Chapter 6: Marketing Research and Analytics

Dana Okafor has a question that could cost Lumen Cycles millions of dollars if the answer is wrong: will families outside the Mountain West and Pacific Northwest buy a $2,999 cargo e-bike? In Denver, the Lumen Cargo sells well. Parents see other parents riding it to school, they can test-ride it at a nearby dealer, and the city has bike lanes. But Lumen is about to spend a large share of its marketing budget taking the Cargo national. Before she commits, Dana wants evidence.

Within an hour of asking, she has plenty of answers. Jess Patel has asked an AI assistant to estimate the U.S. market for cargo e-bikes, and it has produced a confident figure with an annual growth rate and a list of three research reports. Marcus Lee has pulled website data showing visitors from Chicago and Minneapolis viewing the Cargo page. A dealer in Portland has emailed to say "every parent who comes in asks about cargo bikes now." Each answer sounds useful. Some of them may be true. When Marcus tries to find the three research reports the AI cited, he can locate only one, and its figure is for all e-bikes, not cargo bikes.

This chapter is about turning questions like Dana's into trustworthy evidence. It covers where marketing information comes from, how to plan and carry out research, how to measure marketing performance with a few key metrics, and how AI is speeding up research while making it easier than ever to be convincingly wrong. The verification habits you learn here are the ones you'll use for the rest of the book.

6.1 Marketing information and insight

Marketers make decisions about customers they can't fully see. Information reduces, but never eliminates, that uncertainty. The goal isn't to collect as much data as possible. It's to get the information that will change a decision, at a cost that's worth it, and to turn it into insight.

Data and insight are different things. "Thirty percent of Cargo buyers also buy a child seat" is data. "Parents aren't buying a bike; they're buying a way to replace the second car on school runs" is an insight, an explanation of behavior that suggests what to build, say and sell. Insights usually come from combining several sources and thinking hard about what they mean.

Most companies organize their information into what's often called a marketing information system: the people and procedures for assessing what information managers need, developing it, and getting it to them in a usable form. The information comes from three main places.

6.1.1 Internal data

Internal data is information the company already collects in the course of doing business. It's usually the cheapest and fastest information to get, and often the most underused.

For Lumen, internal data includes:

• Sales records from the website, the dealer network and the fleet program, showing what sold, where, when and at what price.
• Website analytics, showing how many people visit, where they come from, which pages they view and where they leave.
• Customer relationship management (CRM) data: purchase history, contact preferences and service interactions for each customer.
• Service and warranty records, showing which parts fail and how often.
• App data, recording rides, battery health and service needs for customers who use Lumen's app.
• Customer feedback: reviews, chat transcripts from the website assistant, emails and dealer reports.

Internal data has limits. It tells you about your customers, not about the people who chose a competitor or never considered you. It records what people did, not why. And it is often scattered across systems that don't talk to each other: the dealer sales report may use different product codes from the website. Much of the practical work of analytics is simply cleaning and connecting internal data.

6.1.2 Competitive and marketing intelligence

Competitive intelligence is the systematic collection and analysis of publicly available information about competitors and developments in the market. Done properly, it's legal and ethical: it uses information anyone could find.

Sources include competitors' websites, prices and product launches; their job postings (a competitor hiring battery engineers is telling you something); patent filings; annual reports and investor presentations for public companies; trade press and industry events; customer reviews of competing products; and social media. Lumen's team regularly checks the prices and promotions of low-cost online e-bike brands, reads reviews of competitors' cargo bikes, and asks dealers what customers say when they compare brands in the store.

The ethical line is clear in principle: gather what's public, and don't misrepresent who you are, pay people to reveal confidential information, or take information that a competitor has tried to protect. Pretending to be a customer to get a competitor's confidential fleet pricing, for example, crosses that line.

6.1.3 Customer insight from research

The third source is marketing research: studies designed to answer a specific marketing question. Internal data and competitive intelligence are collected continuously. Research is usually a project with a defined question, method and deadline.

Companies do research to understand customers' needs and decision journeys, test new product concepts, evaluate advertising, set prices, estimate market size and track satisfaction. Large consumer goods companies like Procter & Gamble are known for in-depth customer research, including watching how people use products in their own homes. Smaller companies like Lumen do research too, often with a few interviews, an online survey and careful use of the data they already have.

6.1.4 What this means for marketers

Before collecting anything, ask what decision the information will inform and what you would do differently depending on the answer. If no answer would change the decision, don't pay for the research. And look inward first: the cheapest, fastest evidence is often already in your own systems.

6.2 The research process

Good research follows a disciplined process. It's easy to skip steps, especially the first one, and end up with a pile of data that doesn't answer the real question. The process has four main steps.

1. Define the problem and research objectives.
2. Develop the research plan.
3. Carry out the plan: collect and analyze the data.
4. Interpret and report the findings.

6.2.1 Defining the problem and research objectives

Defining the problem is often the hardest step and the most important. A manager may describe a symptom ("Cargo sales in Seattle are flat") rather than the problem that needs a decision ("Should we change how we market the Cargo in Seattle, and if so, how?"). The researcher's job is to work with the manager until the decision and the information needed for it are clear.

Research objectives usually fall into one of three types.

Type of research	Purpose	Lumen example
Exploratory	Gather preliminary information that helps define the problem and suggest ideas	Why do some parents who test-ride the Cargo not buy it?
Descriptive	Describe things such as market size, customer characteristics or attitudes	How many families with young children in Chicago live within two miles of their child's school, and how do they get there now?
Causal	Test cause-and-effect relationships	Does offering a free child seat with the Cargo increase online orders enough to cover its cost?

Exploratory research often comes first, because it helps researchers figure out which descriptive or causal questions are worth asking.

6.2.2 Developing the research plan

The research plan spells out what information is needed and how it will be gathered. A written plan, even a one-page one, forces the team to agree before money is spent. It covers the information needed, the data sources, the research approach, the contact method, the sampling plan and the research instruments.

6.2.2.1 Secondary and primary data

The first choice is whether the needed information already exists.

Researchers usually start with secondary data, because it's faster and cheaper. Secondary sources include the company's own internal data, government data (such as the U.S. Census Bureau's population and household data, or the Bureau of Labor Statistics' Consumer Expenditure Survey), trade associations, academic research, research firms that sell syndicated data (such as Nielsen and Circana, which track retail sales), news reports and company filings.

Secondary data has to be judged before it's used. Ask four questions:

• Is it relevant? Data on "e-bikes" may not tell you much about cargo e-bikes. Data on Europe may not fit the United States.
• Is it accurate? Who collected it, how, and do they have a reason to make the number look big or small?
• Is it current? E-bike markets have changed fast. A figure from the pandemic-era boom may badly overstate today's demand.
• Is it impartial? A report published by an industry group promoting e-bikes may present only favorable numbers.

When secondary data can't answer the question, researchers collect primary data. It costs more and takes longer, but it's designed for the exact question.

	Secondary data	Primary data
Cost and speed	Usually cheaper and faster	Usually more expensive and slower
Fit to the question	Often partial; collected for another purpose	Designed for the question
Control over quality	Limited; must judge someone else's methods	High, if the research is designed well
Availability to competitors	Often available to everyone	Proprietary
Lumen example	Census data on households with children in target cities	A survey of parents in those cities about school-run transportation

6.2.2.2 Research approaches

Primary data can be collected in several ways, each suited to different questions.

Observational research gathers data by watching people, actions and situations. Researchers might watch how parents load children onto cargo bikes at a test-ride event, or study how shoppers move through a bike shop. Ethnographic research is a deeper form in which researchers spend time with people in their natural settings, such as following a family's morning routine to see how they actually get kids to school. Observation reveals what people do, including things they can't or won't describe. It can't reveal attitudes or motives directly.

Survey research asks people about their knowledge, attitudes, preferences and behavior. It's the most widely used method for descriptive research, because a well-designed survey can reach many people at low cost. Its weaknesses are that people may not remember accurately, may not know why they do what they do, or may give answers they think are expected.

Experimental research tests cause and effect by changing one factor for some people and comparing their response with a control group that didn't get the change, while holding other factors constant. In digital marketing, the most common form is the A/B test: half of the website's visitors see one version of a page and half see another, and the team compares results. Because visitors are assigned at random, a difference in results can be attributed to the change, not to other factors.

Interviews and focus groups are used mostly for exploratory research. In an in-depth interview, a researcher talks one-on-one with a customer, often for an hour, probing for motives and stories. In a focus group, a moderator leads a discussion with a small group, usually six to ten people. Focus groups can generate ideas quickly, but they're small and not representative, and a few strong voices can dominate.

Approach	Best for	Lumen example	Main limitation
Observation and ethnography	Understanding real behavior and context	Riding along on families' school runs	Shows what, not why; slow; small samples
Interviews	Exploring motives and decision journeys	Hour-long interviews with ten Cargo owners and five people who test-rode but didn't buy	Small, not representative
Focus groups	Generating ideas and reactions	Parents react to three possible ad concepts	Group dynamics; not representative
Surveys	Describing attitudes and behavior across a population	Online survey of parents in five candidate cities	Self-report bias; question wording matters
Experiments and A/B tests	Testing cause and effect	Two versions of the Cargo landing page	Needs enough traffic; tests only what you change

6.2.2.3 Contact methods

Researchers reach people online, by phone, by mail or in person. Online surveys are now the most common: they're fast and cheap, and they can include images and video. Many online surveys use online panels, groups of people recruited by research firms who have agreed to take surveys for small rewards. Panels are convenient, but their members may differ from the general population, and some panel respondents rush through surveys or answer carelessly. Phone and in-person methods cost more but can reach people who don't take online surveys and allow interviewers to clarify questions.

6.2.2.4 Sampling

Researchers rarely study everyone they're interested in. They study a sample, a segment of the population selected to represent the whole.

A sampling plan answers three questions: Who is to be studied (the sampling unit)? How many people should be included (the sample size)? And how should they be chosen (the sampling procedure)?

In a probability sample, every member of the population has a known chance of being selected, as in a random sample. Probability samples allow researchers to estimate how much results might differ from the whole population. In a nonprobability sample, people are chosen for convenience or judgment, such as surveying everyone who signs up for Lumen's email list. Nonprobability samples are cheaper and often good enough for exploratory work, but you can't safely generalize from them.

6.2.2.5 Research instruments

A questionnaire is the most common research instrument, and small wording choices can change results. Common problems:

Problem	Bad question	Better question
Leading	"Wouldn't a cargo e-bike make school runs easier?"	"How do your children usually get to school?"
Double-barreled	"Is the Cargo affordable and safe?"	Ask about price and safety separately
Vague	"Do you bike often?"	"In the past 30 days, on how many days did you ride a bicycle?"
Hypothetical	"Would you buy a cargo e-bike for $2,999?"	Ask about past purchases and current spending, or test real behavior
Jargon	"How important is torque sensing?"	"How important is it that the motor helps more when you pedal harder?"

The hypothetical question deserves a special warning. What people say they'll do and what they actually do often differ, especially for expensive purchases. Behavioral evidence, such as test-ride bookings, deposits or the results of an A/B test, is usually more trustworthy than stated intentions.

Besides questionnaires, researchers use mechanical and digital instruments: website analytics, eye-tracking, and the sensors in products like Lumen's app. These record behavior precisely, but they still need careful interpretation.

6.2.3 Collecting and analyzing data

Data collection is usually the most expensive step and the one most prone to error. Respondents may be hard to reach, refuse to take part, or answer carelessly. Interviewers may introduce bias. Online surveys can be filled out by bots. Researchers check data quality by looking for incomplete answers, impossibly fast completion times, contradictory responses and suspicious patterns.

Analysis turns the cleaned data into findings: tables, averages, percentages, comparisons between groups and, for larger studies, statistical tests of whether differences are likely to be real or due to chance. Two concepts matter for judging any finding:

• Validity asks whether the research measures what it claims to measure. A survey question about "interest in e-bikes" may not measure intent to buy.
• Reliability asks whether the research would give similar results if it were repeated.

One more warning belongs here: correlation is not causation. If Lumen's data show that customers who use the app buy more accessories, it doesn't follow that the app causes accessory sales. Enthusiastic riders may be more likely both to use the app and to buy accessories. To establish cause and effect, you need an experiment or a careful causal study.

6.2.4 Interpreting and reporting findings

Research is only useful if managers understand it and act on it. A good report starts with the answer to the decision question, then shows the key evidence, then explains the method and its limits. It separates what the data show from what the researcher infers. It states how confident the team is, and what would change its mind.

A classic cautionary tale is Coca-Cola's 1985 launch of New Coke. Extensive taste tests suggested people preferred the new formula. But the research focused on taste and missed customers' emotional attachment to the original brand, and the backlash was strong enough that the company brought back the original formula within months. The research answered the question it asked. It asked the wrong question.

6.3 Marketing analytics and metrics

Research projects answer specific questions. Marketing analytics is the ongoing work of measuring marketing performance and using data to make decisions about it. It depends on choosing the right metrics, calculating them consistently and reading them in context.

6.3.1 Key metrics

There are hundreds of possible marketing metrics. A few core ones show up in almost every business. The examples below use Lumen's direct-to-consumer website numbers. All are illustrative.

Last year, Lumen's website had about 1.1 million sessions (visits) and took about 11,000 orders, for about $23.1 million in sales, roughly 55% of Lumen's $42 million in revenue. About 10,000 of those orders came from first-time customers. Lumen spent about $2.2 million on digital marketing aimed at website sales.

Conversion rate is the share of visitors (or leads, or any group) who take a desired action, such as buying.

Conversion rate = conversions ÷ visitors = 11,000 ÷ 1,100,000 = 1.0%

A conversion rate on its own means little. One percent might be good for a $2,000 product and poor for a $20 product. It's most useful for comparing pages, campaigns or time periods, and for spotting where in a funnel people drop out.

Average order value (AOV) is revenue divided by the number of orders.

AOV = $23,100,000 ÷ 11,000 = $2,100

Customer acquisition cost (CAC) is the marketing (and sometimes sales) spending needed to win one new customer.

CAC = marketing spend to acquire customers ÷ number of new customers = $2,200,000 ÷ 10,000 = $220

Return on marketing investment (ROMI) compares the profit a marketing activity produces with what it cost. Suppose a spring campaign for the Cargo costs $50,000 and the team credits it with 120 Cargo sales. If each sale brings about $1,020 in gross margin (the bike plus accessories bought at the same time, minus their cost), then:

ROMI = (gross margin from the campaign − campaign cost) ÷ campaign cost

= ((120 × $1,020) − $50,000) ÷ $50,000 = $72,400 ÷ $50,000 = 145%

ROMI uses margin, not revenue, because the company doesn't get to keep the revenue that pays for the bike itself. The hardest part isn't the arithmetic. It's the phrase "credits it with." How many of those 120 people would have bought anyway? Did the campaign get credit for sales that started with a dealer test ride? This question, called attribution, is covered in Chapter 12. Digital, Social and Search Marketing. For now, treat any ROMI figure as only as good as the attribution behind it.

Other common metrics include:

Metric	What it measures	Why it matters
Market share	The company's sales as a share of total market sales	Shows performance relative to competitors, not just growth
Customer retention rate	Share of customers who stay or buy again over a period	Keeping customers is usually cheaper than finding new ones
Churn rate	Share of customers who leave (for example, cancel Lumen Care)	The flip side of retention
Net Promoter Score (NPS)	Based on how likely customers say they are to recommend the company	A simple, widely used signal of loyalty; it's a stated intention, not behavior
Customer satisfaction (CSAT)	Customers' ratings of a purchase or interaction	Useful for spotting service problems
Brand awareness	Share of a target audience that recognizes or recalls the brand	Critical when entering new markets, as Lumen is
Return or warranty-claim rate	Share of products returned or claimed under warranty	Signals product or expectation problems

6.3.2 Customer lifetime value

Some customers are worth far more than their first purchase. A Cargo buyer might add accessories, enroll in Lumen Care, bring the bike in for paid service and, years later, buy a second bike. Customer lifetime value (CLV) estimates the total profit a customer will bring over the whole relationship.

Here's a simple, illustrative CLV calculation for a Lumen Cargo buyer. The assumptions:

• Gross margin on the first purchase (the bike and accessories bought with it): $1,020.
• In each later year, an active customer brings about $100 in gross margin from Lumen Care, paid service, parts and accessories.
• 80% of customers who were active one year stay active the next. That's the retention rate.
• Lumen counts five years after the purchase, and discounts future profit at 10% a year, because a dollar next year is worth less than a dollar today.

Year	Chance customer is still active	Expected margin	Discount factor (10%)	Present value
0 (purchase)	100%	$1,020.00	1.000	$1,020.00
1	80.0%	$80.00	0.909	$72.73
2	64.0%	$64.00	0.826	$52.89
3	51.2%	$51.20	0.751	$38.47
4	41.0%	$40.96	0.683	$27.98
5	32.8%	$32.77	0.621	$20.35
Total		$1,288.93		$1,232.42

Each year's chance of being active is the previous year's chance times 80%. The expected margin is $100 times that chance. The present value divides the expected margin by 1.10 raised to the power t, where t is the number of years from now.

So a Cargo customer is worth about $1,230 in today's dollars. Why does this matter? Because it tells Lumen how much it can afford to spend to win one. If acquiring a Cargo customer in a new city costs more than the $220 average CAC, say $400 because Lumen is unknown there, the ratio of CLV to CAC is still about 3 to 1, and the investment looks sound. If it costs $1,300, Lumen would lose money on every new customer, however exciting the growth looks.

CLV also shows which levers matter. Raising the retention rate from 80% to 90%, for example through better service or app reminders, raises the value of every customer. Notice too how much of the value comes from the first sale. For Lumen, the bike's margin dominates, so getting the first purchase right is critical, and Lumen Care adds value mainly by keeping customers connected.

6.3.3 Dashboards

A marketing dashboard collects a small number of key metrics in one view, usually updated automatically, so managers can track performance at a glance.

Here's an illustrative quarterly dashboard for the Lumen Cargo's direct-to-consumer sales.

Metric	Last quarter	This quarter	Target	Status
Cargo page sessions	46,000	52,000	50,000	Above target
Test-ride bookings (dealer and demo events)	1,450	1,380	1,600	Below target
Online conversion rate (Cargo page)	0.9%	0.8%	1.0%	Below target
Cargo orders (online)	414	416	500	Below target
CAC (Cargo, online)	$260	$310	$280	Worse
Lumen Care enrollment at purchase	38%	41%	40%	Above target
Cargo NPS	52	54	50	Above target

Reading a dashboard is a skill. A few habits help:

• Look for the story across metrics. Here, traffic is up but test rides, conversion and orders are flat or down, and CAC is rising. More people are looking but not buying. That points to a problem somewhere between interest and purchase, perhaps the page, the price or the lack of a nearby place to test-ride, rather than a problem with awareness.
• Separate leading and lagging indicators. Test-ride bookings are a leading indicator: they predict future sales. Orders and revenue are lagging indicators: they report what already happened. Falling test-ride bookings are an early warning.
• Ask what's behind the number. A rising NPS among existing customers is good news, but it says nothing about why new visitors don't buy.
• Beware of vanity metrics. Sessions, followers and impressions can rise while the business stalls. Favor metrics tied to decisions and revenue.
• Don't let the dashboard become the strategy. When a metric becomes a target, people find ways to move the metric that don't help the business. A team rewarded only for test-ride bookings may book rides with people who have no intention of buying.

6.4 AI in marketing research

AI is changing nearly every step of the research process. It can make research much faster and cheaper. It can also produce convincing nonsense faster than ever. Marketers need to know both sides.

6.4.1 Where AI helps

Desk research and summarizing. AI assistants can summarize long reports, compare competitors' public materials, and draft a quick overview of an unfamiliar market in minutes. AI answer engines that search the web can point to sources. Used well, this speeds up the early, exploratory part of research.

Survey and interview design. AI tools can draft questionnaires, suggest answer options, and flag leading or double-barreled questions. They can also translate surveys and adapt wording for different audiences. A human researcher still needs to check that each question measures what the decision requires.

Analyzing open-ended text. Some of the richest data is unstructured: open-ended survey answers, reviews, interview transcripts, chat logs. Coding thousands of comments by hand into themes used to take days. AI tools can sort comments into themes, summarize what each theme says and pull out representative quotes in minutes. Marcus already uses AI to summarize customer reviews and dealer feedback.

Social listening. Social listening tracks what people say about a brand, its competitors and its category in public online conversation, such as social media, forums and review sites. AI tools classify posts by topic and sentiment (positive, negative or neutral) and flag sudden spikes, such as a wave of posts about e-bike battery fires.

Analytics. AI tools can write spreadsheet formulas and database queries, spot unusual patterns in a dashboard, and explain in plain language what a chart shows. This lowers the barrier for marketers without deep technical training, as long as they can check the work.

6.4.2 Where AI fails

Invented facts and sources. AI assistants generate text that sounds right. When they don't have a real figure, they may produce a plausible one, attach it to a real-sounding research firm and even supply a citation to a report that doesn't exist, or that exists but doesn't say what the AI claims. This is often called hallucination. It is especially dangerous in research, because false numbers can drive real budgets.

Outdated or mismatched data. Even when a figure is real, it may be years old, may describe a different market (all e-bikes rather than cargo e-bikes, or Europe rather than the United States), or may come from a promotional press release rather than a careful study. Published estimates of the size of a market often vary widely, because different firms define the market differently and use different methods. An AI tool may pick one estimate without telling you that others disagree.

Biased summaries. When AI summarizes reviews or comments, it can overweight the most common or most emotional themes and miss rare but important ones, such as a single report of a battery problem. Sentiment analysis can misread sarcasm, slang, or comments in other languages or dialects. Social media conversation itself is not representative: the people who post about e-bikes aren't the same as the people who buy them.

Confident analysis of bad data. An AI tool will happily calculate a CLV or analyze a survey whether or not the inputs make sense. It won't necessarily notice that the sample was Lumen's email list rather than the market, or that half the survey responses came from bots.

6.4.3 Synthetic respondents

One of the newest and most debated uses of AI in research is the synthetic respondent: an AI model prompted to answer survey or interview questions as if it were a particular kind of person, such as "a 38-year-old parent of two in Chicago who drives the kids to school." Some research vendors offer panels of thousands of synthetic respondents that can "answer" a survey in minutes at a fraction of the cost of real people.

Synthetic respondents have some legitimate uses:

• Pretesting instruments. Running a draft questionnaire past synthetic respondents can reveal confusing wording, missing answer options or a survey that's too long, before real respondents see it.
• Generating hypotheses. Synthetic interviews can suggest possible objections, needs or ideas that the team then tests with real people.
• Practicing. New researchers can practice interviewing techniques.

They also have serious risks:

• They aren't customers. A synthetic respondent's answers come from patterns in the model's training data, not from anyone's actual experience, budget or school run. They can tell you what text about parents typically says, not what parents in Chicago will do next spring.
• They can reproduce stereotypes. Asked to play a member of a particular group, a model may produce a caricature drawn from common assumptions about that group.
• They're often too tidy. Real people are inconsistent, distracted and surprising. Synthetic answers tend to be more coherent, more agreeable and less varied, which can make results look more certain than they are.
• They miss what's new. For a new product or a changing market, the training data may contain little or nothing relevant.
• They're hard to validate. Unless you compare synthetic answers with real ones for the same question, you can't know how far off they are.

The safe rule: use synthetic respondents to prepare for research with real people, never to replace it for a decision that matters. If a report includes synthetic results, it should say so plainly, following the "disclose it" rule.

6.4.4 How to check AI-generated research

The "verify it" rule from Chapter 1. Marketing in an AI Era becomes a concrete routine in research. For every figure or factual claim that matters:

1. Trace it to an original source. Not "a report says," but who published it, when, and where you can read it. If the AI gave a citation, open it.
2. Check that the source says what the AI claims. Find the exact figure. AI tools often get the number, year, unit or market slightly wrong, or combine two figures.
3. Check the definition and scope. What market, which countries, which years, revenue or units, and what counts as a "cargo e-bike"?
4. Check the date. Is the figure recent enough for your decision?
5. Check who measured it and how. Is it a government agency, an independent research firm, an industry association or a company's press release? What method did they use?
6. Triangulate. Look for a second, independent source. If two credible sources disagree, report the range and explain why.
7. Label what you couldn't verify. An unverified number may still be useful as a rough guess, but it must be labeled as such, never presented as fact.

What AI does well in research	Where AI falls short	What stays a human judgment
Summarizing long documents and many comments	Inventing figures and citations	Which question the research should answer
Drafting survey questions and flagging wording problems	Knowing whether a question fits the decision	Whom to sample, and whether the sample represents the market
Coding open-ended text into themes	Catching rare but important issues; reading sarcasm	Which themes matter for the business
Monitoring social conversation at scale	Representing people who don't post	What the conversation means for the brand
Simulating respondents for pretesting	Predicting real customers' behavior	Whether evidence is strong enough to act on
Calculating metrics quickly	Checking whether inputs and assumptions make sense	Which metrics to track and what to do about them

6.5 Lumen Cycles: a research plan for the Cargo expansion

Dana asks Marcus to turn her big question into a research plan the team can carry out over the next two months, before Lumen commits most of the national Cargo budget. Marcus starts by separating the decision from the symptoms.

The decision. Which two or three cities outside Lumen's home markets should be the first targets for the Cargo, and what does Lumen need to offer there (test rides, service, price, message) to sell enough bikes to justify the investment?

Research objectives.

1. Exploratory: Why do parents choose, or reject, a cargo e-bike? What role do test rides, safety concerns and nearby service play?
2. Descriptive: In five candidate cities, how many families with young children live close enough to schools and daily destinations for a cargo e-bike to be practical, and how do they make those trips now?
3. Causal: Which landing-page message, "replace your second car" or "safer school runs," produces more test-ride bookings?

Information and methods.

Need	Source or method	Type of data
Household and commuting data for candidate cities	U.S. Census Bureau data; city transportation plans	Secondary
Competitor cargo bikes, prices and dealer presence	Competitors' websites and dealer locators; reviews	Secondary (competitive intelligence)
Current interest from candidate cities	Lumen's website analytics by city; email sign-ups	Internal
Why parents buy or don't	In-depth interviews with 12 Cargo owners and 8 people who test-rode but didn't buy	Primary, exploratory
Attitudes and current behavior in candidate cities	Online survey of parents of children under 12 in five cities, using a research panel	Primary, descriptive
Effect of message on test-ride bookings	A/B test on the Cargo landing page	Primary, causal
What people say about cargo e-bikes and battery safety	Social listening across forums and social media	Secondary, monitored with AI

Use of AI. Jess will use AI to draft the interview guide and survey, which Marcus will edit and pretest with a few real parents on staff. Marcus plans to try synthetic respondents only to pretest the survey wording. AI will help code the interview transcripts and open-ended survey answers, with Marcus reading a sample of the originals to check the themes. Every market-size figure an AI tool suggests will go through the verification routine above, and the final report will include a short note on how AI was used.

Key metrics. Marcus proposes a short dashboard for the expansion: Cargo page sessions and conversion rate by city, test-ride bookings, CAC by city, Cargo CLV (recalculated as data from new cities arrives), Lumen Care enrollment at purchase, and aided brand awareness among parents in the target cities, measured by survey before and after launch.

Your marketing plan: research plan and key metrics. Add a section to your plan, about one to two pages, that does the following:

1. State the most important decision your plan depends on that you don't yet have enough evidence for, and turn it into two or three research objectives. Label each as exploratory, descriptive or causal.
2. List the information you need and where you'll get it. For each need, say whether the data is internal, secondary or primary, and name the method (for example, interviews, survey, A/B test or social listening).
3. Describe the sample for any primary research: who, how many and how you'll recruit them. Explain one way the sample could be biased and how you'll reduce that risk.
4. Say how you'll use AI in the research and how you'll check its output. If you would use synthetic respondents, explain exactly what for, and what you would not use them for.
5. Choose five to seven key metrics for your plan's dashboard. For each, give a definition, a target or benchmark (labeled as illustrative if you're estimating), and the decision it informs. Include at least one leading indicator and one estimate of customer value, such as CLV with its assumptions stated.

If you're writing your plan for a different company, use its public information for secondary data and design the primary research you would do if you worked there. Don't collect personal data from real people for this assignment unless your instructor has approved it.

6.6 Try it with AI

In this exercise, you'll ask an AI assistant for market-size figures and then try to verify each one. The goal is to see for yourself how AI handles numbers and sources, and to practice the verification routine from this chapter. You may find that some figures check out, some come from real sources but are misquoted or out of date, and some can't be traced at all. Whatever the result, the habit you're building matters more than any single number.

Step 1. Choose a market. Use the U.S. cargo e-bike market (for the Lumen plan), or a market relevant to the company you're writing your plan about. Pick something specific enough to be useful.

Step 2. Ask for figures with sources. Use a prompt like this one.

If your AI tool can search the web, try the same prompt both with and without search, and note any differences.

Step 3. Trace every figure. For each figure, follow the verification routine in this chapter and fill in a table like this:

Figure the AI gave	Source the AI named	Could you find the source?	Does it say this? (exact figure, year, definition)	Independent second source?	Verdict
		Yes / no			Verified / partly verified / misquoted / not found

Useful places to look include the publisher's own website, government statistics agencies, industry associations, reputable news coverage that links to the original study, and your library's business databases. Many research firms publish only a summary or press release and sell the full report. If you can see only the press release, record that, since you're trusting the firm's summary of its own work.

Step 4. Ask the AI to account for what you found.

Compare the AI's advice with the seven-step routine in the chapter. Did it recommend anything you would add to the routine, or anything you would reject?

Step 5. Report. Write a short memo (about 250 words) to Dana Okafor, or to the manager of the company you chose, that:

• states the market size as well as you can support it, as a range if sources disagree, with each figure's source;
• lists which AI-provided figures you could verify, which you couldn't, and which were wrong;
• explains in one or two sentences what this means for how the team should use AI for research;
• includes a one-line disclosure of how you used AI, following the "disclose it" rule.

6.7 Summary

Marketing information comes from internal data, competitive intelligence and marketing research, and its value lies in the insight it produces for real decisions. The research process starts with defining the problem and research objectives (exploratory, descriptive or causal), then develops a plan that weighs secondary against primary data and chooses among observation, interviews, focus groups, surveys and experiments, with a sampling plan and carefully worded instruments. Data are then collected, checked for quality, analyzed with attention to validity, reliability and the difference between correlation and causation, and reported in a form that leads with the decision.

Marketing analytics tracks performance through metrics such as conversion rate, average order value, customer acquisition cost, return on marketing investment and customer lifetime value, usually brought together in a dashboard that should be read as a story and tied to decisions. AI speeds up desk research, survey design, text analysis, social listening and analytics, but it can invent figures and sources, rely on outdated or mismatched data, and summarize in biased ways. Synthetic respondents can help prepare research but can't replace real customers. Verifying every important claim, by tracing it to an original source, checking its definition, date and method, and triangulating, is the habit that makes AI-assisted research trustworthy, and it's the habit this book asks you to use from here on.

6.8 Review Questions

1. What's the difference between data and customer insight? Give an example of each for a company you know.
2. Name three sources of internal data a direct-to-consumer company might use, and one limitation of internal data.
3. For each of the following, say whether it calls for exploratory, descriptive or causal research, and suggest a method: (a) a coffee chain wants to know why sales of a new drink are lower than expected; (b) a bank wants to know what share of its customers use its mobile app weekly; (c) an online retailer wants to know whether free shipping on orders over $50 increases average order value.
4. Rewrite each of these survey questions to fix its problem, and name the problem: (a) "How satisfied are you with our fast, friendly service?" (b) "Do you find our app useful and easy to use?" (c) "Would you buy our product if it cost less?"
5. Lumen's website had 150,000 sessions in a month and took 1,350 orders, 1,200 of them from new customers. Lumen spent $300,000 on digital marketing that month. Calculate the conversion rate and CAC. Using the chapter's CLV estimate for a Cargo customer, what would you want to know before deciding whether this spending was worthwhile?
6. In the chapter's CLV example, recalculate the undiscounted five-year CLV if the retention rate rose from 80% to 90%. How much does the change add, and what could Lumen do to achieve it?
7. Explain what a synthetic respondent is. Describe one appropriate use and two risks of relying on synthetic respondents for a product-launch decision.
8. Discussion: A colleague says, "The AI gave us a market-size figure with a citation, so it's good enough for the budget request." Using the chapter's verification routine, explain what you would do before agreeing, and how you would present the figure if you couldn't verify it.