IESE AI Club

Role playbook

AI for founders.

Trying a startup idea is far cheaper. A working product need not require a technical cofounder, though scaling does. The constraint is knowing what to build and getting anyone to care. Everyone has the same superpowers; distribution and taste matter more.

This assumes you have the base. Start there if not

The job

What this role actually does.

Six things, usually at once and with too little money. AI makes each cheaper; it does not pick the problem or persuade the first customer to take a risk.

01Validation 02Building 03Distribution 04Customer calls 05Operations 06Economics

Most examples are vendor customer stories. Their self-reported, unaudited numbers show something was built and used—not a benchmark to expect.

01

Validation

Research
The use case
Testing whether the problem is real before building: reading what people already complain about, sizing the pain, finding who has it.
Job to be done
When I have an idea, I want evidence for or against it within days, so I stop early if it is wrong.
Value to the business
Months of build time not spent on a problem nobody has.
How to evaluate it
Hypotheses validated or killed per month, and whether anyone committed money or time before you built. Synthesised evidence is not a substitute for talking to a person who might pay.

What AI does here

Before building, test the idea against reality. AI maps competitors, scans complaints, compares positioning and finds claims the market has heard. An idea-validation MCP tool can check live developer and community sources. Evidence, not permission.

Workflow map

  1. InputProblem statement, target customer, alternatives and public customer evidence.
  2. AI-assisted processMaps competitors, groups complaints and identifies claims to test.
  3. Human checkpointFounder validates sources and interviews people before choosing a bet.
  4. OutputA testable problem hypothesis and interview plan.

What is still yours

Pick the problem and what people might pay to solve. A crowded market can prove demand; an empty one can mean nobody cares. A score cannot commit you.

02

Building

Build
The use case
Getting a working product in front of users without an engineering team.
Job to be done
When I need something real to show, I want to describe it and get something that runs, so the gap between idea and user test is days.
Value to the business
Launching without a technical co-founder, and iterating at the speed of feedback.
How to evaluate it
Time from idea to first user, and how fast you can ship a change after feedback. Before real users or real data, get security and privacy reviewed: generated code ships generated vulnerabilities.

What AI does here

Coding agents can take a written description to a running narrow product in a week, including interface, database and integrations. They cut test cost; they do not remove engineering once it must scale safely.

Workflow map

  1. InputUser story, success metric, constraints and a deliberately narrow scope.
  2. AI-assisted processGenerates a prototype, tests and iterates on clearly specified changes.
  3. Human checkpointFounder reviews security, quality and whether the build tests the assumption.
  4. OutputA deployable MVP and a feedback loop with real users.

What is still yours

Choose what the first version proves and refuse the rest. Ship the smallest product that puts the central assumption before a real customer.

03

Distribution

Drafting
The use case
Producing and placing content across channels from a small number of ideas.
Job to be done
When I am the entire marketing team, I want one idea turned into every channel's version, so distribution is not capped by what I can personally write.
Value to the business
Reach that would otherwise require a hire.
How to evaluate it
Cost per acquired user by channel, not assets published. Publishing volume is the vanity metric here.

What AI does here

One idea becomes posts, launch email, outreach and positioning tests. AI makes volume cheap and learning faster; it also fills the internet with interchangeable copy. More output is not more attention.

Workflow map

  1. InputCustomer insight, launch goal, proof points, voice guide and channel plan.
  2. AI-assisted processDrafts variations and repurposes approved source material by channel.
  3. Human checkpointFounder chooses the point of view, checks claims and prioritizes tests.
  4. OutputA small campaign with tracked learning rather than undirected volume.

What is still yours

Taste is the filter: what is worth saying, which channel deserves persistence and when copy gives a reason to care. Distribution compounds because a founder shows up.

04

Customer conversations

Meetings
The use case
Recording, transcribing and synthesising customer conversations into something the whole company can search.
Job to be done
When I finish a call, I want the insight captured and findable, so what customers told us survives past my memory.
Value to the business
A compounding evidence base rather than a founder's recollection.
How to evaluate it
Share of conversations captured and later retrieved, and product decisions traceable to them. The founder still runs the calls; outsourcing those removes the point.

What AI does here

Record calls, then synthesise them. AI separates repeated pain from one-offs, finds natural language and exposes where words conflict with behaviour. It gives a fast team memory.

Workflow map

  1. InputConsent-based recordings, interview guide, notes and product context.
  2. AI-assisted processTranscribes, retrieves themes and organizes evidence across conversations.
  3. Human checkpointFounder revisits the source, probes ambiguity and decides what to test.
  4. OutputA searchable customer-insight record and prioritized product questions.

What is still yours

The conversation. Ask the awkward follow-up, notice hesitation and resist pitching long enough to hear the truth. Customers provide evidence, not the product.

05

Operations

Automation
The use case
The recurring back-office work (support, billing questions, scheduling, onboarding steps) handled without a hire.
Job to be done
When the same operational request arrives for the fiftieth time, I want it handled automatically, so headcount goes to work only a person can do.
Value to the business
Lower cost to serve, and a smaller team for longer.
How to evaluate it
Cost to serve per customer, deflection rate, and satisfaction on the automated path. Watch the escalation route: a customer trapped in a loop is worse than a slow human reply.

What AI does here

A one-person company can have a credible back office: prepare invoices, draft support replies, update CRM and assemble reports. Automate repeatable work after understanding it; approve money, access and promises.

Workflow map

  1. InputDocumented repeatable process, permissions, customer data and exception rules.
  2. AI-assisted processClassifies requests, drafts responses and triggers bounded workflow steps.
  3. Human checkpointOwner approves money, access and unusual customer commitments.
  4. OutputA monitored operating process with a clear escalation path.

What is still yours

Accountability and persistence. Customers do not care that an agent erred. You own the exception, apology and decision to automate.

06

Economics

Analysis
The use case
Billing, collections, revenue reporting and the numbers that tell you whether the business works.
Job to be done
When I need to know where the money is, I want billing and reporting to run themselves, so I see the truth weekly instead of at year end.
Value to the business
Cash collected sooner, and decisions made on current numbers.
How to evaluate it
Days sales outstanding, share of invoices collected without chasing, and finance hours per week. Reconcile to the bank: automated revenue reporting still needs a human close.

What AI does here

AI models token costs, support load, price tiers and margins under usage assumptions. AI features carry variable request cost; track cost per successful outcome, not call.

Workflow map

  1. InputUsage events, model and infrastructure costs, prices, margins and scenarios.
  2. AI-assisted processReconciles usage, models unit economics and flags variance or cost drivers.
  3. Human checkpointFounder tests assumptions and decides price, limits and investment.
  4. OutputA reviewed unit-economics model with explicit operating decisions.

What is still yours

Price the value and walk away from broken economics. Customers pay for an outcome, not tokens. Connect them, defend price and change product when usage outruns margin.

Start this week

  1. Run one idea through an idea-validation MCP tool, map the five closest alternatives and write what your customer would choose today instead.
  2. Build only the path that proves the core promise, then put it in front of three prospective customers before adding another feature.
  3. Model revenue, model cost and support time for one light, one normal and one heavy user at your intended price.

Where it fails in this role

  • Mistaking build speed for demand. A product finished in a week still has zero customers until somebody chooses it over doing nothing.
  • Scaling generic distribution. Everyone can generate acceptable content and outreach, so undifferentiated volume mostly creates better-filtered spam.
  • Automating before learning. A bad process running unattended is faster, cheaper and harder to notice.

Other playbooks

All nine playbooks Consulting Product management Sales Marketing Venture capital Operations Forward deployed engineer AI product manager

Now do it somewhere real.

A playbook is a map. The Industry Fellowship is the terrain: a term spent talking to professionals who are implementing AI in one industry, working out where it actually creates value, and building a working prototype against what you find. Most fellows start as beginners.

Explore the Fellowship Apply for the Class of 2028