Short answer: mid-market companies are the missing middle in customer research — too big for founder-led discovery, too small to staff a research team. The tooling market reflects this. Startups get free tiers and self-serve. Enterprises get site licences and a research function. Companies in between get sold one or the other, and neither fits.
This is a playbook for that middle: what to run, what to skip, and how to buy — for companies roughly between $10M and $1B in revenue.
The segment nobody builds for
The National Center for the Middle Market defines the US middle market as companies with annual revenues between $10 million and $1 billion. There are nearly 200,000 of them. Together they represent about one-third of private sector GDP and employ approximately 48 million people — and during the 2007 to 2010 downturn they added 2.2 million jobs while larger firms shed them.
It is the largest under-served segment in the research tooling market, and the reason is structural rather than accidental: mid-market companies do not fit either vendor motion. They are too sophisticated for a free tier and too small to absorb an enterprise contract with seat minimums they cannot fill.
The founder-knowledge cliff
Here is what actually happens, and why so few leadership teams see it coming.
In a startup, the founder is the research function. They talk to customers most weeks, they hold the model of the market in their head, and they can answer "why do people buy this" from memory. It works. It is genuinely one of the highest-leverage things about being small.
Then two things happen at once, somewhere between roughly 50 and 200 employees:
- The founder's sample goes bad. They still talk to customers — but now they talk to the biggest accounts, the loudest complainers, and whoever the sales team puts in front of them at a conference. That is a systematically unrepresentative sample of the customer base, and it gets less representative every quarter as the company grows.
- The number of people who need customer knowledge exceeds the number who can absorb it by hallway conversation. A PM three org layers away cannot ask the founder what customers think before every decision. So they guess, or they cite something the founder said eighteen months ago about a market that has since changed.
The cliff is dangerous because it is invisible from the top. The founder still feels well-informed — they are still having customer conversations, after all. The decay is in the representativeness of the sample, not in the volume of contact, and representativeness is not something you can feel.
By the time someone notices, the company has usually shipped two or three expensive things nobody wanted.
The arithmetic that makes hiring your way out impossible
The middle market went into 2026 growing revenue considerably faster than headcount. From the NCMM's Year-End 2025 Middle Market Indicator, a survey of 1,000 C-suite executives at companies in that revenue band:
| Indicator | Year-End 2025 |
|---|---|
| Year-over-year revenue growth | 11.7% |
| Year-over-year employment growth | 7.8% (post-pandemic average: 9.8%) |
| Companies reporting revenue growth | 85% |
| Companies reporting growth of 10%+ | 53% |
| Companies increasing headcount 10%+ | 36% (down from 44% a year earlier) |
| Will invest in intelligence tools near-term | 53%, up from 44% six months earlier |
| Very confident their industry grows in 2026 | 65% |
The gap between 11.7% and 7.8% is the entire argument for how mid-market companies must operate. Every function is being asked to produce more output per head. The report puts a number on the mechanism, too: on average, companies say AI has replaced or reduced the need for human employees by 18%.
Now apply that to research specifically. Research is the function that will never get proportional headcount in a mid-market company, because the cost of not having it is invisible. A missing engineer shows up as an unshipped feature. A missing researcher shows up as a feature that shipped and did not work — eighteen months later, attributed to something else entirely.
Independent data confirms the un-staffing. The User Interviews State of User Research 2025 report, based on 485 qualified responses collected 25 July to 9 August 2025, found the share of UX researchers working at companies with zero dedicated researchers rose to nearly 14% — up from 6% in 2022. The direction of travel is away from staffed research functions, not toward them.
Meanwhile Maze's survey of 800+ researchers, designers and PMs found 63% naming time and bandwidth as their biggest challenge, and that studies are increasingly run by non-researchers: designers 70% and PMs 42% in 2025. And 13% of organisations offer non-researchers no support at all.
The conclusion is not "hire a researcher." Most mid-market companies will not, and for many the correct decision is that they should not. The conclusion is that research has to become a process the existing team runs, not a role you fill.
What to run: three standing studies and a triage path
Do not build a research programme. Build three studies that never turn off, plus a way to answer the urgent question when it arrives.
Standing study 1 — Why we win and lose. Every closed deal, won or lost, gets a short interview. This is the highest-ROI recurring study for any company with a sales motion, because it is the only one that pays for itself in a quarter. See win/loss analysis and, for the sales-side operating detail, customer research for B2B sales teams.
Standing study 2 — Why people leave. Every churn and every downgrade, plus the renewals that nearly did not happen. The near-miss renewals are the valuable half and almost nobody collects them. See customer renewal interviews.
Standing study 3 — What new customers expected. Interviews at 30 days, while the memory of the buying decision and the first-week friction is still intact. This is the study that catches the gap between what marketing promises and what onboarding delivers. See user onboarding research and aha moment research.
The triage path. For everything else — the exec question, the roadmap disagreement, the pricing decision — you need a route from "we do not know" to "we asked 20 customers" that takes days, not a quarter. That path is what most mid-market companies genuinely lack. Without it, every unplanned question gets resolved by opinion.
Three standing studies plus a triage path is a complete research function for a company of this size. It is not a compromise version of an enterprise programme; it is the right shape for the segment.
Who runs it
Not a researcher. In practice, one of three patterns works:
- A part-time owner. Someone in product ops, a PM, or a senior analyst owns the process — not the interviewing, the process: which studies are live, who reads the output, what happens next. Ten percent of one person's time.
- Distributed with guardrails. Anyone can launch a study from an approved template; a named owner reviews questions before they field. This is research democratization with the safety rail that keeps it from producing junk — and the guardrail matters, because bad questions are the main failure mode. Questionnaire design and The Mom Test are the two things to make everyone read.
- The first researcher, hired late and deliberately. If you do hire, hire someone who will build the system rather than personally conduct every interview. Our guide to research team structure covers the models, and the research team of one covers the reality of the job.
Whichever pattern, write down where findings live. A shared repository beats a folder of decks; see building a research repository and ResearchOps.
How to buy: the mid-market procurement trap
Mid-market buyers get squeezed from both directions, and the squeeze is predictable enough to plan around.
From above: enterprise research platforms sell annual site licences with seat minimums calibrated for organisations with a research department. You end up paying for ten seats so that two people can occasionally run a study — and the per-seat model is exactly wrong for a company whose research usage is bursty. We worked the arithmetic in per-seat vs usage-based pricing for research tools: a seat only beats metered pricing past roughly annual seat price divided by price per unit studies per person per year, which for most mid-market teams is a threshold they will never approach.
From below: self-serve survey tools are affordable and cannot do the job. They collect answers; they do not ask follow-up questions. The moment your question is "why," a form is the wrong instrument.
Four things to insist on at this size:
- Usage-based or credit-based pricing, so an occasional user is not a licence.
- No professional-services dependency. If launching a study requires a vendor solutions consultant, you have bought an enterprise product.
- Output a non-researcher can act on — a report, not a transcript archive.
- Structured measurement alongside open-ended depth, so findings survive a numerate audience.
For the two adjacent segments, best user research tools for startups and best customer research tools for enterprise teams show what the ends of the market look like — useful mostly for recognising when a vendor is selling you the wrong one. If you need to justify the spend internally, the user research cost calculator and stakeholder buy-in playbook do that work.
Where Koji fits
Koji is built for exactly this shape of organisation: real research needs, no research department.
AI-moderated voice interviews mean nobody on your team has to schedule, moderate, or transcribe. Interviews run in parallel, on the customer's own time, which is what makes a triage path measured in days rather than quarters possible at all. No moderator also means no moderator bias — a genuine methodological advantage, not just a cost saving.
Structured questions give findings the numeric backbone a CFO or a head of product will ask for. Six types — open_ended, scale, single_choice, multiple_choice, ranking, and yes_no — combine with open-ended conversation in a single study. See structured questions in AI interviews.
Automatic thematic analysis and one-click reports are the difference between research that gets read and research that gets archived. Nobody at a 400-person company has time to synthesise 30 transcripts by hand, which is precisely why, at most companies this size, the interviews quietly stop happening.
Customisable AI consultants let you encode your own domain — your market, your terminology, your product — so the interviewer probes like someone who understands the business.
And the pricing model fits the segment rather than fighting it: Koji does not charge per seat. Plans start at 29 euros per month for Insights and 79 euros per month for Interviews, credits are consumed per conversation (1 for text, 3 for voice), and only conversations that clear the quality bar consume credits at all. The PM who needs one study this quarter is not blocked by a licence, and the quarter where you run nothing does not cost you a full-price renewal.
Where to start
Pick the standing study whose absence is costing you most right now. For most mid-market companies with a sales motion, that is win/loss. For most product-led ones, it is the 30-day new-customer interview.
Run it for one quarter before you decide whether to build the other two.
Koji gives you 10 free credits when you sign up — enough to field a genuine study and see what your customers say when someone finally asks. No research expertise required, and from question to insight in hours rather than weeks. If you want the vocabulary before you start, the user research glossary covers 120 terms in plain language.
Frequently Asked Questions
What counts as a mid-market company?
The most widely used US definition comes from the National Center for the Middle Market: annual revenues between $10 million and $1 billion. That band covers nearly 200,000 US businesses, about one-third of private sector GDP, and roughly 48 million employees. The NCMM further splits it into lower ($10M to $50M), core, and upper ($100M to $1B) segments, which behave quite differently — upper middle market firms have been driving most of the recent revenue and hiring growth.
Do mid-market companies need a dedicated researcher?
Usually not, and often they should not hire one first. The better move is to make research a process the existing team runs: three standing studies plus a triage path for urgent questions, owned part-time by a PM, analyst, or product ops person. If you do hire, hire someone who will build the system rather than personally conduct every interview — otherwise you have bought a bottleneck rather than a capability.
Why does founder-led customer discovery stop working?
Not because founders stop talking to customers, but because the sample degrades. As the company grows, a founder's contact skews toward the largest accounts, the loudest complainers, and whoever sales puts in front of them. The volume of contact stays high while representativeness falls — which is why the problem is invisible from the top until something expensive ships and fails.
How much research can a mid-market company realistically run?
Three standing studies — win/loss, churn and downgrade, and 30-day new customer — plus roughly one ad hoc study a month is a realistic and genuinely sufficient programme at this size. That is achievable with about 10% of one person's time when interviews are AI-moderated, because the scheduling, moderating, transcribing, and synthesis steps that consume most of a traditional research budget are automated.
Should we buy an enterprise research platform?
Rarely. Enterprise platforms are priced and configured for organisations with a research department, which usually means seat minimums you cannot fill and a launch process that depends on vendor solutions consultants. Look for usage- or credit-based pricing, self-serve study creation, and reports a non-researcher can act on. A platform that requires professional services to launch a study is the wrong shape for a company of this size.
How do we prove the ROI internally?
Start with win/loss, because it is the only research programme that reliably pays for itself inside a quarter — you are attaching a reason code to revenue that already exists in your CRM. Once the sales leader is quoting interview findings in pipeline reviews, the budget conversation for the other studies stops being a debate about research and becomes a debate about coverage.
