MBUS 853 — Session 5: Digital Architecture

Block 1 — Session Theme: What's Foundational vs. What's Tactical

Architecture Is What Makes Everything Else Possible — or Impossible

Session 4 asked what organizational structure enables adoption. Session 5 asks the layer underneath that: which technology and data decisions are foundational — load-bearing walls a company must get right before anything else works — and which are tactical, safe to defer or experiment with? Harley-Davidson's own CDO, Jagdish Krishnan, frames this almost exactly as the session title: he calls core IT/architecture/ERP/data work "digital defense" and everything built on top of it "digital offense" — "everything that becomes possible once the core systems worked."

Digital success exposed constraints earlier, which is a feature, not a failure.

Krishnan's own framing, on why scaling digital tools revealed capacity limits at Milwaukee and Tomahawk — architecture work makes problems visible faster, it doesn't create them.

A Second DBS Cameo

The Session 5 article closes with the same DBS Bank story that opened Session 1 — Piyush Gupta's $300M/year AI investment, DBS's 1,000+ data scientists, its "world's best bank" recognition. That's a deliberate bookend: after four sessions testing DBS's model against GE's execution failure, DeepSeek's fragile moat, and Pernod Ricard's adoption gap, this session asks whether DBS's real, durable advantage was actually architectural — a company that "went all in" on a flexible, cloud-based, unified data foundation years before asking what to build on top of it.

Block 2 — Article: Stop Tinkering with AI (Davenport & Mittal, HBR 2023)

It's Time to Go All In

Based on a 2019 MIT Sloan/BCG survey finding 7 in 10 companies report AI efforts have had minimal or no impact — and that even among the 90% who'd invested something in AI, fewer than 40% saw business gains after three years — Davenport and Mittal studied 30 companies that went "all in" on AI and reaped real returns. Architecture is squarely in the middle of their 10-action framework: not the whole story, but the hinge the rest depends on.

Action 3

Master analytics

Companies need proprietary data, not just access to the same data everyone else has — "if all your competitors have the same data, they'll all have similar machine-learning models and similar outcomes." Seagate Technology's disk-drive defect detection (accuracy from ~50% to 90%+) is the article's example.

Action 4

Create a modular, flexible IT architecture

Traditional data-center software only talks to itself; a flexible architecture can communicate with data from both inside and outside the company. Capital One's 2011 decision to rebuild its entire technology stack around the cloud, years before its AI payoff, is the clearest illustration.

Action 5

Integrate AI into existing workflows

Inflexible business processes are as limiting as inflexible IT. Target workflows that generate high volume/repetition first; don't force AI into seldom-used processes just because a model exists.

Action 9

Invest continually

Aggressive AI adoption is not a decision made lightly — CCC Intelligent Solutions spends $100M+/year on AI and data; DBS invested ~$300M/year over several years before its AI advantage compounded into record profitability.

"Companies with the most aggressive AI adoption, the best integration with strategy and operations, and the best implementation will achieve the greatest business value."

— Davenport & Mittal — architecture without strategic integration, or strategy without architectural investment, both fall short of the 30 companies studied

The Harley test: Krishnan's team already completed something close to Action 4 at York (nearly 80% of legacy technology removed or re-platformed) and Action 3 in narrow slices (real-time quality data, defect traceability). The open question the case leaves for Session 5 is whether Harley should extend that architectural work uniformly across Milwaukee and Tomahawk before layering on new "offense" capabilities — or move faster on offense even with an uneven foundation.

Block 3 — Case: Harley-Davidson: On the Road to Digitization

A Dashboard That Told a Troubling Story

Founded in Milwaukee in 1903, Harley-Davidson is one of the world's most iconic motorcycle brands, built on heavyweight craftsmanship and a loyal rider community (nearly half of buyers already owned a Harley). By October 2025, CDO Jagdish Krishnan had spent five years building an integrated digital dashboard spanning dealer inventory, retail demand, factory utilization, supply-chain logistics, financing risk, and legacy-system health. The dashboard itself was a genuine achievement — but the numbers it revealed were not: nearly $400M in supply-chain cost inflation since 2020, a second consecutive year of revenue decline in 2024 (down $600M+ to $5.2B), dealer inventories over 700,000 units forcing a 30% production cut, and fresh EU tariffs threatening profits on heavyweight motorcycles just as motorcycle revenue fell more than 25% in the first half of 2025.

Krishnan's Three Lenses

Lens 1 — End-User Experience

Consolidated 10+ separate apps and an agency-run website into one internally-governed platform (50M annual visitors). Built a virtual "Bike Builder" customization tool (6.1M builds/year, the site's most-used feature). Adopted a "dealer-touchpoint mindset" — BOPIS gave loyalty points for in-dealer pickup, driving ~20% of e-commerce orders into physical dealer visits.

Lens 2 — New Business Models

H-D1 Marketplace (July 2021) aggregated pre-owned Harley inventory online, fulfilled through dealers — framed by Krishnan as "a funnel for our dealers, not a competitor." LiveWire, spun out as a separate public company (2022), let Harley pilot direct-to-consumer and residual-value financing models without destabilizing the core dealer system — but diverted capital from a business that needed it.

Lens 3 — Modernizing the Core

Brought $10M/year of outsourced data work in-house. Built dealer inventory, sales, and dealership-visit dashboards. Ran a 3-month enterprise architecture review categorizing every application as needed-now, needed-future, or retireable. Pushed cloud migration "not to save money, but to get speed and agility" — cutting IT headcount from 230+ to 160 in the process.

The Conflict This Created

A vocal group of dealers resisted the centralized, connected e-commerce ecosystem, arguing it reduced their parts/apparel/accessory sales and diverted foot traffic. In May 2025, an association of 170 dealerships joined an activist shareholder in open opposition to Harley's direction; CEO Jochen Zeitz stepped down in June 2025.

By the Numbers — Q3 2025 Snapshot

$1.34B

Q3 2025 revenue, up 23% year-over-year

-6%

global retail motorcycle sales, same quarter

$27M

added cost from tariffs in the quarter alone

Dealer inventory was down ~13% year-over-year — but because Harley had limited shipments, not because retail demand had recovered. Operating margin remained near 5% despite the revenue growth, and HDFS (financial services) generated $439M in operating income, helping steady overall results even as core motorcycle sales softened.

Block 4 — Where Architecture Meets the Factory Floor

York: The Site Where Digital Transformation Was Tested First

York's assembly lines handle four motorcycle families and 20+ models — from an entry motorcycle to a $40,000 custom build — using "jellybeaning" (deliberately mixing colors, trims, and complexity levels rather than batching similar builds) to keep workload steady across the day. Layered on top of Toyota Production System-style lean methods (pull signals, takt scheduling, in-station quality checks), York added vision-system inspection, automated guided carriers, and "no fault forward" controls that stop the line if a required fastening step wasn't completed. By 2024, nearly 80% of York's legacy technology had been removed or re-platformed.

What Digital Actually Changed on the Floor

Traceability — the ability to trace a defect back to the specific operator, station, and timestamp — turned root-cause analysis "from guessing to actually fixing the problem," in one operator's words. Manual processes that once took two weeks (on a manual chain, cutting and welding) could be completed in a single shift with the new tools. But automation also shifted where responsibility sat: a vision system "can spot a defect. But it can't feel when the finish is wrong," as one paint technician put it — a direct tension between augmenting craft and threatening it.

The Adoption Gap the Article Would Predict

York became an early digital champion; Milwaukee and Tomahawk lagged, largely due to the complexity of their own operations. Manufacturing director Zach Merovich's rule — "what works at York must scale across Milwaukee and Tomahawk before it deserves capital" — is close to a textbook version of the article's Action 5 (integrate into existing workflows before expanding scope) applied to physical plants instead of software workflows. Annual capital investment across facilities ranged $115M–$225M, all of it competing against tooling, automation, and product redesign for the same limited pool.

Krishnan's own ambition, and his own caveat: digital twins — a full virtual simulation of the factory floor — could "merge all the data in the product lifecycle together, dramatically reducing costs and accelerating assembly speed." But his own assessment: "it's a long way off. I haven't seen it done in any environment as complex as ours." That's the case's clearest example of an architecturally sound idea that isn't yet a responsible near-term bet.

Block 5 — Memo Protocol: Team Case Study Memo #3

Format Reminder Before the Team Writes

Due Thursday 11:59pm before Session 5, based only on the Harley-Davidson case, two pages, 11-point font, written wholly by the team.

To Jagdish Krishnan, Chief Digital and Operations Officer — the case's own closing questions are addressed to him directly ("How should Krishnan allocate resources... What changes, if any, should he make..."), making him the clean single decision maker.

Issues Exactly 5, each grounded in a specific case fact — the York/Milwaukee/Tomahawk maturity gap, the dealer revolt and Zeitz's departure, the LiveWire capital tradeoff, the tariff/inventory/demand headwinds, and the "long way off" status of digital twins are all strong, distinct candidates.

Problem/Decision 40–60 words on the underlying cause — consider whether the root issue is architectural sequencing, capital scarcity, or stakeholder trust, since the three issues pull toward different root causes.

Alternatives Exactly 3, mutually exclusive, feasible, not simultaneous, not status quo.

Criteria Exactly 3 standards for judging the alternatives.

Evaluation/Recommendation 120–140 words, pros/cons per alternative per criterion, no table, ending in a justified pick.

Actions Exactly 3 steps not already taken in the case.

Case-only constraint: this case is dated January 2026 and set in late 2025 — it's easy to accidentally reach for real-world knowledge about Harley-Davidson's actual post-2025 performance or leadership. Stay strictly inside what the case states as of its own close, including the two open questions it leaves unresolved.

Academic integrity — GenAI is banned in submitted work for this course. The diagnostic analysis below is discussion prep, not memo text — the team's actual submission must be written independently.

Block 6 — Case Diagnostic: Issues, Decision, Position (Discussion Prep)

Applying the Case Prep Protocol

Step 1 — Who and What

Decision maker: Jagdish Krishnan, Chief Digital and Operations Officer. Core challenge: Harley has built genuinely strong "digital defense" at one plant (York, ~80% modernized) while Milwaukee and Tomahawk lag, and it's facing the sharpest external and stakeholder pressure of Krishnan's tenure (tariffs, declining sales, a dealer-led governance revolt) at precisely the moment it needs to decide where to place its next, more limited round of digital investment.

Step 2 — Candidate Issues Grounded in Case Facts

Uneven architectural maturity across plants. York is ~80% modernized; Milwaukee and Tomahawk still run largely on legacy systems, meaning company-wide digital capability is inconsistent even though it's proven where it exists.
A dealer-network trust crisis. 170 dealerships plus an activist shareholder publicly opposed Harley's centralized digital direction in May 2025, arguing e-commerce (especially discounted, free-shipping online sales) undercut their margins — a conflict serious enough to contribute to CEO Zeitz's June 2025 departure.
Capital diverted to an unproven bet. LiveWire's carve-out let Harley pilot direct-to-consumer and residual-value financing safely outside the core dealer model, but every dollar invested there was, in Krishnan's own words, a dollar not invested in factories the core business needed.
Severe external headwinds compressing the investment window. $400M in supply-chain cost inflation since 2020, a second straight year of revenue decline, a 30% forced production cut, and new EU tariffs all shrink the capital and patience available for further digital investment.
The most ambitious architecture play (digital twins) isn't ready. Krishnan's own assessment — "a long way off," unproven "in any environment as complex as ours" — means the most transformative available architecture bet isn't a responsible near-term allocation choice.

Step 3 — A Position

Underlying problem, one sentence: Harley proved its "digital defense" playbook works at York but hasn't yet extended it company-wide, and it now faces a capital and trust environment (tariffs, declining sales, a dealer revolt) too constrained to fund ambitious new "digital offense" bets before that foundational work is finished — so Krishnan's real choice isn't which new capability to build next, it's whether to finish the foundation or keep building on top of an uneven one.

Counterargument to weigh: One could argue the dealer trust crisis is the more urgent fire regardless of architecture maturity — a governance and stakeholder-relations fix (clarifying H-D1/BOPIS economics, formally renegotiating the digital-commerce revenue split with dealers) addresses a problem that already cost Harley its CEO, while further architecture work addresses a problem that hasn't yet visibly cost the company anything beyond internal friction. The strongest response has to weigh whether trust, once broken with 170 dealerships, is repairable through governance alone or whether it requires the underlying architecture (and the transparency it enables) to be finished first.

Second counterargument — maybe the answer is neither offense nor defense: with revenue down for a second straight year, dealer inventory still elevated, and fresh EU tariffs compressing margin further, one could argue Krishnan's real constraint isn't sequencing digital work at all — it's that Harley may not be able to afford either ambitious offense or full defense modernization right now, and the responsible near-term move is capital preservation (finish only what's already funded, e.g. Milwaukee/Tomahawk parity with York) rather than a new push in any direction. The strongest response has to weigh whether "pause everything" is actually available to Krishnan as CDOO, or whether the case's own framing — Starrs asking him to "set the pace for the next phase" — makes doing nothing new itself a decision with a cost.

Step 4 — 30-Second Cold-Call Answer

Krishnan's dashboard didn't create Harley's problems — it just made them impossible to ignore at the same time: $400 million in supply-chain inflation since 2020, a second straight year of revenue decline to $5.2 billion, and now 50% EU tariffs on heavyweight bikes. But the real architecture story is uneven progress — York is nearly 80% modernized and can trace a defect back to the exact operator and station, while Milwaukee and Tomahawk still run on the same brittle legacy systems Krishnan inherited in 2020. Building more "offense" — H-D1 expansion, digital twins Krishnan himself admits are "a long way off" — on top of that uneven foundation is exactly the mistake Davenport and Mittal warn against: DBS and Capital One both spent years on defense before their AI payoff compounded. So the real decision isn't which new capability to chase next, it's whether Krishnan finishes what York already proved works before a dealer network that's already lost one CEO loses patience with a second round of unproven bets.

Block 7 — Discussion Questions & Sharp Answers

Likely Professor Questions

Framing to expect: (1) How should Krishnan allocate resources across digital offense and defense? (2) What changes should Harley make to how it leads future experiments? (3) Was the dealer revolt a digital-strategy failure or a communication failure?

Q1: Should Krishnan prioritize finishing "digital defense" (Milwaukee, Tomahawk) or continue building "digital offense" (H-D1 expansion, digital twins)?

Defense first, on both the article's evidence and the case's own internal logic. Merovich's rule — nothing scales past York without proving out first — is already an internal admission that offense investments outrun defense readiness when pursued too early. Davenport and Mittal's Actions 3–4 (mastering analytics, building flexible architecture) precede Action 6 (building solutions across the organization) in their own sequence for a reason: DBS, Capital One, and Seagate all built the data/architecture foundation for years before their AI payoff compounded.

At Redamo Labs, the enterprise IAM platform needed a consistent verification architecture across all 50,000+ users before any new feature (fraud detection, personalization) could be layered on reliably — building on an uneven foundation would have multiplied technical debt, not capability.

Q2: Was the dealer revolt caused by Harley's digital strategy itself, or by how that strategy was communicated and structured?

Structure and communication, more than the strategy's substance. Krishnan's own "dealer-touchpoint mindset" and BOPIS program show real intent to protect dealer economics — and the data (20% of e-commerce orders converting to in-dealer pickups) shows it partially worked. The revolt happened anyway because 170 dealerships experienced the accumulation of changes (H-D1, direct online parts/apparel sales, discounting) as erosion of trust and margin, regardless of Krishnan's internal framing — a gap between designed intent and dealer-perceived reality that better governance and earlier dealer co-design might have closed.

Stutern's 2,500+ business partnerships required constant, proactive communication about how the platform's growth would or wouldn't cannibalize partners' existing revenue — assuming good intent would be understood without explicit reassurance was never a safe bet.

Q3: Was spinning off LiveWire the right way to pursue new business models, given it diverted capital the core business needed?

Yes, specifically because it was a spin-off and not an internal bet — Krishnan's framing ("LiveWire was about learning fast") only works because the carve-out let Harley test a direct-to-consumer model without destabilizing the century-old dealer system the core business depends on. The capital tradeoff is real, but the alternative (running the same experiment inside Harley's core P&L) would have created exactly the kind of dealer-trust conflict the case shows already happening with H-D1 — at a moment Harley can least afford a second front in that fight.

This is the strongest question to lead class discussion with — it forces a genuine tradeoff (capital discipline vs. structural experimentation) rather than a clean right-or-wrong call, which is exactly the kind of judgment this course rewards testing out loud.

Block 8 — Participation Hooks & Taju's Edge

How to Contribute Distinctively

Open Strong

Don't open with "Harley needs to modernize its factories." Open with Krishnan's own vocabulary: digital defense (core architecture) makes digital offense (new business models) possible — and the case shows Harley building offense faster than it finished defense.

Push the Consensus

Class will likely say "the dealer revolt was a communication failure." Push further: it's a structural incentive misalignment — dealers are compensated on foot-traffic-driven sales, and every digital efficiency gain that removes friction from a rider's path to purchase mechanically reduces exactly the traffic dealers are paid on.

Bridge to Session 1

The article's closing DBS callback is a gift for participation — explicitly connect Krishnan's "defense enables offense" framing back to DBS's own sequencing (core tech 2009–2015, AI experimentation 2013–2017, data centralization 2018+, scaled AI 2020+) from Session 1.

Taju's Edge — Redamo Labs

99.9% uptime and 95% CSAT for the enterprise IAM platform depended on getting the underlying verification architecture right before layering on new features — a direct parallel to York's "prove it here before it scales" discipline.

Taju's Edge — Prodigy Education

A/B testing infrastructure across a 150M-user platform only became possible once the underlying data pipeline was reliable — the same defense-before-offense sequencing Harley is now retrofitting under much tighter capital constraints.

Taju's Edge — Owo

Building a stock-valuation tool for NGX retail investors meant getting the underlying data architecture (accurate, current market data) right first — any "offense" feature (recommendations, alerts) built on shaky data would actively erode user trust rather than build it.

Block 9 — Reflections Journal Prep (Fill In After Class)

Concept Must Come From Live Slides

As with Sessions 2–4, the journal's concept half needs to be captured live from Prof. Mufti's Session 5 slide deck. The example half is ready to pair with whatever concept the session covers.

Concept — Fill In From Session 5 Slides

[3–7 word identification — write after class]

150–200 words, from class slides only. Architecture-focused candidates to listen for: technical debt, platform vs. point-solution thinking, build-vs-buy frameworks, or the sequencing of foundational versus tactical technology investment — write down whichever the professor actually presents.

Candidate Example — Ready to Pair With Whatever Concept Fits

Redamo Labs — Resisting a Feature Request That Skipped the Foundation

A client asked for a new personalization feature on top of the IAM platform before the underlying data model had been fully standardized across all 50,000+ users. Building it quickly was technically possible and would have looked like progress in a weekly status update. Declining to build it until the data foundation was consistent meant a harder conversation in the short term, but it avoided a much larger rebuild six months later, once the inconsistent data would have surfaced as broken personalization for a subset of users. The lesson mirrors Harley's own plant-by-plant approach: proving something works cleanly in one place, on a solid foundation, is worth more than deploying it everywhere quickly on an uneven one — even when the pressure to show visible progress points the other way.

Block 10 — Key Takeaways

What to Walk Away Knowing

Architecture work makes existing constraints visible — it doesn't create them. Krishnan's own reframe ("digital success exposed constraints earlier, which is a feature, not a failure") is the session's sharpest one-liner.

Foundational and tactical investments compete for the same limited capital. Harley's $115–225M annual capital range had to cover tooling, automation, and product redesign simultaneously — architecture never gets a blank check.

Uneven architecture maturity creates uneven adoption, independent of the tool itself. The York/Milwaukee/Tomahawk gap mirrors Pernod Ricard's D-STAR/Matrix adoption gap from Session 4 — the pattern recurs at the plant level, not just the product level.

Stakeholder trust can break faster than architecture can be fixed. The dealer revolt and CEO departure happened on a faster timeline than the multi-year modernization program — a reminder that technical sequencing decisions have to account for political and relational clocks too.

Looking Ahead — Session 6: Digital Implementation

→ Session 6 (Implementation)

Session 5 asks "what architecture is foundational?" Session 6 (Discovery-Driven Digital Transformation; ANZ Bank case) asks "once the foundation and organization are ready, what implementation methodology actually ships it?"

↔ Recurring Thread: Prove, Then Scale

York's "prove it here before it scales" rule and Pernod Ricard's TLO (test-learn-optimize) periods from Session 4 both anticipate Session 6's discovery-driven implementation logic directly.

Memo #4 Due Before Session 6

The team's fourth Case Study Memo (ANZ Bank) is due the Thursday before Session 6 at 11:59pm.