your notes, and run ten conversations. If nothing scores above 3, you learned something valuable and cheap. If several score 5 or 6, you have earned the right to build.