1.19 Customer discovery from inside the lab

The core idea#

Customer discovery is the scientific method applied to the market. You write down what you believe about customers as testable hypotheses, test them through structured interviews with the people who would use, buy and approve your product, and change the hypotheses when the evidence says so. The benchmark is the NSF I-Corps standard: at least 100 interviews in seven weeks (NSF). You can run it with or without a programme, while still in the lab.

Stage: minus-one · Lens: Deeptech · Reading time: about 35 minutes

Who this is for#

Researchers who want to know whether their result is a business before they leave the lab.
Teams preparing for I-Corps, an incubator, a BIRAC or NIDHI application, or a first pitch.
Founders who have done "some conversations" but can't say what they learned.

What you leave with#

A stakeholder map of everyone who touches a purchase in your target market.
40–100 structured interviews, logged and coded.
A business model canvas with each box marked as confirmed, rejected or untested.
A clear go, pivot or stop recommendation for each customer segment you tested.
Evidence that persuades grant panels and investors: names, numbers and costly signals.

Why researchers are good at this, and where they go wrong

Researchers already know how to form a hypothesis, design an experiment and change their mind on data. The skills transfer. What doesn't transfer is the instinct to explain your work: in a customer interview, every minute you spend describing your technology is a minute you aren't learning about theirs. The single most useful habit is to ask about the customer's past behaviour and let them do 80% of the talking.

How customer discovery works#

Customer discovery comes from Steve Blank's customer development method, which treats a new venture as a set of untested guesses and sends founders out to test them. I- Corps organises those guesses on the business model canvas, a one-page grid of nine components; teams write hypotheses for each box and change them based on what customers say (NCI I-Corps webinar).

Step 1: Map everyone who touches the purchase#

In deeptech, the person who would use your technology is rarely the person who pays for it, and neither of them can approve it alone. Interview all of these roles; a segment where you've only spoken to users is untested.

Stakeholder map for a B2B deeptech sale · 8 roles

The economic buyer is the most important and the hardest to reach. If you finish 40 interviews without knowing who owns the budget, you haven't done discovery yet. Blockers matter too: an incumbent supplier's internal champion, or the team whose work your product replaces, can stop a purchase that everyone else wants.

Step 2: Write hypotheses for each box of the canvas#

Write each hypothesis so it can be wrong, with a number where possible. Mark each one untested at the start.

Canvas box	What it asks	Example hypothesis (membrane startup from playbook 1.07)
Customer segments	Who exactly are the customers?	Mid-size pharma API plants in India and Europe, 50–500 employees
Value proposition	What problem do you solve, and how much better?	Recover 30% more solvent per batch than current distillation
Channels	How do you reach and deliver to them?	Direct sales to plant heads; installation by a local engineering partner
Customer relationships	How do you win and keep them?	Paid 3-month pilot, then annual supply contract
Revenue streams	How and how much do they pay?	Membrane modules priced on solvent saved, with a replacement every 12 months
Key resources	What must you have?	A module production line able to make 200 a year
Key activities	What must you do well?	Membrane manufacturing yield above 90%
Key partners	Who must you work with?	An equipment maker that already sells to these plants
Cost structure	What are the main costs?	Materials under 30% of module price at pilot volumes

For regulated sectors, add the questions the NIH version of I-Corps adds for life sciences: regulatory path and reimbursement (NCI I-Corps webinar).

Step 3: Test, update, repeat#

Each week: interview, update every hypothesis that changed, and decide what to test next. A hypothesis is confirmed only when several independent interviewees, ideally including a buyer, support it with specifics. One enthusiastic person is not evidence.

Setting up#

Two hours of setup saves weeks of muddle. Do all of this before the first interview.

Team roles#

I-Corps teams have three roles, and the split works well even outside the programme (Georgia Tech I-Corps hub):

Role	Usually	Job in discovery	Time
Entrepreneurial lead	PhD student or postdoc who may run the company	Books and runs most interviews; owns the tracker and canvas	15–20 hours a week
Technical lead	The PI or senior researcher	Joins technical interviews; keeps the science claims honest; opens doors through their network	5–10 hours a week
Industry mentor	Someone with experience selling into the target industry	Makes introductions; spots when you're hearing politeness rather than interest	About 2–4 hours a week

Two people per interview is ideal: one asks, one takes notes. Rotate so everyone hears customers first-hand.

The hypothesis board#

Keep the canvas in a shared document or on a wall, with each hypothesis on its own card or row. Use three colours: untested, confirmed, rejected. Update it every Friday. Keep the old versions; the sequence of changes is your best evidence for investors and grant panels that you learn from data.

The target list#

Build a list of 150–200 names before you start, covering every role on the stakeholder map in two or three segments. For each name record: role, company, segment, how you'll reach them (warm intro, cold email, event) and who could introduce you. Aim for a mix: roughly half users and technical evaluators, a quarter economic buyers and approvers, and a quarter regulators, integrators and advisors.

Tools#

A tracker. A spreadsheet is enough. Use the template in the templates section below. A calendar link so interviewees can pick a slot.
Recording and transcription, only with the interviewee's permission. Many industry people will say no; take good notes instead.
A shared folder for notes, named by date and interviewee code, not full names, if you'll share it outside the team.

Ethics and confidentiality#

Tell people who you are and why you're asking. Don't pose as a student doing a survey if you're planning a company.
Don't ask for confidential information. If they volunteer it, don't record it in shared notes.
If your institution requires ethics approval for interviews used in research, check whether that applies. Commercial discovery usually doesn't need it, but writing it up as research might.
Don't disclose unfiled inventions in technical detail. Talk about the problem and the performance you can reach, not how you reach it (playbook 2.04).

Getting the interviews#

Booking interviews is the hardest part, and it is a skill you will need all through the company's life. I-Corps instructors report teams using forums, LinkedIn, email, cold calls and personal referrals to reach the interviews they needed (University of Georgia). The pace to aim for is about 15 a week during an intensive programme (Illinois I-Corps); parttime teams should aim for 6–8.

Channels that work for researchers#

Channel	How to use it	Notes
Your own network and your PI's	Former students now in industry, industry co-authors, sponsors, consultancy clients	Fastest start; watch for bias toward friendly answers
Referrals from	End every interview by asking	Over time the best source;
interviewees	for two names	buyers trust peers
Your institution's industry links	TTO, incubator, industry liaison office, alumni association	Often under-used; TTOs know who has licensed similar technology

Channel	How to use it	Notes
LinkedIn	Search by job title and company; send a short note naming a specific shared connection or interest	Personalise every message; expect a low reply rate from cold messages
Conferences and trade shows	Industry events, not academic ones; book meetings before you go	One day at a trade show can produce 10–20 short conversations
Industry associations	In India: CII, FICCI, NASSCOM, sector bodies such as the Indian Drug Manufacturers' Association; their events and committees	Associations can introduce you to members if you explain the research angle
Government labs and PSUs	For defence, energy, space and railways: DRDO labs, ISRO centres, public sector companies	Slower to book, but they are often the first buyer in India
Online communities	Professional forums, Slack or WhatsApp groups for an industry	Useful for users; less so for budget owners

Making the ask#

Ask for advice, not a sale. "I'm trying to understand how plants handle solvent recovery" gets more meetings than "I'd like to show you our membrane."
Be specific about time. 20 minutes, two slots offered.
Use your status. Researchers at known institutes get meetings that startups don't. Use that while you still can.
Go in person where you can. Site visits produce better interviews and show you how the problem looks on the ground. Most of the teams in one I-Corps short course did their interviews in person (University at Buffalo).

India-specific tips#

Phone and WhatsApp often work better than email for plant managers and small business owners. A short voice note introducing yourself can get a reply where an email doesn't.
Visit industrial clusters (for example, pharma in Hyderabad and Gujarat, auto in Pune and Chennai, textiles in Tiruppur). A day in a cluster can produce a week's interviews.
Senior people in family-run firms often make purchase decisions themselves. Ask early whether the person you are speaking to also controls the budget.
Government buyers often decide through committees and tenders. Interview the technical committee members and the people who write specifications, not just the department head.

Running interviews#

The three rules#

Rob Fitzpatrick's The Mom Test reduces good customer conversations to three rules: talk about the customer's life rather than your idea, ask about specific past events rather than general opinions or the future, and talk less while listening more (SuperSummary). Once you have something to offer, he adds a test of real interest: commitment and advancement, meaning the customer gives up something they value, such as time, reputation or money, or takes a concrete next step (Bookmate quotes).

In practice for a researcher:

Instead of	Ask
"Would you use a membrane that recovers 30% more solvent?"	"Tell me about the last batch where you lost solvent. What happened?"
"Is solvent recovery a big problem for you?"	"What did solvent losses cost the plant last year? How do you know?"
"How much would you pay for this?"	"What are you spending on this now, and who approved that spend?"
"Do you think this could work?"	"What have you tried before? Why did you stop?"
"Would you be interested in a pilot?" (at the end)	"Could we test it on one line next quarter? What would you need from us to set that up?"

Structure of a 25-minute interview#

Opening (2 min). Who you are, why you're asking, how long it will take, and that you're not selling.
Their role (3 min). What they're responsible for and how their success is measured.
The last time (8 min). Walk through the most recent occurrence of the problem. What happened, who was involved, what it cost, how they solved it.
Current solutions (5 min). What they use, from whom, at what cost, and what they dislike. What they've tried and abandoned.
How buying works (4 min). Who else is involved, what tests a new supplier faces, how long the last change took, where the budget sits.
Close (3 min). "Who else should I talk to?" "Can I follow up when we have results?" If appropriate, ask for a specific next step.

Question banks by role#

End users: What does a typical day involve? Which part of [task] takes longest or goes wrong most often? What workarounds do you use? What would you stop doing if the problem disappeared?

Technical evaluators: What specifications does a supplier have to meet? How do you test a new material or device? What made the last new supplier fail qualification? Which metric would you look at first?

Economic buyers: What does this problem cost the business each year? Where does that show up: scrap, downtime, energy, labour, penalties? What size of saving would justify a trial? How are trials funded?

Approvers (procurement, safety, finance): What paperwork does a new supplier need? Are there approved-vendor lists? What payment terms are normal? Is there a minimum company age or size?

Regulators and certifiers: Which standards apply? How long does certification usually take, and what does it cost? Are rules about to change?

Integrators and channels: How do you choose components? What margin do you need? What would make you recommend a new supplier to customers?

Watch for false positives#

Compliments ("That sounds amazing") carry no information. Thank them and return to their problem.
Fluff ("We'd probably use something like that") means you need a specific: "When did you last look for an alternative?"
Feature requests are clues to underlying needs. Ask why they want it.
Academics' enthusiasm is not purchase intent. A professor who loves your result is rarely the buyer.

Notes#

Write up within an hour. Separate what they said (quotes and facts) from what you think it means. Record every number they gave. Tag each note with the hypotheses it supports or contradicts.

Making sense of it#

Interviews without synthesis become a pile of anecdotes. Hold a 60-minute synthesis meeting every week.

The weekly synthesis meeting#

Read out the week's interviews (20 min). Each interviewer gives a 2-minute summary: who, the most surprising thing, numbers heard, any costly signal.
Update hypotheses (20 min). For each canvas box, decide whether this week's evidence changes the status. Write down why.
Spot patterns (10 min). Which problems came up in more than a third of interviews in a segment? Which segment is pulling ahead?
Plan next week (10 min). What is the riskiest untested hypothesis? Who must you talk to next to test it?

Grading evidence#

Not all evidence is equal. Grade each data point so that a pile of weak signals doesn't look like a strong one.

Grade	What it is	Example
A: Paid or signed	Money, a purchase order, a signed LOI, a paid pilot	"They issued a PO for a₹5 lakh pilot"
B: Costly commitment	Time, data, samples, site access, an introduction to their boss	"Plant head sent 6 months of solvent loss data and invited us to visit"
C: Specific past behaviour	Detailed account of a real problem with numbers	"They lost about 12% of solvent per batch last year and tried two vendors"
D: Opinion	General views, predictions, compliments	"This would be really useful for the industry"

A segment is promising when it has several C-grade accounts from different organisations and at least one B. You are ready to commit when you have an A or several Bs from buyers.

Pivot signals#

A pivot means changing one major part of the canvas: the customer segment, the value proposition, the channel or the revenue model (UVA Licensing & Ventures Group). Signals that you need one:

Users love it, but no one with a budget sees the problem as a priority.
The problem is real but already solved well enough by a cheap incumbent.
A different role or industry keeps coming up as the one with real pain.
Buyers want a different form of the product (a service rather than a device, or a component rather than a system).
Academic customers dominate the interest. Academics often have limited funding and buy in low volumes, which can make them poor first customers (Carnegie Mellon).

Pivots in discovery are cheap. Teams in I-Corps cohorts change customer focus, find new beachheads and change their business models as a normal outcome (University at Buffalo).

Programmes that fund and structure discovery#

A programme gives you deadlines, instructors who challenge your conclusions, and sometimes money. You can follow the method without one, but most teams do more interviews, and better ones, inside a structure.

NSF I-Corps (US)#

	Regional / Hub programmes	National Teams
Length	A few weeks; some short courses run about 10 days ( UC Berkeley)	Seven weeks, online ( CASRAI)
Interviews	About 20 ( Clarkson)	At least 100( NSF)
Funding	Usually small or none	Up to $50,000, from which a $10,000 participation fee is paid ( Angels Partners)
Eligibility	Open to university teams in the hub's region	Recent NSF award, or completion of a regional programme with a recommendation( Angels Partners)

	Regional / Hub programmes	National Teams
Rule to	—	No R&D or product work during the
know		seven weeks ( NYU)

The NIH and other US agencies run their own versions for life sciences, which add regulatory and reimbursement questions (NCI I-Corps webinar). National I-Corps completion is often used as preparation for an SBIR Phase I proposal (Angels Partners).

India and elsewhere#

India has no single national programme matching I-Corps. Options:

Institute pre-incubation programmes. Many IIT, IISc and IISER incubators run customer discovery bootcamps for researchers before incubation. Ask your incubator.
NIDHI-EIR. The stipend (playbook 1.04) funds the founder's time while they do discovery full-time.
BIRAC and C-CAMP programmes for life sciences teams, which often include mentoring on market validation.
Run it yourself. Use this playbook, set the 100-interview target, find an industry mentor, and ask your incubator manager to sit in on weekly synthesis.

In the UK, Innovate UK's ICURe programme funds early-career researchers to spend several months testing the market for their research. In Europe, check your university's proof-of-concept and venture-building programmes.

Case files: researchers who changed course after interviews#

I-Corps has run long enough to produce many documented examples. Steve Blank, who designed the curriculum, wrote in 2014 that the programme had already taught more than 700 teams of scientists (Steve Blank). The cases below show the most common outcomes: a new customer, a new business model, or a new beachhead.

Impact Proteomics (Carnegie Mellon): from academic labs to paying customers

The team started with academic researchers as their target. Interviews showed that academics struggle for funding and buy in small volumes, so they were poor customers for a startup. The team pivoted toward customers with budgets, and after many more interviews found that 78% of the potential customers they spoke to were willing and ready to spend money on the product (Carnegie Mellon).

Lesson: For research-tool startups, the first instinct is to sell to people like yourself. Test that instinct early; academic buyers rarely fund a company.

Haro: finding a partner instead of a customer#

After 120 interviews, the team found that oncology organisations acting as NCI-funded clinical development partners would take Haro's compound and develop it at their own expense, all the way into the clinic (Steve Blank).

Lesson: Discovery can change the business model, not just the customer. For drug and biotech teams, the "customer" may be a development partner who funds the next stage.

Cardiax: catching your own bias#

After 80 interviews, this team realised they had been jumping to conclusions and reading their own bias into what they heard, and corrected course (Steve Blank).

Lesson: Bias builds up gradually. Weekly synthesis with someone outside the team (an instructor or mentor) is the best defence.

QuikCal (Duke students via NC State I-Corps): wrong segment to exit#

An undergraduate team building a construction-planning app joined an I-Corps cohort. Through more than 30 interviews in the construction sector, they quickly found they had been targeting the wrong segment (NC State). The process also helped them map the construction ecosystem and find champions outside their direct customers. The company went on to a successful exit (NC State).

Lesson: Even 30 interviews can reveal a wrong segment. Mapping the whole ecosystem finds allies you wouldn't otherwise meet.

What these cases have in common#

Team	Interviews	What changed
Impact	Many, over 7	Customer segment: academia to budget-holding
Proteomics	weeks	buyers
Haro	120	Business model: sale to development partnership
Cardiax	80+	Method: corrected for their own bias
QuikCal	30+	Customer segment within construction

Templates#

1. Interview tracker (one row per interview)#

ID,Date,Interviewer,Name code,Organisation,Segment,Role
(user/evaluator/buyer/approver/regulator/integrator/advisor),How
reached,Problem described,Last occurrence,Cost of problem (number +
unit),Current solution,Spend on it,Buying metric,Threshold,Who
approves,Evidence grade (A-D),Costly signal offered,Hypotheses
supported,Hypotheses contradicted,Referrals given,Next step

2. Weekly synthesis memo (one page)#

# Week [n] synthesis – [dates]

## Numbers

Interviews this week: [n] (total [n]). By segment: [A n, B n, C n]. By role:
[users n, buyers n, ...]
Evidence this week: A [n], B [n], C [n], D [n]

## Biggest surprise

One paragraph.

## Hypothesis changes
| Canvas box | Old hypothesis | New status / hypothesis | Evidence (interview
IDs) |

## Segment ranking
1. [Segment] – why
2. ...

## Riskiest untested hypothesis

## Next week

Target interviews (names/roles), questions to add, decisions needed.

3. Segment summary for the go/pivot decision#

# Segment: [name]

Interviews: [n] across [n] organisations; buyers interviewed: [n]
The problem in their words: "..." (3 short quotes, different organisations)
Cost of the problem: [range, with source interviews]
Current solution and spend: ...
Buying metric and threshold: ...
Who buys, who approves, typical cycle: ...
Strongest evidence: [A/B items]
Kill criteria met? [yes/no, which]
Recommendation: commit / keep testing / drop

Mistakes and how to recover#

Mistake	Why it hurts	How to recover
Pitching in every interview	You learn what they think of your pitch, not about their problem	Hold back any description of your technology until the last 3 minutes, and only if asked
Interviewing mostly academics and friends	Polite, uninformed or non- buying answers	Set a quota: at least half of interviews with strangers in industry, a quarter with budget holders
Asking hypothetical questions	People are poor predictors of their own future behaviour	Rewrite every question to ask about the last time something happened
Counting compliments as validation	Compliments carry no information	Grade every data point A–D; ignore D grades in decisions
Never reaching the economic buyer	You don't know if anyone will pay	Ask every user and evaluator who owns the budget, and for an introduction
Doing R&D instead of interviews	Discovery stalls while the science feels productive	Ring-fence discovery time; I- Corps forbids R&D during its seven weeks for this reason
No weekly synthesis	Interviews pile up without changing anything	Hold the 60-minute meeting every week, with notes

Mistake	Why it hurts	How to recover
Disclosing unfiled inventions	Can destroy patent novelty (2.04)	Talk about problems and performance, not how the technology works; file before detailed technical conversations
Stopping at 15–20 interviews	Too few to see patterns across roles and segments	Keep going until each target segment has 15+ interviews including buyers
Ignoring a pivot signal because you love the technology	Years spent on a market that doesn't want it	Agree kill criteria in advance (playbook 1.07) and review them in synthesis