NP Nithin Palal

Chapter 27: The Investor’s Playbook

Part V: Conscious Strategy

· 17 min read · Discuss · Plain text

AI-Proof: What Machines Will Never Be — Volume II: Application, Part V: Conscious Strategy, Chapter 27: The Investor’s Playbook

The call came on a Thursday evening, and Rajan almost did not answer it.

He was in his office on the second floor of the factory in Coimbatore, reviewing export documentation for a consignment of brushed cotton headed to Copenhagen. The looms had stopped at six. Below, the factory floor was dark except for the security post, where a radio played something filmi and indistinct. His phone showed an unfamiliar Singapore number. He let it ring.

Eighteen months of scar tissue had made him cautious. He had installed AI systems in a burst of competitive anxiety — quality inspection, demand forecasting, production scheduling, automated client communications — because three competitors had done it first and his buyers in Hamburg and Tokyo were asking what he was doing with AI. The honest answer was nothing, and nothing did not sound competitive. So he called his nephew and said: do everything. The technology worked. The dashboards glowed green. And on the factory floor, quietly, a catastrophe unfolded. His most trusted floor supervisor — a man who had been with him nineteen years, whose eye for fabric quality Rajan trusted more than his own — stopped offering opinions. The younger workers stopped learning. Why develop judgement when a camera judges faster? Productivity climbed. Something harder to name collapsed.

Rajan caught it too late. That was the wound. Not the technology’s failure — the technology had performed exactly as designed. His failure. The failure of a man who had run a factory for twenty-seven years and had not thought to ask the people on his floor how they felt about a machine that was designed to replace their judgement.

The recovery came through his son Vikram — not Karthik the nephew who had brought the pitch deck, but the quieter presence who ran the factory floor alongside him. Vikram had introduced AI demand forecasting as a subordinate tool — something Rajan reviewed, adjusted, and sometimes overrode based on knowledge no dataset contained. A conversation with a buyer whose tone had shifted. A rumour about a competing mill’s problems. The political situation in a cotton-growing region. That arrangement worked because it preserved the thing the first installation had nearly destroyed: the human judgement that made the data worth having.

So when the phone rang with a Singapore number, Rajan’s body remembered before his mind caught up. Wariness. The particular tension in the chest that comes from having trusted something that looked right and turned out to be incomplete.

He answered on the fifth ring.

It was Karthik, his nephew. Not the breathless Karthik who had once forwarded articles about digital transformation with the urgency of someone delivering prophecy. A different Karthik — older by two years and what felt like a decade, working now at a venture capital firm in Singapore, speaking with a deliberateness that Rajan recognised as the sound of someone who has learned that not every exciting thing is important.

“Uncle, I have a company I want you to look at.”

“I am not an investor, Karthik.”

“You are a man who has deployed AI in a textile supply chain and watched it fail and watched it succeed. This company is building AI for textile supply chains. My firm has people who understand the technology and people who understand the finance. We do not have anyone who understands what happens when you put this kind of system in a real factory with real workers and real suppliers.”

A lorry passed on the road outside. The security radio switched to a new song.

“Send me the deck,” Rajan said.

What arrived the next morning was a presentation from a company called ThreadMap. Two founders — one with an engineering background from Chennai, one from Zurich — who had built a platform for textile supply chain optimisation. Sourcing, quality prediction, logistics, demand forecasting, supplier risk assessment. The deck was polished. The market analysis was thorough. The financial projections showed the upward curve that Rajan, even with no investing experience, found suspicious — not because the numbers were wrong, but because they were smooth. In twenty-seven years of business, nothing had ever been smooth.

He read it twice. He understood more of the technology than he expected — because he had lived it. The demand forecasting module resembled what Vikram had installed. The quality prediction tool was a cousin of the inspection cameras that had caused such trouble on his factory floor. He knew, from experience, what these tools could do. He also knew where they broke.

He knew what the deck was missing.

27.1 What the Deck Does Not Say

Karthik’s firm was not blind to the standard risks. The venture capital industry had been discussing wrapper mortality and model dependency openly since 2023. Any competent analyst in 2026 could identify a company that had built a thin interface over someone else’s intelligence and called it a product. These were not novel observations. They were the table stakes of AI investing.

What the analysts could not see was what Rajan saw.

The deck’s pilot results described reduced sourcing time in a trial with mills around Tiruppur. Rajan knew those mills. He had worked with three of them for over a decade. He knew that sourcing time in the Tiruppur cluster was not a metric. It was a tangle — relationships, seasonal rhythms, festival schedules, monsoon disruptions, the particular stubbornness of mill owners who would rather lose a contract than be told how to run their operations by software. The pilot had measured one dimension of a reality with many dimensions, and the deck presented that slice as though it were the whole.

The competitive landscape showed a grid with ThreadMap in the favourable quadrant. Better technology, stronger data. Rajan recognised the format. What he noticed was the absence. The grid mapped software competitors. It did not map the people who actually performed the function ThreadMap was trying to automate: exporters like Rajan, brokers in Surat, agents in Karur, buying houses in Delhi. These were not technology companies. They were relationship networks, performing with human judgement what ThreadMap proposed to perform with algorithms. The deck treated them as irrelevant. They were the actual competition.

The technology section described the AI model’s capabilities in detail. It did not describe what happened when the model contradicted the people on the ground. Rajan had lived this. His own production scheduler had optimised away his speciality line — the hand-finished textiles with traditional South Indian motifs that sold in small quantities to boutique buyers in Europe and Japan — because the algorithm could not perceive that a low-volume product was his identity in the market. The correction came not from a dashboard but from a phone call. A buyer in Tokyo whose voice carried a faint note of disappointment that no sensor could detect.

ThreadMap’s deck was careful and thorough within its frame. The frame was wrong. It assumed that textile supply chain management was primarily a data problem. It was primarily a trust problem, with data in a supporting role.

27.2 The Durability Test

Rajan did not use the vocabulary of venture capital. He did not think in terms of defensibility or addressable markets — the language investors use to describe why a company might endure and how much it might be worth. He had his own questions, built from twenty-seven years of evaluating suppliers, buyers, and partners.

Is this person serious? Will this last? What happens when things go wrong?

Simple questions. They cut through complexity that more technical analysis often obscured.

What Rajan’s experience adds is a truth about how frameworks are applied: a framework wielded by someone who has never lived inside the industry being evaluated is an instrument played by someone who has never heard the music. The notes are correct. The feeling is wrong.

Rajan’s instinct was about the balance between computation and judgement. The more a company’s value depended on computation alone, the more fragile it was — because computation gets cheaper with every passing month. A company whose core value was a clever algorithm sitting on a foundation model had a useful life measured in quarters. A company whose value included human relationships, regulatory expertise, physical knowledge, or the kind of trust that takes years to earn had a useful life measured in decades. The distinction between these two — between a business and a feature — was the distinction that mattered. In 2026, making it required judgement, not calculation.

Some readers will object that every venture capital firm already evaluates defensibility and team quality — that none of this is new. The objection is fair. What the AI era changes is not the principle but the difficulty. In a traditional technology company, defences can be assessed by examining patents, market share, switching costs — all approximately measurable. In an AI company built on someone else’s model, the most important defences are often the least measurable: the depth of the founding team’s domain understanding, the quality of their relationships, their capacity to adapt when the technological ground shifts. These are assessments of consciousness. No screening tool has learned to exercise them.

27.3 Three Phone Calls

Rajan spent a week with the ThreadMap deck before agreeing to fly to Singapore. During that week, his stubbornness — the quality that has proven both his greatest asset and his most persistent limitation — was on full display.

He called Karthik on Monday evening. He had found that the competitive analysis omitted the entire category of human intermediaries. Exporters, brokers, buying houses — invisible.

“They are not serious people,” Rajan said. “They have mapped the companies that do not matter and ignored the people who do.”

“That is exactly why you should come,” Karthik said.

“To tell them their deck is incomplete?”

“To tell them what is missing. That is a different thing.”

Rajan hung up and stared at the deck on his desk. He had the uncomfortable sense that Karthik was handling him — patiently, deliberately, the way Rajan himself handled mill owners who resisted a change they needed to make.

He called again on Wednesday, after calculating that the growth projections assumed mill owners would adopt the platform faster than most of them had adopted mobile banking. “They are dreamers,” he told Karthik. “They do not understand how slowly this industry moves.”

“Come anyway,” Karthik said.

He called a third time on Friday. He had read the technology section four times and still could not determine which capabilities the founders had built and which they were licensing from a model provider. The distinction mattered — it was the difference between owning your equipment and renting it — and the deck blurred it.

“They are unclear,” Rajan said.

“Uncle.” A pause. “You have now spent a week with a forty-page deck for a company you say you have no interest in. You have identified three problems that my entire team missed. Come to Singapore. Tell them to their faces. If you are right, you will have done them a service. If you are wrong, you will learn something. When was the last time you did something you have not done before?”

Rajan did not have an answer for that.

He booked the flight on Saturday morning. It was, by a margin of decades, the most unusual thing he had done in his professional life. A man who had built his entire career within the triangle of Coimbatore, Tiruppur, and the port at Chennai, travelling to a glass tower in another country’s financial district to evaluate a company he had every reason to reject. His wife asked what had changed. He was not sure. Perhaps it was Karthik’s last question, sitting inside him like a stone in a shoe. When was the last time he had done something he had not done before? He could not remember.

27.4 The Dependency Beneath the Deck

The conference room was high enough above Singapore’s harbour that the container ships below looked like toys in a bathtub. The ThreadMap founders were young, articulate, and prepared. They had answers for the financial questions, the technology questions, the market questions. They had rehearsed their responses until the responses were smooth.

They had not expected Rajan.

He began with a question about the platform’s relationship to its AI provider. The deck described the model’s capabilities across many slides. It never addressed the dependency this created.

Rajan had lived a version of this dependency — not with AI, but with buyers. Early in his career, a single buyer in Manchester had accounted for nearly half his revenue. When that buyer went bankrupt, Rajan nearly followed. The lesson was permanent: never let one relationship determine your survival.

“If your model provider decides to build what you have built,” he said, “what do you have left?”

The founders exchanged a glance. The answer, when it came, was honest: they had their data, their customer relationships, their domain-specific workflows. Whether these would be sufficient without the underlying model was, they admitted, an open question.

“That is the right answer,” Rajan said. “The wrong answer is ‘that will not happen.’”

The deeper risk of model dependency — the one that keeps thoughtful investors awake — is not that the provider will raise prices, though pricing shifts occur. It is that the provider will release the startup’s core capability as a native feature. Free. Integrated. Available to everyone. This has happened repeatedly in technology platforms: the cloud company that builds the service its customers were selling, the social platform that absorbs the feature its partners invented. When the intelligence layer belongs to someone else and that someone decides your market is worth entering, you are competing against your own foundation. Investors who understand this — and by 2026, the serious ones do — ask about architectural independence: how tightly coupled is the product to a specific model, and what survives a switch? Rajan’s version was less technical. “If your key supplier disappears, what do you have?” He had asked it about cotton suppliers, shipping agents, dye houses. The context changed. The logic did not.

But Rajan added something the analysts could not. ThreadMap’s platform recommended which mills to source from. The platform had data about those mills — capacity, quality metrics, pricing. It did not have the knowledge that comes from walking through a mill at dawn and watching how the floor manager speaks to the weavers. Which owner will honour a commitment during a crisis. Which will prioritise a bigger order over a smaller one. Which seasonal constraints never appear in any database because they are governed by festivals, family obligations, and the unwritten codes of communities that have been making textiles for generations.

A platform that recommends suppliers without understanding the trust architecture of the industry will make recommendations that look right on screen and fail on the ground. Rajan had seen the pattern from the inside. His own scheduling system had optimised for the measurable and, in doing so, eroded the immeasurable. The pattern repeated at every scale: computation solving the visible problem while creating an invisible one.

27.5 The Question of Value

The financial slide said the company was worth eighteen million dollars.

Rajan stared at the number. He had spent twenty-seven years building a business that employed one hundred and forty people and shipped to fourteen countries, and he suspected his own company would not attract a comparable figure. Two engineers with a platform and eighteen months of existence were claiming to be worth more than what Rajan had spent a lifetime building.

He was not bitter. He understood, in a general way, that startup valuations reflect expectations about the future rather than a photograph of the present. What troubled him was not the number but what the number rested on.

Early-stage valuations have always involved more judgement than precision. The standard approaches — comparing to similar companies at similar stages, projecting revenue, assessing the size of the market being targeted — are reasonable starting points. But they share a weakness when applied to AI startups: they evaluate a company’s current output without adequately weighting the durability of its position. A startup with impressive early revenue but no structural advantage — no proprietary data, no trust-based relationships, no regulatory expertise, no physical presence — may be highly valued today and irrelevant in eighteen months, when a competitor builds the same product on the same model, faster and cheaper.

The reverse is also true. A company with modest revenue but genuine structural depth — real domain knowledge, relationships that resist replication, data that no one else has gathered — may be undervalued because its strengths resist quantification. The market pays for growth. It should also pay for resilience. In practice, the two are confused, and growth without resilience is rewarded as though it were growth with resilience — until the ground shifts.

Rajan traced the valuation backward. The projections rested on customer acquisition assumptions. Those assumptions rested on the belief that mill owners and exporters would adopt the platform because it was more efficient. The efficiency claim rested on the AI model’s performance. The model’s performance rested on a foundation the company did not own. At every layer, the chain was only as strong as a question no spreadsheet could answer: would real people, in a real industry, with real habits and loyalties and scepticisms, actually change their behaviour because a platform told them to?

Rajan had his doubts. He also had his stubbornness — and here it cut against him. He had spent so long building a business through handshakes and accumulated trust that he found it difficult to believe a platform could accomplish even a fraction of what it promised. Vikram had named this pattern during the demand-forecasting integration. “Appa, you resist every new tool until it proves itself three times over. Some things are good on the first try.”

Vikram was right. Rajan knew Vikram was right, which made the stubbornness more interesting than simple resistance. He was watching himself resist. He was asking whether the resistance was wisdom or habit. This is consciousness in operation — not the guarantee of a correct decision, but the awareness of the specific ways in which you might be wrong. Rajan was exercising it imperfectly, in a conference room in a country he had never visited for a purpose he had never imagined. The imperfection was the point.

27.6 Senthil’s Mill

The meeting had lasted two hours when Rajan asked the question that changed it.

“Does your platform recommend Senthil’s mill?”

The founders looked at each other. They did not know Senthil’s mill. They did not know Senthil.

Rajan explained. Senthil ran a mill in Tiruppur that produced what Rajan considered the finest brushed cotton in Tamil Nadu. Not the cheapest. Not the fastest. The finest — where finest included qualities no inspection metric captured: consistency across seasons, a willingness to hold production for a trusted buyer, a feel to the finished fabric that came from decades of knowledge passed from Senthil’s father to Senthil. Rajan had placed his first order with the family in 1998, when he was twenty-six and the mill was a fraction of its current size.

By every measurable criterion — price, speed, documented capacity — Senthil’s operation would rank in the middle of the Tiruppur cluster. A platform selecting suppliers on quantifiable data would steer users toward mills that measured better on paper and performed worse in practice. Mills that looked efficient but cut corners under pressure. Mills that met specifications but lacked the embodied knowledge that separates adequate fabric from fabric a buyer in Copenhagen will reorder without comparing alternatives.

“Your platform measures what can be measured,” Rajan said. “The things that make Senthil the best supplier I have ever worked with cannot be measured. They are known because I have been in his factory at five in the morning. They are known because my first order was placed with his father. They are known because trust is not a data point.”

Silence.

Then one of the founders said: “We know.”

He described the pilot. Manufacturers had kept contradicting the platform’s recommendations. For three months, the ThreadMap team assumed the contradictions were irrational — human bias overriding superior data. Then, gradually, they realised the manufacturers were being rational in ways the system could not see. They were making decisions based on relationships and history and knowledge that existed nowhere in the platform’s data.

“We do not know how to encode trust into software,” the founder said. “That is why we need someone like you.”

Rajan had been prepared to walk away. His list of objections was long, his scepticism deep, his instinct toward caution reinforced by eighteen months of hard experience with AI systems that promised more than they delivered. He had spent two hours finding more wrong with ThreadMap than right.

But the founders had done something Rajan valued above almost everything else in a business relationship. They had told him what they did not know. Not hidden the limitation. Not reframed it as a feature. Sat in a room full of investors and said: we built something powerful, and it is not enough, and we need help understanding why.

The gap between “we have built the future” and “we are building toward something we cannot yet see clearly” is the gap between confidence and consciousness. Rajan had spent his career on the receiving end of sales pitches — from fabric suppliers, from technology vendors, from the consultants who circled his factory promising transformation. The pitches that earned his trust were never the most polished. They were the ones that included the sentence most founders are afraid to say: here is what we do not know yet.

For founders seeking investment in the AI era, this is the counterintuitive lesson that the best investors have learned. The pitch that leads with the model’s capabilities tells the investor what the company has today. The pitch that leads with what the model cannot do — and what human layer the company is building to compensate — tells the investor what the company could become. Products can be copied. A vision grounded in honest self-assessment is harder to replicate.

27.7 An Act of Judgement

Rajan invested. A modest amount — appropriate for a man making his first angel investment, not a sum that would alter his life if lost. But he invested more than money. He agreed to serve as an industry adviser. He agreed to introduce the founders to mill owners he trusted — beginning with the curious ones and working toward the sceptical ones, because trust cannot be transferred in bulk. It moves through introductions, shared meals, small tests of reliability that accumulate into something larger.

He did this not because the numbers justified it. The numbers were uncertain, the technology unproven in real conditions, the founders capable but incomplete. He did it because his judgement — built from twenty-seven years of evaluating people and materials and situations, tested by failures that had cost him real money and real sleep, complicated by a stubbornness he was learning to recognise as both shield and cage — told him these founders had a quality worth investing in. Not their intelligence, which was considerable. Their willingness to see what their intelligence could not reach.

Whether this was wise belongs to the future. Judgement does not come with guarantees. That is what separates it from calculation. A calculation can be checked. A judgement can only be lived with — owned, examined, revised when new evidence arrives, and, when wrong, absorbed as the kind of failure that makes the next judgement more honest.

On the flight home, somewhere over the Strait of Malacca, Rajan called Vikram. The connection was poor. He told his son what he had done. There was a pause — the particular silence of a child recalibrating their understanding of a parent.

“Did the financials support it?” Vikram asked.

“Not entirely.”

“Then what did?”

“They knew what they were missing,” Rajan said. “That is rare.”

The seatbelt sign came on. Rajan ended the call and sat with the sentence he had just spoken, turning it over like a fabric sample, feeling its weight. It was the same quality he looked for in a supplier, a buyer, a partner. The awareness of limits. The willingness to seek what you cannot provide alone. The understanding that the gap between what you know and what you need to know is not a weakness to be hidden but a space to be filled — with relationships, with trust, with the patient accumulation of shared experience that turns strangers into partners.

↑ Back to top

Discuss

Loading the discussion…

More in Artificial Intelligence
  •  ·  26 min read

    Epilogue

    Late in the Gītā, the one who knows most says: ponder it thoroughly, then act as you like. A machine can now set out the reasons on every side. Saying “I answer for this” then is not pride. It is a reply, for your part, to those owed one.

  •  ·  37 min read

    Chapter 40: What This Series Cannot Tell You

    The print edition’s last chapter reduced the book to one sentence, and rested human worth on what machines cannot do. I withdraw both. A book can date its claims and correct them in public. It cannot sit in the room where you decide.

  •  ·  6 min read

    Part IX: The Unfinished Chapter

    The print edition’s last Part said that consciousness never concludes. This one is a single chapter that turns the series on itself: what a series can stand behind, what it owes those who read the print edition, and what it leaves open.