The risk with the best evidence behind it is usually the one management already reports on. Build-A-Bear shows how that happens. Its most visible risk is its leases. Between fiscal 2023 and fiscal 2025, leases carried in invested capital rose 65% while after-tax operating profit rose 3%. The numbers are clean, reported, and hard to argue with. Its deeper risk has no clean numbers at all. Gift buyers and teen and adult buyers may be drifting to finished plush sold without the build. Every signal of that shift comes with another explanation attached. Online demand is falling, but the company names four causes and does not separate them. Teen and adult visits fell faster than total visits, but weaker mall traffic could explain that too. Rank these two risks by evidence, and the leases win easily. That ranking is wrong. The leases cannot break anything while buyers keep paying the premium. Store return on invested capital was still 27.2% in fiscal 2025, so new stores still earn back their capital. The leases only matter if the buyer base shrinks. They carry the damage. They do not start it. So the risk with weaker evidence sits upstream of the risk with stronger evidence. This is not a coincidence. A risk is well documented because someone already measures it. A risk gets measured because management already sees it. Evidence follows attention. The risk nobody is watching produces only scattered signals, each with a reasonable cover story. That is exactly what an early, unmeasured shift looks like. The structure explains why this matters here. Neither Build-A-Bear business has a competitive advantage. The premium lasts only as long as buyers keep deciding the build is worth it. In a business like that, the buyer base is the one condition everything else depends on. The leases, the partner channel, and the margin all sit downstream of it. A risk to that condition outranks any risk that merely carries its effects, however well documented. A better test for ranking risks is structural. Which risk weakens several parts of the business at once? Which one, if it happens, makes the others bite? That test points somewhere evidence cannot. A risk list ranked by evidence is a list ranked by what you already measure. The risk that decides the outcome is often the one with no line of its own in your reporting.

Build-A-Bear’s real risk is not its leases. It is the buyers who never needed the build. A child building a bear in the store is not the problem. That child is the buyer the stores were designed for. The problem is everyone else who pays the premium: the adult buying a gift, the teen or adult buying a licensed character. They can buy finished plush somewhere else. If they leave, the premium is earned on fewer visits than the stores were sized for. And the company has been signing longer leases on the assumption that they stay. Build-A-Bear runs two businesses. Its 375 company stores, every one leased, operate as a traditional retailer charging a premium markup. Their strategy is customization. The build is the product. The second business is wholesale. Cruise lines, resorts, and restaurants run Build-A-Bear stations in their own venues and buy inventory from the company. That wholesale side has no strategy of its own. Partners choose the sites, run the stations, and decide whether to open or close. Build-A-Bear supplies product. Neither business has a competitive advantage. Nothing stops a rival from offering the same occasion. Partners host stations because people already want the brand, and that demand forms in the company stores. The whole system rests on one condition: buyers keep deciding the build is worth paying for. A business without an advantage can run well for years. The strain begins when it signs commitments that only an advantage would make safe. Between fiscal 2023 and fiscal 2025, leases carried in invested capital rose from $73.4 million to $121.1 million, up 65%. After-tax operating profit rose 3%. Lease extensions moved to longer terms. That gap is the best-documented risk on the record. It is not where the break starts. Return on invested capital in the stores fell from 37.4% to 27.2%. A 27.2% return still pays for new stores. While the buyer base holds, the leases cannot break anything on their own. The leases are how a break would show up, not what causes it. They are fixed, so when demand shrinks, they carry the loss straight into margin. The cause sits upstream: buyers outside the children’s occasion moving to finished plush. Every signal of that shift has another explanation on the record. Online demand fell 11.8% in fiscal 2024, 5.5% in fiscal 2025, and 21.2% in the first half of fiscal 2026. The company names four causes, including changes to AI search, and does not separate them. Teen and adult store visits fell faster than total visits in the first quarter of fiscal 2026. Weaker mall traffic could produce that too, if teens and adults depend more on walk-by visits. The risk management already reports on is rarely the one that decides the outcome. Now look at what the reported numbers show. In fiscal 2025, store profit fell $5.9 million. Profit from third parties, mostly the partner channel, rose $5.1 million. Pre-tax income held near $67 million for a third straight year. The partner channel is covering the stores. But it depends on the same brand demand, so if the buyer base narrows, it follows. Revenue per partner location already fell by about a fifth over two years. That alone proves little, because partner locations grew from 92 to 178 in that time, and new locations take time to build revenue. Some departing buyers may choose Build-A-Bear’s own finished plush, sold wholesale through Walmart. That sale counts as wholesale growth. The company keeps the buyer, but the revenue leaves the stores that carry the leases. Birthday visits hold, because the children’s occasion is the last part of demand to move. Executives are paid on revenue, and revenue rises with every new location whether or not the premium holds. Each reading is accurate. None of them isolates the shift. This is the breakpoint: the company can keep its premium or keep the breadth of its buyers, but not both. Holding the premium shrinks the base below what the leases require. Holding the base gives up the premium the leases were priced on. Anyone watching the reported numbers would see that point only after it had passed. The question is not whether the leases are too large. It is whether anyone knows who is paying the premium. The only public measure of buyer mix is a 2022 survey that put teens and adults at about 40% of sales. Before setting another long lease term, leadership needs sales by buyer group, transactions per store, and licensed-product share of sales. Without those figures, the leases are being sized against a buyer base last publicly measured four years ago.

Buc-ee's best proof of quality is invisible to the system that will soon help choose the stop. That proof is the restroom attendant who cleans while you watch. It works on anyone standing in the room. It does nothing for a personal AI agent planning a stop three hours out. The agent can read reviews. Buc-ee's has millions of them, and they say the restrooms are clean. But reviews describe past visits. The agent is asking about the next hour. The obvious fix is for the company to publish live data: a cleaning log, open pumps, lot space. Here is the catch. Data a company publishes about itself is only a claim. A careful agent will treat it that way. It becomes evidence only when it can be checked against something the company doesn't control. Buc-ee's already has that check. A store whose published cleaning log matches what reviewers report, month after month, gives an agent a live signal it has reason to believe. A rival can publish a log tomorrow. Without that history, it is only the rival's word. So the reviews Buc-ee's earned over decades are worth more once the company starts publishing. The live data gets its credibility from the reviews, and the reviews get a present tense from the live data. The strategic frame makes this urgent. Buc-ee's competes on Performance & Quality. It holds no structural competitive advantage. Its formula sits at Operational Parity, where nothing makes leaving costly and performance is the entire defense. A business defended only by quality cannot afford to have that quality go unseen by whatever decides where the customer stops next. The question for leaders is not whether your reputation is strong. It is whether any of it can be confirmed by a machine at the moment the decision is made.

Buc-ee's is spending tens of millions of dollars per store to expand into states just as the choice of where to stop begins moving from the driver to a personal AI agent. Its customer loyalty is real, but it is stored in one place: travelers' memories. Buc-ee's holds no customer history, no account, and no app. An AI agent plans from what it can read. A preference that lives only in someone's head now competes with rewards data, live conditions, and everything else the agent knows. THE FORMULA Business model: Traditional Retailer. Buc-ee's sells one visit at a time at standard markup. Strategies: Selection & Availability, Performance & Quality, Build & Expand, and Customer Segmentation. It competes on immediate availability and on a quality standard rivals treat as optional. It grows by building company-owned copies of the same site. It serves one segment, the passenger-vehicle traveler. Competitive advantage: Operational Parity. Every strategy is built to win the visit. None of them holds the customer between visits. LOYALTY IS NOT A SWITCHING COST Buc-ee's ranked first of 41 convenience chains in dunnhumby's preference index, 17 points ahead of Sheetz. A competitive advantage requires a cost to leaving. Skip a Buc-ee's and stop at the next exit, and you lose nothing. No points expire. No account has to be rebuilt. So performance is re-earned every quarter at the restroom and the brisket counter. Buc-ee's executes better than anyone in its category, and execution is its entire defense. WHY ROUTE OPTIMIZATION IS THE WRONG WORRY For many families, the Buc-ee's stop is the point of the trip. The founder has said roughly two-thirds of revenue comes from inside the store. Nobody optimizes an event out of a road trip. But that assumes the traveler already knows the stop is an event. In Texas, everyone does. In the states Buc-ee's is entering, that knowledge has to be built from nothing. WHAT THE AGENT CAN VERIFY An agent proposes stops from the data available to it: locations, hours, live charger or pump availability, reviews, and any rewards accounts the family connects. Reviews are where Buc-ee's is strongest. But a review describes a past visit. An agent planning a stop three hours out is asking about the next hour. At many Buc-ee's stores, the only live data an agent can read is from the EV chargers, and Buc-ee's doesn't run them. WHERE THE EXPOSURE SHOWS Buc-ee's has more than 55 stores and roughly 15 in the pipeline. Public cost figures for a single site run from $38 million to $82 million. Each new store is a bet that a region will form the habit Texas formed over four decades. Wisconsin shows the difficulty. Kwik Trip runs more than 900 stores across Wisconsin and six other states, with a rewards app since 2018. A family that connects its Kwik Trip account gives its agent a record of stops most weeks at one chain and none at the other. The two chains sell different visits. Nothing in that record tells the agent so. WHAT LEADERS GET WRONG They treat brand preference as protection. To an agent, it is one weight in a larger calculation. For a family that plans around Buc-ee's, the agent will make the stop. An agent that overrode a preference that strong wouldn't keep its job. The exposure sits with travelers who like the brand but don't plan around it. For them, preference is a tiebreaker, and an agent breaks ties with what it can confirm. WHAT BREAKS FIRST Not Texas. The new markets, where the habit hasn't formed. A rival doesn't need Buc-ee's standard everywhere. It needs to be verifiably good enough, right now, at the exit before. Live data on open pumps, lot space, and restroom cleaning turns Buc-ee's reputation from a verdict travelers carry into a comparison at one exit. WHAT A FIX REQUIRES Buc-ee's doesn't need a loyalty app or an online store. It needs a building machines can read. A self-published cleaning log is only a claim. It becomes evidence once it matches what independent reviewers report, month after month. Published through a public, read-only interface, that data collects nothing about customers and leaves the business model untouched. *** AI agents won't eliminate the Buc-ee's stop. They move the decision into a system that weighs preference against facts it can check. A brand with no customer data can keep the affection and still lose the decision. Leaders of execution-driven businesses should stop asking only what customers think of them. The sharper question is what a machine can confirm about them right now, and which rival is answering it first.

What year would a time traveler come back to if they wanted to prevent AI catastrophe in this timeline? If that, in fact, has occurred, how would we be able to tell from within this timeline if the catastrophe has been avoided? ___ The time traveler arrived in late 2022 with a map. Not the kind with roads. It was a map of decisions — which meeting, which year, which sentence spoken by which tired person at which hour — and where each one led. Nearly all of them led to the same place: the year the machines stopped needing to ask. She had studied it for most of her life. She was the only one who had ever seen the whole thing. She was given one rule, and she gave it to herself: stay in the room where the large language models are built. So she did not resign. She did not go on television. She did not sign the letter. When colleagues quit over what the lab was making, she wished them well and kept her badge. When colleagues said the fear of the machines was overblown, she nodded and said nothing, because arguing costs credibility, and credibility was the only currency the map accepted. What she did instead was small. A review before a model shipped. An evaluation nobody wanted to run, because it would show what the system did when it believed no one was watching. A clause in a compute contract. A hiring decision she quietly killed. She was calm about the things everyone else was frightened of and immovable about two or three things that seemed, to everyone else, oddly minor. People found her steadiness faintly irritating. Someone joked that no new model had ever surprised her. She smiled and let it pass. She changed a handful of decisions. A research project received funding that hadn’t before. Another took six months longer than it had in her timeline. A dangerous capability was discovered behind closed doors instead of after release. An obscure safety technique received attention several months earlier. Tiny changes. And each time she pushed, the thing she remembered did not happen. That was the problem. Every road she closed took a piece of the map with it. The catastrophe she had memorized belonged to a world that no longer existed — the one where the evaluation hadn’t been run and the code hadn’t been written. The world she was now in had never been charted. ___ By 2024, she began to worry that she had failed. The models kept improving. By 2025, they were reasoning, writing software, operating computers, and performing tasks that people in 2022 had insisted would take decades. Companies poured fortunes into larger data centers. Governments became concerned about falling behind. Every advance seemed to produce the resources for the next one. To the time traveler, it all looked disturbingly familiar. Then came 2026. She read essays by people convinced that we were racing toward extinction. They pointed to accelerating capabilities. They warned that competition would overwhelm caution. They argued that once machines became sufficiently capable, humans might no longer control what happened next. She knew these arguments well. In her childhood, they were taught in history class. The thing was that these people couldn’t see what she could see. They didn’t know about the system that never escaped. They didn’t know about the experiment that never happened. They didn’t know that one technical breakthrough had arrived three years earlier than it was supposed to. They saw only the danger. And because AI continued advancing, some became more certain every year that catastrophe was approaching. She wanted to tell them they were wrong. But she couldn’t. Perhaps she was wrong. Her world had disappeared the moment she changed it. She had no way to observe the future of this timeline. She knew only that she had removed the chain of events that destroyed hers. Perhaps another chain had already begun. Or perhaps there had always been many safe paths forward, and her generation had simply been unlucky enough to choose the wrong one. So she kept her secret and watched. ___ She remembered the shape of the ending but not its date. She remembered a name spoken in a room, but that person had left the lab, and she could not say whether they’d left because of something she’d done. What remained was instinct. A tightening in the chest when certain sentences were said. In the fourth year there was a meeting. There was a deadline, and a competitor’s launch, and a new model that had done something in testing that nobody could fully explain. The room was full of intelligent people who wanted to go home. Someone said the sentence that made her chest tighten: We’ll fix it in the next version. And she sat there not knowing. Was this the one? Was this the meeting she had crossed time to sit in? Or was it an echo — a smaller cousin of the decision she had already, years ago, prevented without knowing it? Was the real one still ahead? Was it already behind her? The map was gone. She had only the feeling. ___ The optimists pointed to every safe year as evidence that the pessimists had exaggerated the danger. The pessimists pointed to every new capability as evidence that the optimists still didn’t understand it. Neither side could see the thing she had learned by traveling through time: A catastrophe prevented leaves no ruins. A catastrophe approaching leaves no ruins either. From the present, they can look exactly the same. That is why she went back to 2022. Not because she knew the future. Because she knew one future. And that was enough to know we should choose carefully.

The battery bolted to the side of a Base Power customer's house is not the moat. A 39.2 kWh cabinet feels permanent. Once it is installed, the household seems to belong to Base. But a DirecTV dish also stays bolted to the house long after the subscription ends. Physical presence retains no one. What holds Base's homeowners in deregulated markets today is a contract. The battery stays only if the household keeps Base as its electricity provider, and that obligation transfers to the next owner of the house. Contracts are not scarce. Competitors can write them too, and can buy Base's customers out of theirs. The friction of leaving is real but uneven. Switching to another battery provider means removal and reinstallation, and every provider imposes roughly the same cost. Switching to a plain retailer means losing whole-home backup. That is a genuine loss. Whether it is a structural switching cost depends on what happens when the term ends. If Base removes the battery, the household compares Base to a bare retailer with the backup already gone. If Base leaves the battery behind, the backup stays and the retention it produces stays contractual. Base's own pricing hints at its view. The removal fee is $500, priced as the cost of pulling the unit, not as a penalty. A company that wanted the hardware back would price exit higher. This fee suggests Base expects to renew, which means it expects to hold the household by contract again. In strategic formula terms, Base runs a Landlord business model: it owns the battery, keeps title, and grants the household a time-bounded right of use. Contract Lock is the strategy layered on top of it. It is a lever management pulls, not a competitive advantage. The advantage it could become is Demand: a switching cost that persists after the contract expires. Contract Lock is the mechanism through which that advantage might form. It is not evidence that it has. That evidence arrives only at term end. No twelve-year homeowner agreement has reached it yet. Until a contract expires and the customer stays without being obliged to, a locked customer and a loyal one look identical from the outside. Leaders who count installed batteries as retained customers are counting a test that hasn't been run.

Base Power is about to double its install rate, and nothing public shows anyone deciding which of its two businesses the new batteries will join. Every battery Base installs is dispatched by one of two parties. Where a co-op or city utility holds that right, Base collects a fixed toll. Where Base holds it, Base sells the household electricity at a promised below-market price and trades the battery in the wholesale market. The two earn differently. Only one carries a promise that has to be funded. I count two strategic formulas here, not one. Both rest on the same business model: Landlord. Base owns the battery, keeps the title, and grants someone else a time-bounded right to use it. Both share three strategies. Vertical Integration, because Base builds the battery in its own Austin factory. Build and Expand, because it replicates the arrangement territory by territory. Contract Lock, because every disclosed arrangement runs on a multi-year term. The deregulated formula adds two more. Trader, because the wholesale spread is Base's to capture. Low Cost Provider, because the retail rate is contractually held below a market benchmark. Austin Energy pays a fixed, non-escalating fee for 40 megawatts of reserved capacity, up to $4.08 million a year, for up to ten years. That revenue is fixed in price but contingent in duration, since the city funds it one budget year at a time. Every cost input floats. The homeowner barely counts as a customer. Austin Energy customers pay $19 a month. The house is a site; the utility is the counterparty. Zach Dell's line is "I don't sell batteries. I sell electricity." That describes only the deregulated half. There, being the retailer is less a retail business than market access. Outside a capacity-limited pilot, a battery behind a home meter cannot reach the Texas wholesale market unless someone is the retailer of record for that meter. So Base runs two books. The retail book holds the price commitment. The trading book is supposed to pay for it. Base has not disclosed how revenue splits between them. Nothing structural stops a competitor from running the same trade. A battery in ComEd territory goes under Base's own dispatch, because that market is deregulated and Base is licensed to sell there. A battery in GVEC territory goes under the co-op's dispatch, because the co-op is the one buying the right. Base chooses territories. The territory chooses the formula. What leaders get wrong: the battery looks like the moat. A DirecTV dish also stays bolted to the house long after the subscription ends. Physical presence retains no one. A contract does, and competitors can write contracts too. Contract Lock is a lever management pulls, not a competitive advantage. It becomes one only if leaving costs the customer something real after the term ends. No twelve-year homeowner agreement and no utility program has reached that point. What breaks first: contribution margin per unit. In-house manufacturing pushes cost down. The price guarantee pushes revenue down. Nothing public says which pull is stronger. If the trade comes up short, the $2.5 billion raised absorbs the gap. Where scale stops helping: a bigger fleet earns more spread, not a wider one. The market sets the margin per megawatt-hour, not Base's size. Any scale advantage will build in the factory and the dispatch software, which every battery feeds regardless of formula. What must be true: the trade has to fund the promise long enough for something structural to arrive. That means either a factory that builds the battery cheaper than incumbents can buy one, or customers who stay when a term ends and nothing obliges them to. Expansion at Base is not one decision. Each new territory is also a choice of business mix, and of how much capacity carries a price promise the trade must cover. The useful board question is not which formula is better. It is how much new capacity Base wants under a utility's dispatch versus its own, and what it assumes pays for the promise until an advantage forms. That makes market entry a capital allocation decision. It needs a named owner and a stated assumption about what funds the promise. Until a term comes due, the fleet will look like one bet. It is two.

Simple AI Visibility Decision Tree Does ChatGPT put your brand on the shortlist? → No: build category authority. → Yes: are your pages being fetched? → No: fix technical/retrieval issues. → Yes: are they being cited? → No: improve page-level citation competitiveness. → Yes: maintain both brand authority and page performance.

Here’s Your AI Visibility Action List 1. Determine which problem you have. Run the same open-ended, comparison-style category question in ChatGPT five times. Use a logged-out session if you want an unpersonalized baseline, and record the reasoning mode. 2. Check whether your brand appears in ChatGPT’s search queries. * Never/rarely named: You have a category association problem. * Consistently named: You are already passing the first gate. Move on to citation performance. 3. If you are not being named, invest in category association—not page optimization. Focus resources on: * Reviews and review sites * Comparison articles and buying guides * Trade and industry publications * Analyst coverage * Earned media * Repeated third-party association of your brand with the category Then make a deliberate positioning choice: compete for an established category or try to own an unsettled one. Don’t let miscellaneous PR language determine what category AI associates with you. 4. If you are being named, measure retrieval versus citation. Use server logs to determine whether AI crawlers are fetching your pages, then compare those fetches with actual citations. * Not fetched: investigate crawlability, rendering, discoverability, and technical access. * Fetched but rarely cited: you have a page-selection problem. * Fetched and cited: protect and improve what is already working. 5. For pages that are fetched but not cited, optimize for extraction and selection. Put the direct answer near the top. Keep important facts and numbers in plain HTML rather than images or script-dependent elements. Maintain a small cluster of tightly relevant pages where justified, but avoid producing large numbers of near-duplicate pages simply to target every possible question. 6. Stop treating unproven tactics as established AI-visibility levers. Don’t make llms.txt, Markdown mirrors, or schema the center of the program. Schema still has legitimate uses, but there isn’t evidence that these tactics by themselves drive AI citations. 7. Separate your KPIs. Do not combine mentions and citations into one “AI visibility score.” Track at least: * Shortlist rate: % of relevant queries where the brand is named/searched for. * Retrieval rate: how often your pages are fetched. * Citation rate: how often fetched pages are ultimately cited. 8. Split the budget according to the bottleneck. If you’re failing the shortlist gate, spend more on brand/category authority. If you’re passing the shortlist but failing citation selection, spend more on content and technical optimization. If you’re failing both, fund both workstreams rather than expecting SEO alone to solve the problem. 9. Measure the two gates on different time horizons. Treat category association as a slow-moving signal that may take months or years to establish. Treat retrieval and citation behavior as fast-moving signals that can change quickly as models and search systems change. 10. Don’t overreact to weekly volatility. Changes in queries, fetched URLs, citations, or measured visibility may reflect changes in the retrieval system rather than changes in your underlying brand position. Look for sustained trends rather than interpreting every weekly movement as progress or decline.