Articles on growth, written for operators.

You have one. Every multi-unit operator does. The manager whose store just runs. Labor lands, shrink stays low, the team stays, the reviews are good, and when something goes sideways at another location, she’s the one you send. You’ve thought about cloning her. You’ve probably tried: put her in charge of training, had new managers shadow her for two weeks, asked her to write down how she does it. And the other stores still don’t run like hers.

That’s not a talent gap in the rest of your bench. It’s a transfer problem, and it has a specific cause.

Why "Just Write It Down" Produces a Folder Nobody Opens

Roughly 80% of the processes in a typical business are undocumented. That number gets quoted a lot. The part that matters more is what the documented 20% actually contains: the official process, not the real one. Documentation captures the workflow. Your best manager’s edge is the judgment she applies to the workflow, and those are different things.

Ask her to write down “how to run a store” and you get a bad manual, for the same reason a great chef asked to write down “how to cook” produces a bad cookbook. Expertise is stored in the situations where it gets used. She can’t narrate it cold. So the shared-drive folder gets built, the calendar invite goes out, the two-week shadow happens, and six months after she gets promoted or leaves, her successor has quietly rebuilt half of what she knew, badly, and is still learning the other half by breaking things.

Documentation captures the workflow diagram. Tribal knowledge is the judgment that gets applied to the workflow. Those are different substrates.

Verti

Median tenure in most operations sits around three and a half years. Knowledge isn’t walking out the door at retirement. It’s walking out continuously.

What Actually Gets the Judgment Out

The method that works is unglamorous: sit with her, one real case at a time, and ask why. Not “how do you schedule,” but “why did you pull Tuesday’s mid shift last week and not the close.” Not “how do you handle shrink,” but “why did you count the candy wall on a Wednesday.” Never let it float into the abstract.

A knowledge-capture team ran this with a senior insurance underwriter who could not describe her renewal process on a blank page. Case by case, asked why this file gets approved and that one doesn’t, she talked for an hour on every decision. Every reason came out. None of them were in her team’s SOPs.

That last step is where most operators skip ahead. Encoding one person’s habits into the system institutionalizes the quirks along with the judgment. The output you want is a working profile: the case types she sees, the reasoning behind each call, the exception library, the triggers that make her escalate, and where she looks first when a store feels off.

Documents Don't Run Stores. Routines Do.

Here’s the trap on the far side of the capture: producing a document and calling it done. A documentation project produces documents. What you need is something that runs, every store, every week, whether or not she’s in the building.

The delivery mechanism at store level is the routine, not the manual. A routine defines when the work happens, who does it, what sequence is followed, how completion is confirmed, and how exceptions escalate. That’s the form your best manager’s judgment needs to take to survive contact with a store she’s never seen.

One prerequisite worth being honest about: this works against whatever data she already uses to read a store. POS, labor, inventory, reviews. If you don’t have that layer, build it first. Buying software before the method exists is paying to digitize the chaos. The value split is roughly 70% method, 30% tool.

When to Do This

Most operators start the capture after a near-miss: she gives notice, or gets sick, or the store she left behind falls apart within a quarter. That’s the expensive version. The cheap version is doing it now, while she’s engaged, while the cases are fresh, and while the second and third best managers are around to validate what she says.

Three things worth un-learning:

• Promoting her preserves the knowledge. It moves the person. The judgment still lives in exactly one head, now further from the stores.
• Having her train everyone else transfers the system. Training moves explicit knowledge well and tacit knowledge poorly. Judgment has to be recorded in action, not lectured.
• Her method is the system. Until two or three other strong operators confirm it, it’s one person’s logic mixed with personal habit.

The opening date was set the day the lease got signed. Everybody agreed to it. There’s a checklist, a long one, and every line on it will get done. And the store will still open three weeks late, because the checklist never had the one thing that actually decides an opening date: who owns each handoff, what it depends on, and the date by which a decision has to be made.

Most openings run 90 to 180 days from lease signing to doors open. About 40% of them slip. When you go back and look at where the time went, it’s almost never the construction crew. It’s the space between teams.

Every boundary between teams is a risk point.

Hubler

A typical opening moves through roughly three dozen milestones and touches 15 to 20 people across real estate, legal, construction, IT, ops, HR, and the landlord’s side. Each time the work passes from one of them to the next, it costs two to five days on average. Five handoffs decide most of the schedule. Here they are, with what your checklist needs to say about each.

1. Lease to Pre-Construction: The Site Handover

A signed lease does not mean the site is ready for work. Delays start here because the premises condition, landlord obligations, access rights, permit status, and utility connections never got validated before the schedule got locked.

Treat the handover like a delivery you’re inspecting. Possession date. Loading access and delivery restrictions. Work-hour limits. Ceiling heights, floor condition, storefront condition. HVAC, electrical, and plumbing capacity. Fire and sprinkler coverage. Signage rights. Landlord-controlled areas. Have real estate, construction, ops, IT, and facilities all look at it, and get the answers in writing.

2. Design to Permits: The Third-Party Clock

Permitting is the single largest source of delay, for three reasons: it happens before construction can start, it depends on somebody else’s review queue, and it varies wildly by jurisdiction. The same tenant build-out clears in four to eight weeks in a fast suburb and takes eight to twelve or more in a major coastal city.

The control here is a permit tracker, owned by someone on your side, not assumed to be handled by the general contractor. Each required approval, who’s responsible, submission date, expected approval date, what it depends on, current status, and the trigger that escalates it. Building, electrical, plumbing, signage, health, fire, occupancy, food handling, business registration. Whatever your format needs.

3. Procurement to Construction: Long-Lead Equipment

Rooftop HVAC units and electrical switchgear have run twenty to forty weeks to procure in recent years. That breaks the old assumption that you order them once construction is underway. Signage and landlord deliverables round out the usual hiding places. All of these run on their own clock, in parallel with the build, which is exactly why they get missed.

4. Construction to Operations: The Readiness Gate

Two mistakes live here. Delaying the opening over a scratched wall, and opening with a failed fire inspection. The punch list needs to distinguish opening-critical defects from post-opening corrections, and it needs to do that before the final walkthrough, not during it.

Define who has authority to sign off on construction completion, IT, security, merchandising, operations, HR, compliance, and the final go or no-go. Then require evidence, not confidence: inspection certificates, test results, photos, training records, inventory reports, system screenshots, vendor completion forms. If leadership decides to open with something unresolved, the rationale and the mitigation plan get written down. That protects the business from informal risk acceptance.

5. Project Team to Store Team: Day One

The opening sequence itself is a chain: signage, final inspections, certificate of occupancy, punch list, merchandising, staff training, soft open, grand open. The last handoff is the project team handing the building to the people who’ll run it. Put store associates in the final walkthrough. They think in terms of daily use and catch what headquarters misses. It also gives them ownership before the doors open.

Then remember the opening isn’t the finish line. New stores typically take 12 to 24 months to reach maturity, and the first 90 days separate the ones that get there from the ones that stall.

Why It Keeps Happening

There’s no single source of truth. The project plan lives in someone’s spreadsheet. Vendor contracts are in email threads. The compliance checklist is in someone’s drawer. Finance doesn’t know the fit-out is over budget until month end. Ops doesn’t know the fire approval is pending until opening week.

Openings run by a project manager with a standardized playbook finish on average 43 days faster than ones run ad hoc. Not because the PM is a hero. Because every dependency carries an owner, a date, an evidence requirement, and an escalation path, so control survives whoever’s in the seat.

Three things worth un-learning:

• The checklist is the plan. A checklist without owners, dependencies, and decision dates is a document, not a control system.
• The general contractor owns permits. Permits run on a third-party queue. You need a tracker, not an assumption.
• Delays are construction’s fault. Most delays originate in the handoffs and the third-party clocks, not in how fast the crew works.

The corner is perfect. Good visibility, strong traffic, the right neighbors, a landlord who wants you. Everybody who walks it says the same thing. And you can still lose money there for nine years, because the address is only one of the questions a lease answers, and it’s the one operators spend the most time on.

The lease is the least reversible decision you’ll make. Staff, pricing, hours, product mix, all of that can change in a week. Square footage on a nine-year term can’t. One in four new establishments doesn’t reach its first birthday. Some of those were bad concepts. A lot of them were good concepts in the wrong box.

Two sites on the same street can differ by 30% or more in passing traffic because of a corner, a crossing, or a bus stop. So measure the address, not the neighbourhood.

MyTraffic

Six Checks, In This Order

The order matters. Cheap checks first, so you disqualify weak sites before you pay a lawyer or a surveyor. Any one of these can rule a site out.

Footfall at the door. Twelve months of it, not one. A street that lives off a summer market or a single office tower has a traffic profile that collapses for part of the year. Your rent doesn’t.

The real trade area. Where your customers actually come from is almost never a neat radius. It follows roads and transit and habit, and it stops dead at a river, a highway, or a rival cluster. Draw drive-time polygons, not circles. The primary trade area typically supplies 50 to 80% of revenue.

Demographics against your customer. Write your customer profile down before you look at the data, or you’ll talk yourself into whatever the site happens to show. Set pass-or-fail numbers: household income above a threshold at a minimum share of the trade area, residents within a ten-minute walk, enough daytime workers to carry a Tuesday.

The competitive set. Direct competitors inside the primary trade area, counted, not felt.

Accessibility. Visibility, ingress and egress, parking, transit.

The lease clauses themselves. Which is where good sites go bad.

Where Good Sites Go Bad

An office lease negotiates the box: square footage, rent, term. A retail lease negotiates the ecosystem: who else is in the center, how traffic reaches your door, whether you can grow nearby, and what share of your sales the landlord takes on top of rent. Those provisions can’t be unwound at renewal, so they deserve disproportionate attention now.

Done well, percentage rent aligns you and the landlord. Done poorly, with a low breakpoint and a broad definition of sales, it makes the lease far more expensive than the base rent suggests. Get the property, finance, operations, and legal reads before the lease process starts, in a deal memo, not a phone call.

The Question That Decides the Deal

If you already operate nearby, there’s a check the six above don’t cover. A new location posts real revenue. If a chunk of it walked over from your store three miles away, the brand didn’t grow. It redistributed.

Score all of it against your best existing stores. If you can’t explain why your top performers work in these terms, you can’t tell whether the new site will.

Four things worth un-learning:

• A great address is a great site. Measure the door, not the neighborhood.
• The rent is the rent. Percentage rent, CAM, exclusives, and co-tenancy are the real occupancy cost.
• Strong footfall means strong sales. Traffic that converts matters. Volume without fit is a trap.
• A busy summer street is a good site. Test twelve months, not the peak.

A candidate is six weeks into your pipeline and everything has gone right. Discovery Day landed. The unit economics conversation felt honest. They’re ready to sign. Then their attorney finishes the FDD review and the deal goes quiet. Not because the numbers in Item 19 were bad. Because they weren’t the numbers the candidate had already been told.

That gap has two costs. One is the deal. The other is legal, and it’s bigger than most sales teams understand.

Three Conditions, All Required

Item 19 is the only place in the franchise sales process where a financial performance representation is allowed to exist. The definition is broad: any oral, written, or visual representation that states or implies a specific level or range of sales, income, or profit. Under the FTC Franchise Rule, making one is unfair or deceptive unless three independent conditions are met: you have a reasonable basis for it, you have written substantiation at the time you make it, and it’s in Item 19.

A representation can be perfectly accurate and still unlawful because it is not in Item 19.

HDD Law Firm

Read that twice. Accuracy isn’t a defense. A rep who says “our strongest stores tend to run higher” has just made an earnings claim, whether or not it’s true. The rule reaches oral and visual statements specifically, which means sales conversations, slide decks, and webinars. And it reaches brokers: anyone under contract and compensated on sales is a franchise seller under FTC guidance, and their projection is covered the same as yours.

How It Actually Happens

Nobody sits down and decides to break this rule. It happens over a six-to-twelve-week relationship as a rep fields the same question again and again and starts filling the silence. “That number’s a system average, yours would probably do better in a market like this.” None of that is in the FDD. All of it is now what the candidate believes.

That last point is the trap in a blank Item 19. Nobody buys a franchise without forming a view of what it will earn. If Item 19 is empty, that view came from somewhere: a conversation, a pro forma emailed during diligence, a figure mentioned at Discovery Day, a loan projection someone helped prepare. A blank Item 19 doesn’t stop the number. It pushes it into channels you don’t control.

What the Case Law Says

The franchisor sometimes wins these. In Sellstate, a Minnesota court dismissed because once the franchisees read the disclaimers, they couldn’t reasonably rely on the oral projections. Rocky Mountain Chocolate Factory, Ben & Jerry’s, and Kiddie Academy all won on similar grounds. But outcomes depend as much on which state you’re in as on what happened. New York and others have anti-waiver laws that don’t let franchisees sign away those rights. Treating the disclaimer as a shield is betting your system on venue.

And winning the case still means you lost the deal, the legal fees, and a franchisee who’s now telling every candidate in your pipeline what happened.

The Fix Isn't Silence

Telling your sales team to stop talking about money doesn’t work. Candidates will ask regardless. The fix is giving them a specific, compliant answer that maps directly to whatever Item 19 says this year, so there’s nothing left to freelance.

If your FDD makes no claim, that answer is short and it gets memorized word for word, not improvised under pressure. If it does make one, the team needs the plain-English version of exactly what it covers and, just as important, what it deliberately leaves out: which markets, which unit ages, which outliers were excluded. The deliberate gaps are exactly where reps guess. Review that mapping every time Item 19 changes, not once when it’s written.

Three things worth un-learning:

• The numbers were accurate, so we’re fine. Accuracy is one of three conditions. All three are required.
• Oral statements don’t count. They’re named in the definition, and so are slides and webinars.
• Our disclaimers make us untouchable. They’ve won cases. They’ve also lost them, and they never win back the deal.

Sales are climbing. You’re busier than you’ve ever been. And somehow, everything takes longer, costs more, and breaks more often than it did a year ago when you were smaller. If that sounds familiar, you’re not imagining it, and you’re not failing. You’re running into a well-documented pattern: growth is a stress test, not a reward, and most businesses fail the test at a predictable point.

The infrastructure that supports five people rarely survives fifty, and the systems that worked at $2M in revenue tend to collapse somewhere between $10M and $20M.

The dangerous part is the assumption underneath it: if the numbers are going up, the business must be working. It isn’t necessarily. Here are the seven signs that tell you the operating system, not the market, is what’s actually failing.

1. Decisions Stall Because Nobody Actually Owns Them

Something needs a call. It sits. Not because anyone’s negligent, but because it was never clear whose call it was to make. This is one of the clearest tells that you’ve outgrown informal coordination. In a five-person team, everybody just knows who decides what, because you’re all in the same room constantly. At thirty or fifty people, that shared awareness quietly disappears, and nothing has replaced it.

2. Your Best People Are Burning Out Fixing What Should Be a System

If your strongest managers are the ones constantly stepping in to catch things that keep almost falling through, that’s not a compliment to their work ethic. It’s a warning that you’re running on heroics instead of process, and heroics are not a scaling strategy. A DevOps industry survey found 47% of engineers directly tie their burnout to exactly this pattern: relying on someone heroic instead of fixing the underlying design.

3. Every New Hire Costs More Than the Last One

If onboarding is “shadow someone for a week and figure it out,” your hiring costs are multiplying quietly with every person you add, because nothing compounds. A new hire without structured onboarding takes about 6.5 months to reach full productivity. With it, Brandon Hall Group found structured onboarding improves retention by 82% and productivity by more than 70%. That gap is entirely a documentation problem, not a talent problem.

4. Nobody Remembers Where Anything Is Written Down, Because It Isn't

This is the single point of failure problem in plain terms: if a process only lives in one person’s head, that person leaving, getting sick, or getting promoted means the work stops or degrades. It’s not a hypothetical. It’s the default state of most growing businesses, and it’s fixable in weeks, not years.

5. Leadership Spends the Week Fighting Fires Instead of Building Anything

If your calendar (or your best manager’s calendar) is full of status updates and putting out the same fire for the third time this month, that’s not a busy week. That’s a business where communication has become a design problem that nobody’s designed yet.

6. Someone Is Running More Locations Than the Math Actually Supports

There are real numbers here, and they’re worth knowing cold: a general manager runs one location well. An above-store leader can hold three to five (six, if they’re close together and performing well). A regional director can hold four to six districts. Past those numbers, the leader stops leading and starts triaging, and everything underneath them starts to drift without anyone quite noticing why.

7. Management Turnover Is Climbing Even While Hourly Turnover Looks Fine

Hourly turnover has actually been improving industry-wide. Management turnover is the part that’s gotten stickier, and it’s the expensive part: replacing a GM runs roughly $17,651 in hard costs and takes 42 days to fill. If your best assistant managers keep leaving, that’s not a hiring market problem. It’s usually a sign the layer above them isn’t built well enough to support them.

Where the Org Chart Should Actually Change

Restaurant and retail org structure research points to specific inflection points: roughly 3, 7, 15, and 30 locations. Strong GMs come first. Above-store leadership gets added around 5 to 7 units. Functional support (ops, marketing, finance, people) comes in around 15. Regional structure shows up around 30. Skipping a stage doesn’t save money. It just moves the cost into variance, turnover, and firefighting, where it’s harder to see and harder to fix.

The Fast Way Almost Always Loses

There’s a well-known argument for growing fast and fixing the mess later, sometimes called blitzscaling: speed matters more than efficiency in winner-take-most markets, so tolerate the chaos and clean it up once you’ve won. The data on this is fairly blunt. Across a dataset of 3,200-plus high-growth startups, premature scaling was present in 70% of the failures, and companies that scaled in a disciplined, sequenced way grew roughly 20 times faster than the ones that scaled prematurely. The surprising part: disciplined scalers didn’t end up smaller. They ended up with teams 38% bigger at the same stage, because their growth was correctly sequenced instead of just fast.

The Reset That Preceded Growth, Twice

Starbucks spent nearly two years fixing throughput and service consistency before committing to its next wave of store growth. Only once average service time across café and drive-thru dropped under four minutes did leadership commit to roughly 400 net-new U.S. stores. Their own framing: ”Growth doesn’t require us to become something new, it requires us to be exceptionally good at who we already are.”

Chipotle is the cautionary version of the same lesson. 2025 was its worst year as a public company, same-store sales fell for the first time since 2006, largely traced back to speed and throughput problems that never got fixed before the company kept expanding into new formats like catering.

Two things worth un-learning:

• Growth doesn’t prove your foundation is solid. It’s a stress test of it, and stress tests find the weak point.
• Technology doesn’t fix a process problem. Most small businesses now use AI tools, but only 14% have them actually embedded in how work gets done. Tools without a workflow underneath them don’t compound.

You’re the person everyone calls. Vendor issue, staffing crisis, a customer complaint that escalated, a lease question, someone always ends up in your inbox or your voicemail. You’ve gotten good at it. Fast decisions, right most of the time, nobody has to wait on you long. It feels like strength. It’s also, without you necessarily noticing, the reason your business can’t grow past a certain point, and the reason it’s worth less than it should be if you ever sell it.

Why the Business Slows Down Exactly When You Get Busier

There’s a real ceiling here, and it shows up at a predictable size. Founder-led businesses typically hit the wall somewhere between $2 million and $15 million in revenue, not because the market changed, but because the founder became the operating system. At ten employees, being in every loop is fast and effective. At thirty, it’s manageable. At fifty, it’s the bottleneck. At a hundred, it’s actively working against growth.

One consultancy ran a decision audit on a real business (an $8 million luxury goods company, 65 employees) and tracked every decision the founder made over two weeks: 347 of them. When they broke those down, 41% were routine approvals that followed a clear rule and could be automated. Another 34% were judgment calls that could be delegated with the right framework. Only 25% actually required the founder specifically.

What This Actually Costs You at Exit

If you plan to sell someday (and 75% of owners say they want to exit within ten years), this isn’t a soft, someday problem. It’s priced, right now, into what a buyer will pay.

Enterprise value is not what your business earns. It is what a buyer will pay to own the cash flows after you are gone.

Buyers interview your team, your top customers, your vendors, during diligence. If every answer traces back to you, they start discounting on the spot. Depending on how they frame it, that shows up as a straight 10 to 25% valuation cut (up to 50% in extreme cases), or as a multiple compression from the 7 to 8x a founder-independent business commands down to 3 to 4x, or as an earnout: they’ll pay you over 24 months instead of at close, tied to the business surviving without you. Worth knowing before you agree to one: on average, only about 21 cents of every earnout dollar promised actually gets paid.

"Just Delegate More" Doesn't Work, and Here's Why

The instinct is to hire good people and get out of their way. Brian Chesky tried exactly that at Airbnb and called it, in hindsight, disastrous. The problem wasn’t the people. It’s that delegation doesn’t scale when the decision-making logic still lives only in the founder’s head. You can hand someone the authority to decide, but if they don’t have the reasoning you use, you’ve just moved the bottleneck, not removed it.

The actual fix has a name: decision architecture. It means classifying the decisions your business makes, assigning clear ownership to each type, setting thresholds for when something needs to escalate, and then tracking how long decisions take and how often they get reversed. It’s less romantic than “hire great people and trust them.” It’s also what actually works, because it gets your judgment out of your head and into a system other people can run.

The Test That Tells You Where You Stand

There’s a simple, uncomfortable way to find out how dependent your business actually is on you: leave for thirty consecutive days. Take no calls. Don’t delegate anything in advance beyond what’s already written down. Then look at what broke.

Most owners are afraid to run this test because they already suspect what they’ll find. That’s exactly why it’s worth running.

Two Ways to Build a Business That Doesn't Need You

Greg Flynn started with eight Applebee’s in Seattle in 1999. Flynn Group now runs roughly 3,000 units and employs 75,000 people, and Flynn is explicit about why: ”What matters is that we’ve built a business with a reason to exist beyond any one individual, one that can keep growing, evolving and creating opportunity.”

Chesky went the other direction after Airbnb’s COVID collapse: he re-centralized hard, redrew the org chart, and personally ran product decisions again. Revenue went from an 80% booking collapse to $6 billion within a year and nearly $10 billion two years later. That’s not a contradiction of the founder-independence argument. It’s the same idea from the other side: Chesky rebuilt as the deliberate designer of the decision system, not as the answer to every decision inside it. Both operators ended up in the same place. The founder decides what the rules are. The founder stops being the rule.

A few things worth un-learning:

• Being indispensable doesn’t make your business more valuable. It’s priced as a discount, not a premium.
• Documentation alone doesn’t fix this. SOPs are worth something (roughly a 0.25x to 0.5x multiple bump), but buyers want two years of proof that someone other than you has actually run the thing, not a binder.
• Working harder in the business isn’t the path out. Decision demand scales with headcount. Your personal capacity doesn’t.

You did the hard part. Unit one, you built with your own hands. Unit two and three, you ran on hustle: eighteen-hour days, personally training every manager, personally answering every 11pm call. It worked. Now you’re staring at unit four or five, and something’s off. It doesn’t feel like the business is compounding. It feels like it’s fraying. Margins are softer than they were. Turnover’s up. You’re more tired for less result than you were a year ago.

You haven’t done anything wrong. You’ve hit a stage that has a name, and it hits almost everyone at almost the same spot.

The Dead Space Between Three and Five Units

Consultants who study multi-unit operators call the zone between roughly three and five units the “dead space.” Below three units, one hands-on owner can cover everything. Above five or six, you’ve usually built (or been forced to build) enough management layer to run the business without being in every room. In between, you’ve outgrown what you can personally do, but you haven’t built what should replace you yet.

The greatest risk in multi-unit franchising is not growing too quickly. It is becoming trapped between operator scale and enterprise scale.

Mellon Franchising

That’s the trap in one sentence. It’s not that you grew too fast. It’s that your management infrastructure didn’t grow with you.

The Unit-3 Failure Mode

Here’s exactly how the dead space breaks businesses, and it’s specific enough to recognize. An owner hires a GM for unit one, runs unit two personally, and opens unit three at the same time. What happens next is predictable: unit one’s revenue dips 8 to 12% during the unit-three build-out because the owner’s attention is split. Unit two, which never got the owner’s full focus to begin with, plateaus 10 to 15% below what it should be doing.

Nobody did anything wrong. The math just didn’t work, because one person was doing three jobs at once.

The fix is boring and it’s the whole point: hire the area manager before you open unit three, not after you feel the pain of not having one.

Unit two typically doesn’t turn profitable until 12 to 18 months after opening, and during that stretch it needs $100,000 to $200,000 in working capital, usually pulled straight out of unit one’s cash flow. That’s a dependency loop, and it’s exactly why the timing of your hires and your financing matters more than the timing of your leases.

Two Camps, One Shared Truth

There’s a real debate in this space, worth knowing so you can place yourself. One camp argues you should accelerate through the dead space, because additional unit count is what generates the revenue base that eventually pays for the management layer. Slowing down just prolongs the pain. The other camp argues the opposite: build the systems and leadership readiness first, because roughly 70% of multi-unit development agreements stall out when leadership and systems aren’t in place before the growth commitment is signed.

Both camps actually agree on the variable that matters. It isn’t speed. It’s whether organizational infrastructure is being built in step with unit count, or lagging behind it. Fast growth with the infrastructure in place works. Fast growth without it produces exactly the unit-3 failure mode above, just at a bigger scale.

The Money Got Easier in 2026

One thing changed this year that’s worth knowing if you’re financing the next few units: the SBA doubled its combined 7(a) and 504 lending cap from $5 million to $10 million, effective July 2026. That means a franchisee can now hold $5 million in 7(a) operating debt and $5 million in 504 real estate debt at the same time, a structure that wasn’t possible before. Lenders are also increasingly building portfolio-level structures where unit one’s performance pre-approves financing for unit two on better terms. If your unit-1 numbers are strong and documented, you have more leverage right now than operators had two years ago. Use it.

What Building Ahead of Need Actually Looks Like

Jim D’Aquila, who built a multi-brand portfolio (Shipley Do-Nuts, Saigon Hustle, Sweet Paris) starting at age 62, put it plainly: ”We made a deliberate choice to build systems and infrastructure ahead of need. From our tech stack to our corporate structure to our talent bench, everything is designed for where we will be in five years, not just where we are today.” That’s not a personality trait. It’s a purchasing decision, made before the pressure forces it.

Three Things That Aren't True

”A strong first unit proves the model.” It proves the concept works once. It says nothing about whether the model works when you’re not personally in the building.

”Unit one’s cash flow will fund expansion.” It will, until it can’t. That dependency loop is exactly what strands operators mid-growth.

”The next ten will be as hard as the last three.” The data says the opposite. Growth compounds once the infrastructure exists. The hard part is building the infrastructure, not sustaining the growth after.

Every multi-unit operator I talk to has been pitched an AI tool this year. A bot for the operations manual. A drive-thru voice. A lease reader. A candidate screener. The pitch always answers the same question: what can this do?

That’s the wrong question. The right one is: what will this amplify?

The Oldest Rule in Business Technology

There’s a line usually credited to Bill Gates that every operator should tape to the wall before a software demo: “Automation applied to an efficient operation will magnify the efficiency. The second is that automation applied to an inefficient operation will magnify the inefficiency” (IT History Society). Even the attribution is shaky (the archive that carries it lists the source as unknown), but the rule has held up for decades, and AI is the strongest test of it yet.

The evidence so far points one way. McKinsey’s 2026 State of AI survey found that about 6% of companies saw significant financial impact from AI, almost exactly the same share as the year before, despite heavier adoption. The difference between that 6% and everyone else wasn’t which model they bought. About three-quarters of the top performers had fundamentally redesigned their workflows, compared with about a quarter of everyone else (McKinsey via WSJ Partners). McKinsey’s own summary is blunt: “Value flows only when organizations redesign their operating models to capture the new technology’s promise” (McKinsey).

You’ve probably also seen the claim that “95% of AI pilots fail.” Be careful with it, for the same reason you should be careful with the old line that 80% of franchises succeed. The MIT report behind it actually found that around 5% of custom enterprise AI tools reached production with sustained impact, drew on 52 interviews and 153 conference surveys, and was never peer reviewed (People Team AI). What’s worth keeping is its explanation: tools stalled because of “brittle integrations misaligned with day-to-day operations,” and the divide “does not seem to be driven by model quality” (AI Wiki summary of MIT NANDA). Different number, same lesson.

What It Looks Like in Our World

The drive-thru is the cleanest experiment our industry has run.

McDonald’s tested IBM’s automated order taker in more than 100 restaurants and shut it down in 2024 after it struggled to understand customers and get orders right (Yahoo Finance). Yum slowed its voice AI rollout across about 500 restaurants in August 2025 after complaints and the now-famous order for 18,000 cups of water (Restaurant Dive).

Now look at Bojangles. Its voice system runs in 452 locations, needs a human only 1 to 2% of the time, and cut drive-thru workload by 35%, according to the company. The detail that matters isn’t in the headline. The rollout came after six months of preparation, and the team added around 550 unique menu items to the system in the first month of testing (Franchising.com).

Read that again. The hard part was writing down, exactly, every way a customer can order. That isn’t AI work. That’s operating system work. Bojangles did it first, and the AI worked.

AI Is a Stress Test, Not a Fix

Here’s the pattern I’d expect in most multi-unit companies. Put a chatbot on top of the operations manual and ask it how to close the store. If three versions of the closing procedure are floating around (the binder, the district manager’s version, and the way the best store actually does it), the bot will pick one and deliver it with total confidence. You haven’t fixed the inconsistency. You’ve made it faster and harder to see.

The research on knowledge workers shows why that’s dangerous. In the Harvard and BCG study of 758 consultants, AI made people faster and better on tasks it was suited to. On a task outside its range, accuracy dropped from 84.5% without AI to 70.6% with it, and to 60% for the group that also got prompt training. The researchers’ warning: “polished output is not the same as sound reasoning” (Harvard Business School Digital Data Design Institute).

There’s a second trap: AI speeds up one step and exposes the next one. A study of more than 100,000 developers found AI coding tools raised coding activity by 180%, but software releases rose only 30%. The bottleneck moved to the human steps of reviewing, integrating, and releasing (Knowledge at Wharton).

Translate that to our four lanes:

A lease abstract in 30 seconds doesn’t help if no one owns the renewal date. A beautiful site summary doesn’t help if the approval meeting has no criteria. MRI Software, which sells lease AI, makes the same point: success depends on “the controls and processes to verify its outputs” (MRI Software).

Where the Other Side Is Right

The strongest counterargument deserves a fair hearing, because parts of it are correct.

First, AI can be the thing that finally forces the process work. Bojangles’ six months of preparation was process repair. Arguably the AI project was the only reason it got done. If a tool purchase is what finally makes a company write down its menu, its closing procedure, or its lease calendar, that’s a real benefit.

Second, AI can capture know-how the company never documented. In a study by economists Erik Brynjolfsson, Danielle Li, and Lindsey Raymond of 5,179 customer support agents, an AI assistant raised productivity 14% on average and 34% for newer, less-skilled workers, with “suggestive evidence” that it spread the practices of the best performers (Stanford SIEPR). In other words, the tool did some of the operating system’s job.

Third, waiting for a perfect process is its own way to stall. The MIT report found that tools built for specific processes could deliver value “even without major organizational restructuring” (AI Wiki summary of MIT NANDA). And Taco Bell didn’t quit after the parent company’s stumble. It kept expanding voice AI through a separate partner, Omilia, which now runs in more than 890 U.S. Taco Bells (Restaurant Dive). Learning in public worked for them.

All fair. But notice what each example has in common: the value came when the company treated the project as process work, not as a software purchase. That’s the argument, not a rebuttal of it.

Five Questions Before You Buy

Before any AI tool touches your operation, ask:

1. Is there one written, current version of this process, and does someone own it?
2. If a new hire asked two of your best managers, would they get the same answer?
3. Who checks the AI’s output, and how quickly would an error get caught?
4. What happens when the situation isn’t covered? Is there a human path?
5. Do you know how the process performs today, so you can tell whether the tool made it better?

If you can’t answer the first two, you don’t have an AI problem yet. You have an operating system problem, and the AI will find it for you, at scale and in front of your customers.

That’s where I land. If AI can strengthen an operating capability, use it. If it just automates a bad process, don’t. AI doesn’t install an operating system. It audits the one you already have.

When a new technology arrives, most companies reach for the same first move: train people.

It’s a sensible instinct. Nobody wants managers pasting questions into a chatbot with no idea how it fails. But training answers one question, whether people know how to use the tool, and skips the one that decides the outcome: is the work set up so the tool can be used safely?

Those are different questions. Treating them as one is how a team can complete the course and still hand an AI assistant a process nobody owns.

Two Words, Two Problems

Here are the plain-language versions.

AI literacy is what a person knows: what the tool is good at, how it gets things wrong, and how to ask it for something useful.

AI readiness is a property of a specific piece of work. It asks whether the procedure is written down, someone owns it, someone checks the output before it reaches a customer, and there is a place for the cases the tool shouldn’t handle.

One lives in a person’s head. The other lives in the operating system. You can raise literacy for everyone and still have a workflow that isn’t ready.

What The Evidence Does And Does Not Say

Start with the finding that surprised researchers. In two Aalto University studies, 246 and then 452 participants used AI to solve logical-reasoning problems from the LSAT. Their scores were 3 points higher than a comparison population, but they overestimated their own performance by 4 points. Higher AI literacy was correlated with less accurate judgment of their own results: more technical knowledge went with more confidence and less precision. Aalto University

Read that carefully. It’s a correlation, not proof that learning about AI makes people overconfident. It covers one kind of task, logic puzzles. It is worth taking seriously because it points at a gap: knowing more about the tool didn’t make people better at noticing when its answer was wrong.

A second study shows the gap in a work setting. Harvard researchers and Boston Consulting Group randomly assigned 758 consultants to no AI, GPT-4, or GPT-4 plus a brief prompt-engineering overview. On a creative product task, AI users completed 12.2% more subtasks, worked about 25% faster and received quality ratings about 32% higher. On a brand-strategy case that required spotting a subtle clue in interview notes, accuracy was 84.5% without AI, 70.6% with it and 60% with the added overview. Harvard Business School

The study’s own explanation was over-reliance: people trusted the analysis and followed it to the wrong conclusion. Harvard Business School It also found that professionals had “no obvious signal” telling them which tasks sat inside AI’s strengths and which outside. Harvard Business School

That last point is the readiness argument. If individuals cannot reliably tell when the tool is out of its depth, you cannot depend on each person’s judgment in the moment. The check has to be built into the work.

The limits matter here too. It was one study, two assignments, consultants rather than store managers, and the extra instruction was a short overview, not a serious training program. The published summary also doesn’t say whether the gap between the two AI groups is statistically reliable. The safe reading is that a little instruction did not protect people on the hard task. It does not show that training is useless.

Your People May Not Be The Bottleneck

The other half of the evidence is almost an opposite warning. McKinsey’s 2025 workplace study concluded that employees were more ready than leaders assumed: it found three times as many employees were using generative AI for a third or more of their work as leaders imagined. McKinsey

If that holds in your business, many of your managers are already using AI. A training session isn’t the starting point for a skill gap; it is catching up with behavior that exists. The decision that remains is where AI use is allowed to touch customer promises, lease dates, candidate answers or site decisions.

Readiness answers that. Literacy doesn’t.

Five Questions For One Workflow

Pick one workflow, not the whole company. For example, answering field questions from the operations manual, tracking opening checklist items, summarizing a lease, or responding to prospective franchisee questions.

1. Source: Is there one current, approved version of the procedure or information the tool will use?
2. Owner: Can you name the person who corrects the source when it changes?
3. Check: Who reviews the output before it affects a customer, landlord, candidate or opening date, and how quickly?
4. Boundary: What must the tool not answer, and where does that case go?
5. Baseline: Do you know how long the work takes and how often it goes wrong today?

If you can’t answer one or two, the workflow isn’t ready. That is not a reason to ban AI. It is the work to finish before you scale it.

Where The Other Side Is Right

Literacy matters, and the case for it is stronger than a simple contrast makes it seem.

A person who doesn’t understand how these tools fail will not design a good check. Someone has to know why a polished answer can be wrong, and that is a form of literacy. A workflow can look ready on paper and still fail if the people running it rubber-stamp the output.

The evidence on training is thinner than the evidence on overconfidence. The Aalto result is an association, not a test of training (Aalto University), and the Harvard study tested a brief overview, not a full program (Harvard Business School).

The tools are also changing. The research group METR found in 2025 that experienced developers took 19% longer with AI, with a confidence interval of +2% to +39%. In February 2026 it said developers were likely more sped up by early 2026 than that estimate suggested, although its newer data was too unreliable to size the effect. METR Studies of how people misjudge AI can age quickly.

Some regulators have also decided literacy is worth requiring. The EU AI Act, as amended in 2026, tells providers and deployers of AI systems to take measures to support their staff’s AI literacy, while saying this does not require guaranteeing any specific level for an individual. EU AI Act National authorities began supervising Article 4 on August 2, 2026. European Commission That matters only if you operate in the EU. It does show that literacy is easy to require and measure, which may be one reason companies reach for it first.

The practical position is both: train people, and don’t mistake the training record for readiness.

Most lists of AI use cases are lists of tasks: write the email, summarize the lease, answer the question.

A better place to look is where work sits still.

In a multi-unit business, a lot of delay doesn’t happen while someone is working. It happens between people: a site package waiting for review, a permit waiting for a response, a field visit waiting for follow-up, a candidate waiting for a call. The work didn’t get harder. It got stuck at a handoff.

That is where AI is most likely to pay off, and it is also where it is easiest to waste money if you skip the basics.

The Time Is In The Gaps

McKinsey calls the cost of human handoffs between workflow steps the “coordination interface tax.” It reports that coordination (aligning, verifying, reconciling, waiting and handing work along) consumes 35 to 60 percent of total work time in knowledge-intensive organizations. McKinsey

Its worked example is a single planning workflow at a company it does not name. The actual work inside the steps took 12 to 24 hours. The waiting between steps took 9 to 18 days, which McKinsey says is nine to 36 times longer. McKinsey

Treat that as a pattern, not a benchmark. It comes from a consulting firm that sells workflow redesign, and the case is an industrial planning process, not a store opening. But the pattern will be familiar to any operator who has watched a ready project wait on one signature.

Store openings show it in our world. One industry source estimates 2 to 5 days per handoff and 15 to 20 stakeholders touching a typical opening (Consultevo), and a franchise software vendor claims coordination overhead alone adds 4 to 6 weeks (Franly). Those are vendor figures, so use them as prompts to measure your own, not as facts to quote.

Speed Up One Step And The Next One Shows

The software industry has the cleanest evidence of what happens when AI accelerates one step. A study of more than 100,000 developers found AI tools raised coding activity by a cumulative 180%, but software releases rose only 30%. The authors said the constraint appears to be shifting from writing code to reviewing, integrating and distributing it. Knowledge at Wharton

A lease summary produced in 30 seconds is the same kind of speed-up. If the committee that approves the site meets monthly, or no one owns the renewal date, the faster summary doesn’t make the project faster. It makes the next waiting point easier to see.

A Handoff Map For The Four Lanes

For each lane, find the handoff, then decide what AI can do and what has to stay human.

The development row follows the handoffs that commonly hold up an opening: lease to pre-construction, design to permits, procurement to construction, construction to operations, and project team to store team.

One Documented Example, And A Warning

Chipotle’s hiring tool is the clearest restaurant example of the pattern. Chipotle says its “Ava Cado” assistant cut the time from application to first day from 12 days to four. Fortune The tool answers candidates’ questions, collects basic information and schedules interviews. Chipotle’s chief human resources officer said the manager interview remains the opportunity to judge whether a person fits the company’s values. HCAmag

That is a clean split. The tool handles the waiting. The manager keeps the conversation. Chipotle’s operating chief also said the hours saved were reinvested in the restaurant. Fortune

Two cautions. These are company-reported results, and it is a hiring example for restaurant employees, not a franchise-candidate example. Use it as evidence of the pattern, not as a promise for your own candidate pipeline.

Contract review carries a different warning. One roundup of benchmarks reports that a model scored above 0.9 on easy clauses like governing law and parties but near zero on the notice period to terminate a renewal. The roundup also notes that dataset performance is not the same as live review, and that aggregate scores can conceal poor results on important clause types. Legal AI Insights Those tests used general commercial contracts, not retail leases, and tools change. Still, the clause that most affects a tenant’s next decision can be the one the tool is weakest at, which is exactly why a person must verify critical dates.

Where The Other Side Is Right

The strongest objection is that the biggest gains aren’t between steps at all. They are inside one step, at the front line.

Bojangles says its drive-thru voice system needs a person only 1 to 2% of the time and cut employees’ drive-thru workload by 35%. Franchising.com Those are company figures, but they describe a task done faster inside a step, not a handoff between teams. If you only look for gaps between steps, you could miss that kind of gain.

The objection also has a point about the source of the handoff argument. McKinsey’s framing comes from a firm that benefits when clients redesign workflows. The software evidence comes from developers, not store operators. The honest summary is that the pattern is plausible and well illustrated, not proven for multi-unit retail.

There’s also a human cost to removing handoffs. Some of them are where trust gets built: the call between a franchise candidate and a development leader, or the district manager’s visit. Automating that moment because it is slow may remove the thing the business was selling.

So there are two places to look: the repetitive task inside a step, and the wait between steps. Both can pay. The mistake is starting with the tool instead of the delay.

Measure The Wait First

Pick one handoff and track ten recent cases. Record when the work arrived, when someone started it, and when it was done. Compare the time someone spent working with the time the item sat waiting.

Before adding AI, check that the handoff has four basics: one named receiver, a shared definition of done, a timestamp and an acknowledgment. Quire

If the biggest wait is a missing owner, AI won’t fix it. If it is routine information gathering, summarizing or routing that a person has to check anyway, you have a reasonable first test.

Where AI shortens a wait and a person still checks the result, keep it. Where it produces a faster summary that nobody acts on, it is decoration.

Think about the manager everyone calls when a store has a problem the manual doesn’t cover.

She notices that a labor problem is really a training problem. She asks the question that changes how a customer complaint gets handled. She knows when a new manager needs coaching and when the issue needs to move up the chain.

Now imagine asking her to help build an AI assistant so the other locations can get some of that judgment without calling her.

The business case is clear. The question worth asking is what changes for her.

If the system gets better because her knowledge travels further, but her role, recognition and workload stay exactly the same, the technology may be ready before the employment bargain is.

Who Benefits From The Knowledge?

One of the strongest workplace studies makes this question concrete. Economists Erik Brynjolfsson, Danielle Li and Lindsey Raymond studied an AI assistant used by 5,172 customer-support agents; the published study found a 15% average increase in issues resolved per hour. The Quarterly Journal of Economics

The average hides the important part: the lowest-skilled group improved by 36%, while the highest-skilled agents had no significant increase in issues resolved per hour and small declines in resolution rates and customer satisfaction. The Quarterly Journal of Economics

The system had been trained on customer conversations, with greater weight given to conversations handled by top performers. The Quarterly Journal of Economics

In other words, the people whose work helped make the assistant useful were not necessarily the people who benefited most from using it. The Quarterly Journal of Economics

That does not prove a chatbot can reproduce a great general manager’s judgment. Customer-support conversations are not store leadership. It does establish a mechanism worth testing: examples of strong performance can help less-experienced people handle situations better.

The authors also raise the incentive question directly: “compensation policies that provide incentives for people to contribute to model training could be important.” The Quarterly Journal of Economics

For a multi-unit operator, that turns “How do we capture her knowledge?” into a second question: “Why would she want to help us keep improving it?”

Capture Decisions, Not Personality

The opportunity isn’t to upload every document the manager has written or record everything she does. It is to identify decisions other managers struggle with and make the reasoning behind good decisions easier to use.

Take an illustrative field question: “Sales are down. Should we cut the schedule?”

A useful answer needs context. Is traffic down, or are fewer visitors buying? Which hours are affected? Are the people on the floor trained? Would cutting coverage make conversion worse?

The valuable knowledge is not simply “keep two people on Saturday.” It is what to check before changing coverage, what conditions would change the answer, and when the decision requires approval.

That gives you a practical capture method. Start with an actual case and ask:

• What did you notice first?
• What explanations did you rule out?
• What information would have changed your decision?
• What did you do, and what happened afterward?
• Where would another manager need to stop and ask for help?

Then have another experienced manager review the reasoning. One vendor guide to capturing expertise recommends asking “why” questions against real cases rather than abstract ones, and validating the result with two or three experts so personal habit is not mistaken for operating logic (Verti). Treat that as vendor advice, but it matches common sense: a successful store’s habit is a candidate for the standard, not proof of it.

AI can help organize those cases, retrieve the relevant example and explain an approved approach. That does not require building a custom model. A small, reviewed case library may be the right starting point. The test is whether another manager makes a better decision—not whether the answer sounds like the original expert.

The Incentive Is Operating Work

Willingness to share cannot be assumed. In an August 2025 survey of 4,000 knowledge workers in four countries, published by technology consultancy Adaptavist, 35% said they were hoarding knowledge for fear of being replaced. It was a self-reported survey, not a study of store managers. Adaptavist

A September 2026 IESE analysis makes the same organizational point: expertise transfer can meet resistance when employees see it as taking away what made them valuable. IESE Insight

Calling that resistance “a bad attitude” would miss the design problem.

If the manager contributes cases during a scheduled working session, someone has to cover her store. If she reviews disputed answers, that is work. If her contribution helps other locations improve, the review should recognize more than her own store’s results.

A sensible arrangement could include protected time, an explicit coaching responsibility, recognition for validated contributions, and an incentive tied to demonstrated use—not to the number of documents produced.

Be equally clear about what gets recorded, who can access it, whether it leaves the company’s approved environment, and whether the project changes anyone’s role. Don’t promise job security you cannot guarantee. Don’t hide the purpose behind a vague request to “help us with AI.”

Where The Pattern Can Reverse

The “AI lifts the bottom” result is not a law. In a Berkeley Haas field experiment with Kenyan entrepreneurs, access to a GPT-4 business assistant produced no detectable average effect on revenue or profit, while low performers did nearly 10% worse and high performers may have benefited by over 15% (Berkeley Haas). The researchers said the outcome depended on which tasks people chose for the AI: stronger firms asked for help with straightforward tasks, while weaker firms asked about difficult ones (Haas Magazine).

Store leadership looks more like open-ended business problems than a scripted support chat. That is why the pilot should start with a narrow, repeatable decision, not a general “ask the best manager” assistant.

The Owner's Case Is Fair

The strongest opposing argument is straightforward: knowledge sharing is already part of a manager’s job.

The company pays for the work, supplies the setting in which experience develops, and carries the consequences when knowledge stays with one person. It should not have to create a separate royalty arrangement for every useful answer. A manager who deliberately remains indispensable by keeping everyone else dependent is not building a healthy business.

There is also a plausible benefit to the expert. If newer managers learn faster, the best manager takes fewer routine calls and has more time for difficult problems and coaching. The support-agent study found evidence that AI assistance facilitated learning among less-experienced workers, which is the mechanism that would make that true. Stanford GSB

All of that is fair. The obligation to share knowledge does not disappear because AI enters the picture.

Nor does every capture project need a new compensation scheme. A modest pilot may need nothing more than scheduled time, clear expectations and honest recognition.

The distinction is between expecting ordinary collaboration and quietly adding an ongoing system-building job. If the business wants the latter, it should define the responsibility and make it workable.

Test One Transfer

Before buying a platform, run a small transfer test:

1. Choose one recurring field question and one experienced manager willing to help.
2. Review three resolved cases, including one where the normal answer would have been wrong.
3. Turn the reasoning into an approved guide. Remove private employee and customer information.
4. Let a newer manager use it on fresh cases. An experienced reviewer checks the decisions before action.
5. Record answer quality, review time, follow-up calls and errors—not just how quickly a response appeared.

Use AI only if it makes that guide easier to find, understand or apply. If a one-page decision card works just as well, use the card.

Your best manager’s knowledge may be an AI opportunity. The larger opportunity is building a business in which helping other managers succeed is part of how the best manager succeeds too.