The R Doctrine  /  The technological reset

The AI Doctrine

A methodology for moving an institution from AI deployments to business outcomes.

the evidence

The results from the first wave of enterprise AI are in — and they show spend running far ahead of business return.

This is no longer a matter of opinion. The most sophisticated organisations in the world are reporting it about themselves.

95%
of enterprise GenAI pilots deliver no measurable impact to the bottom line.
MIT Project NANDA · 2025
$3.4B
Uber's entire 2026 AI-coding budget, spent in four months.
Uber, reported 2026
80%
of AI projects fail to deliver business value — four times the rate of conventional IT.
RAND · 2025
The budget I thought I would need is blown away already.— Uber's Chief Technology Officer, after the company's entire 2026 AI-coding budget was spent in four months

These are not timid institutions — most embraced AI sooner than their peers. What was missing was not enthusiasm, and not technology. It was the architecture to turn either into business results. This is not a failure of AI. It is the failure of deployment without the right model.

before the journey

Before the doctrine, the decision: should the organisation pursue the AI journey at all, or wait?

That decision is genuinely hard — not because the right answer is difficult to execute, but because it is difficult to know, and you are expected to know. It is best understood through seven dimensions — technology, investment, ambition, risk and liability, ownership, people, and readiness — and the culmination of all seven is what informs the decision.

The full structure is reasoned through in Issue Two of the Letter — Before the Journey: How a Board Decides on AI, and the board-level scan from that letter is a printable question set. For a board or executive team, the Senior Team Briefing runs the conversation live; the session is complimentary.

The R Doctrine’s nine dimensions examine the institution whole; these seven belong to one decision within it.

how it fails

Six ways AI deployments fail — and I have watched every one of them happen.

These are not theoretical risks. They are the recurring structural patterns behind the numbers above — the predictable result of treating an architectural opportunity as a deployment one.

01

Technology-first reversal

Start with what the tool can do, then hunt for a problem — instead of starting with a painful, measurable business problem.

02

The pilot-to-production chasm

Pilots succeed in controlled conditions; production collides with integration, data quality, governance, and cost.

03

The data-foundation gap

Roughly half fail at data readiness; most discover their infrastructure is inadequate only after they have launched.

04

Hidden cost overruns

Spend compounds far faster than outcome — runaway cost without proportional return, even at the most capable firms.

05

AI bolted onto legacy

New capability layered onto operating models built for a different era — throttled by the very architecture it was meant to transform.

06

The governance gap

Support functions become bottlenecks or are bypassed entirely — and unchecked AI becomes a serious institutional risk.

The failure modes are not random or unlucky. They are the precise, predictable result of treating an architectural opportunity as a deployment one — which is exactly why they are addressable, by anyone willing to ask the harder question first.

the discipline

The AI Doctrine runs as four stages, in a continuous loop — each built to defeat a specific way deployments fail.

It was not assembled as good practice in the abstract. Each stage was designed against a documented failure mode — a stronger claim than "this is a sound methodology." It targets the precise failures the evidence has already recorded.

Stage one
ReFrame

Before a single tool is chosen, the Board and Executive team build a shared understanding of what AI makes possible — and define the business outcomes it must serve. The conversation begins where it matters most, not where it is most convenient.

Built to defeat
01 Technology-first reversal05 AI bolted onto legacy
Stage two
Re-Form

The heaviest stage. It rewires the institutional operating system — removing the legacy gating that throttles execution, and turning the support functions (risk, legal, compliance, HR) from gatekeepers that vet what AI produces into enablers that help produce it.

Built to defeat
03 The data-foundation gap05 Legacy retrofit06 The governance gap
Stage three
ReBuild

Execution, on a new mental model: build fast, fail fast, rebuild fast. AI collapses the cost and time of building, so the old software-development gates only throttle it. Cost realities surface early — before mass infrastructure commitment.

Built to defeat
04 Hidden cost overruns
Stage four
React

The stage of institutional honesty. Do the pilot's results actually hold in production? It resists both premature celebration and premature abandonment, distinguishes operational wins from strategic shifts — then feeds back into ReFrame for the next wave.

Built to defeat
02 The pilot-to-production chasm
ReFrame → Re-Form → ReBuild → React → ReFrame
the shape of the journey

Begin along three tracks, run in parallel — not in sequence.

The four stages are how each wave of work runs. The journey itself is best begun along three parallel tracks — each feeding the others, all of them in motion together rather than in a queue.

Adopt & Automate

Deploy AI broadly across the organisation, get people using it in their own work, let them automate tasks and find efficiencies. The point is not the efficiency itself: it is to make the organisation conversant with AI, and to tap into hinterland innovation — the unprompted use that surfaces ideas no roadmap would have planned. The headline results do not come from here; its value is fluency, and a stream of surfaced ideas the funded work draws on.

Reimagine the Work

Organisation-directed: a deliberate effort to re-engineer a particular product, business, or function toward a defined business outcome — a business and a technology leader as joint sponsors, a cross-functional team doing the work, and a report back to the board carrying both the results and the lessons.

Build the New

A conscious, separate effort to launch a genuinely new, AI-native business: a dedicated, named leader, specific revenue objectives, and the mandate to build from the ground up. Begin it on day one, in parallel with the others — it is what allows an organisation to reach real revenue from AI sooner than most.

funding the journey

Fund in cycles: the board approves one quarter at a time, and the teams earn the next.

There is a base investment any start requires — getting the organisation ready, training a meaningful proportion of people, acquiring the core licences and platform. Beyond that base, the board approves funding only one cycle at a time, each cycle carrying specific objectives to be reached by its end, usually a quarter.

At the end of the cycle, the sponsors come back to the board: what was achieved, what failed, what was learned, what they propose for the next cycle, and the funding it requires. It pushes the teams to pick a focused area and demonstrate the outcome before rolling it out across the whole organisation — a far stronger discipline than a blanket cheque written up front.

governing the risk

Build on an architecture of multiple models — no single provider is worth the concentration risk.

The frontier-model capability a business runs on has a switch, and that switch can be controlled by a government or a vendor. The mitigation is architectural: local and open models alongside the frontier ones, treated as interchangeable components. Done well, that one choice de-risks continuity, mitigates cost, and keeps private and proprietary data on models you control.

Two further things belong on the board’s agenda: a dedicated working session on managing risk in the AI world — too consequential to fold into a general update — and the understanding that compliance here is a moving target: as organisations adapt to AI, so do the regulators.

the half most miss

An AI transformation that ignores the people will fail — however good the technology.

The four stages rewire the technology and the work. But the same transformation is asked of the people who must carry it — and this is the half most institutions underestimate. For both the leaders and the wider workforce, it is a move from one state to another.

The leadersWhat the reset asks of those who run the institution
Before
  • Authority drawn from experience — from having seen it before
  • Mastery of a function, a part of the whole
  • A playbook that worked, refreshed every few years
  • Change handled as a programme with an end date
After
  • Authority earned through present capability
  • Owning the whole, not a part
  • Reading the reset and acting with discernment, continuously
  • Leading through change as the operating condition

The hardest part: helping capable leaders unlearn the very things that made them successful.

The peopleWhat the reset asks of everyone who carries the institution forward
Before
  • A defined role, with a stable definition of doing it well
  • "The best people" measured by yesterday's criteria
  • A destination set, then executed toward
  • Uncertainty carried privately, alone
After
  • Work redefined, with honest reskilling toward it
  • A new, clearly communicated definition of valued contribution
  • Committing to a destination that keeps moving — held openly
  • Trust deliberately held while everything else is in motion

Some roles carry forward transformed; some will not — and that is handled with directness, not hidden behind process.

Carrying people across these two transitions is the substance of two further apertures of the R Doctrine — the Leadership Doctrine and the People Doctrine. They go far deeper in their own right. But no AI transformation should proceed without holding both from the start.

where this belongs

This is not a technology rollout to be delegated. It is a business-architecture decision the Board and Executive team must own.

The most consequential decision an institution makes about AI is not which tools to buy or which pilots to run. It is whether the question gets owned where business architecture is actually decided — the Board and the Executive team — or whether it is delegated down to be solved as a technology rollout. That single choice separates the institutions that will climb from the ones that will spend.

And for those who own it early: the reset is repricing every accumulated advantage at the very moment new capability becomes universally accessible. This is the most level playing field in living memory — and the advantage at Level 3 is still uncontested. Those who embrace it early will compound generational advantage over the next decade.

the operating model

Beneath the board and the chief executive, the structure is a choice — four models.

Chief-executive-led

Responsibility at the top, the executive team executing. Its strength: it treats AI as business architecture, owned where it should be. Its weakness: it depends entirely on a chief executive who is truly fluent and engaged — and even then, that person cannot run it day to day.

Technology-led

AI run as a programme out of the CIO or CTO organisation. Genuine technical depth — but the business-architecture decisions the work requires sit outside the owner’s remit, so the effort drifts toward automating existing processes.

Federated

Business and technology co-owning, with a dedicated transformation lead reporting to the chief executive and AI embedded across the functions. The ownership matches the shape of the problem — AI is horizontal. The risk: the lead becomes a programme office with no real responsibility.

The Chief AI Officer, as builder

Not a coordinator: a senior owner who works across the existing businesses to transform them, and also holds the mandate to build the new AI-native business itself. It demands an exceptional individual and a true mandate; done weakly, it recreates the very silo the title is usually criticised for.

There is no right model in the abstract — only a fit to the organisation, the culture, and the depth of its business and technical talent. My own view, plainly: consider a dedicated Chief AI Officer charged with building the AI-native business in parallel. It gives the company a head start on real revenue, and a genuine distinction in the market.

the practitioner thinker

I have not written this as a thinker. I have lived it as a practitioner.

For three decades I have operated at the intersection of business strategy, technology, and markets — turning around and transforming businesses across financial services, fintech, and global data, at different sizes and across India, the US, the UK, Japan, and Southeast Asia. Decisions made when the path was genuinely unclear and costly, with real people on the other side of them. That is where this thinking comes from — not from the outside.

And in addition to transforming businesses, I have the most direct kind of evidence on AI specifically: I have built two enterprise-grade, multi-agent AI systems myself, as a team of one, with AI as my collaborator. Work that would once have taken a team of twenty and two years, built in weeks and months.

Built & working
The AI Cognitive Engine

A four-agent system that discovers trading strategies from first principles — given five years of raw market data and one deliberate constraint: no technical indicators, ever, so it must derive structure from the data itself. The agents explore, refine against return and risk thresholds, validate against unseen data, and paper-test before a signal goes live. It has surfaced genuine strategies across option buying and selling, intraday, and multi-week horizons.

In build · pilot-proven
Yukti

A multi-agent platform built on a single idea: that intelligence and execution can each come from a human or a machine, and the real work is composing the four combinations deliberately. Strategy agents decide; a separate execution backbone carries orders across brokers asynchronously. The layer being added now is a fund-manager agent that allocates capital and detects regime shifts ahead of them — engineered for stable returns in an unstable world.

This page is the what and the why.
The how cannot be written down — it is lived.

Thirty years of judgment, and systems built by hand. That is the work itself — and it is where I come in.

carry it with you

This page is an instrument. Three pieces of it travel.

Built to be carried into your institution. Take it to the room that decides.

An invitation

If your institution is somewhere between the 95% and the few, I would welcome the conversation.

No intake process, no gatekeeper. A note reaches me directly, and I reply myself — usually within a day or two.

This comes straight to my inbox. I read every note personally and reply myself — not a team, not an auto-responder.

Message received

Thank you — your note has reached me.

It comes straight to my inbox, and I reply personally — usually within a day or two. I look forward to the conversation.

Or reach me directly
Mobile
Call or message — India (IST)