AI Partnerships That Speed Up Innovation (Lessons for SG)

A US$2.75 billion number tends to focus the mind. On 30 March 2026, Reuters reported that Eli Lilly expanded its AI-powered drug discovery partnership with Insilico Medicine, adding an exclusive worldwide license for certain oral treatments currently in preclinical development. Insilico will receive US$115 million upfront, with the rest tied to development, regulatory, and commercial milestones—plus tiered royalties if products make it to market.

If you run a business in Singapore, you’re not in pharma. You probably don’t have a wet lab. But this story still matters to you more than most “AI trend” headlines.

Because the headline isn’t “AI finds drugs faster.” The headline is: serious companies are using partnerships to turn AI from experiments into execution—with clear commercial terms, shared risk, and measurable outputs. That’s the part Singapore SMEs can copy.

Snippet-worthy truth: AI doesn’t create value by existing. It creates value when it’s attached to a workflow, a team, and a decision that moves money.

What the Lilly–Insilico deal really signals about AI adoption

The fastest way to understand this deal: Lilly is buying speed and certainty, not novelty.

Drug discovery is slow, expensive, and full of dead ends. Even after a target is chosen, most candidate compounds fail somewhere between preclinical work, clinical trials, and regulatory approval. That’s exactly the kind of environment where AI earns its keep: narrowing search spaces, prioritising hypotheses, and improving the odds that the next experiment is worth running.

In the reported agreement, Lilly will use Insilico’s AI engine and gets an exclusive worldwide license to develop, manufacture, and commercialise specific oral therapies in preclinical stages. This isn’t a “pilot.” It’s an operating model: AI partner produces candidates and insights; pharma partner turns them into approved products.

Why milestone-heavy pricing is the tell

A big number gets the attention—up to US$2.75B—but the structure is the story:

• US$115M upfront: enough to fund serious work now.
• Milestones: payments only trigger if the project hits defined outcomes.
• Royalties: alignment continues after launch.

That structure is how executives manage AI risk: pay for progress, not promises. In Singapore terms, it’s the difference between “we bought an AI tool” and “we bought a result.”

The bridge to Singapore: AI is a throughput problem, not a hype problem

Most Singapore SMEs approach AI like a tech procurement exercise: pick a tool, get a demo, hope staff uses it.

The better approach—what this partnership illustrates—is throughput thinking:

• What’s the bottleneck in our revenue engine?
• Where do decisions slow down because information is messy?
• Which steps are repetitive, rules-based, or pattern-heavy?

Drug discovery has bottlenecks (target identification, biomarker selection, candidate generation). Your business has them too:

• Sales: slow lead qualification, inconsistent follow-ups
• Ops: manual scheduling, inventory decisions made on gut feel
• Finance: invoice matching, anomaly detection, cashflow forecasting
• Customer service: long response times, inconsistent answers

AI business tools in Singapore are most valuable when they remove friction from these chokepoints and push more work through the pipe—without hiring at the same rate.

A practical mapping: “biomarkers to models” → “signals to decisions”

Insilico’s CEO described deploying AI “from biomarkers to life models.” You don’t need “life models,” but you do need the business equivalent: a way to move from raw signals to better decisions.

Here’s the mapping I’ve found useful:

• Signals: customer enquiries, web traffic, POS data, delivery timings, support tickets
• Models: churn prediction, demand forecasting, lead scoring, ticket routing
• Decisions: who to call, what to stock, which promo to run, which tickets to escalate

When people say “AI adoption,” this is what it should mean in practice.

Partnership models Singapore SMEs can copy (without billion-dollar budgets)

Lilly didn’t try to build everything internally. They partnered—first via software licensing (2023), then a research collaboration (Nov 2025), and now an expanded commercial deal (Mar 2026).

That gradual deepening is a pattern worth copying.

Model 1: Start with a narrow AI tool + clear KPI

Good for: marketing ops, customer support, admin automation.

Examples of narrow scopes:

• Reduce first-response time in customer service from 6 hours to 30 minutes
• Increase sales qualified lead (SQL) rate by 15% via lead scoring
• Cut invoice processing time by 40% with automated extraction and matching

The rule: don’t start with “AI transformation.” Start with one metric that matters.

Model 2: External AI partner as your “applied R&D team”

Good for: companies with unique data (and messy processes) where off-the-shelf tools don’t fit.

What this looks like locally:

• A vendor builds a lightweight data pipeline (CRM + accounting + e-commerce)
• They train/use models for forecasting, segmentation, or anomaly detection
• You keep ownership of your data and operational decisions

You’re not outsourcing your brain—you’re outsourcing the build so your team can focus on running the business.

Model 3: Outcome-based contracts (milestones) to control risk

This is the big pharma move, scaled down.

Instead of paying fully upfront, structure agreements around:

1. Baseline: measure current performance (2–4 weeks)
2. Milestone 1: proof in a controlled workflow (4–6 weeks)
3. Milestone 2: production rollout with monitoring (6–10 weeks)
4. Milestone 3: measurable lift sustained for 1–2 cycles

If a vendor won’t talk milestones and measurement, you’re buying theatre.

What “AI-powered discovery” teaches about your data and processes

AI in drug discovery is glamorous, but it’s also brutally operational. Models only help if the underlying inputs are usable and the feedback loop is real.

Singapore SMEs hit the same wall. The wall isn’t the model—it’s the workflow.

The minimum data hygiene that makes AI tools actually work

You don’t need perfect data. You need consistent data.

• A single source of truth for customers (even if it’s just one CRM)
• Standard fields (industry, deal stage, product type, margin, channel)
• Timestamped events (lead created, contacted, quoted, closed)
• A habit of closing the loop (won/lost reasons, ticket categories)

AI tools fail quietly when teams treat data entry as optional.

The “human-in-the-loop” step you can’t skip

Drug candidates don’t go straight from model to market. Scientists test them.

In business, the equivalent is:

• Let the AI recommend, but have a person approve at first
• Track exceptions (when staff overrides the AI)
• Use overrides as training data for refinement

Snippet-worthy truth: The safest AI system is one where humans can disagree—and the business learns from the disagreement.

A 30-day AI adoption plan for Singapore businesses (simple, not easy)

If you want to apply the lesson from Lilly–Insilico using AI business tools in Singapore, here’s a plan that works without turning your company upside down.

Week 1: Pick one bottleneck and one owner

Choose one workflow that:

• happens daily/weekly,
• affects revenue or cost,
• has repeatable steps.

Assign a single accountable owner (not a committee).

Week 2: Define metrics and a “before” snapshot

Write down:

• baseline cycle time (e.g., ticket resolution time)
• baseline cost (hours spent)
• baseline quality (CSAT, error rate, conversion)

No baseline = no proof.

Week 3: Run a controlled pilot with real users

Keep it small:

• one team,
• one channel,
• one dataset.

Make the AI produce outputs inside the workflow (CRM, helpdesk, accounting system), not in a separate dashboard nobody checks.

Week 4: Decide: scale, iterate, or stop

Be disciplined:

• Scale if you hit the KPI and users stick with it.
• Iterate if results are promising but inconsistent.
• Stop if it’s adding steps, not removing them.

That last option is healthy. Pharma kills candidates all the time. That’s how they win.

People also ask: “Will AI replace my team if I adopt it?”

No—and the companies who adopt AI fastest usually keep hiring.

What changes is the hiring profile:

• fewer purely administrative roles,
• more customer-facing and judgment-heavy roles,
• more operators who can run systems and interpret outputs.

The business goal isn’t headcount reduction. It’s capacity expansion: more quotes sent, more tickets solved, more campaigns shipped, more financial control—without chaos.

Where this fits in the “AI Business Tools Singapore” series

This post is part of our AI Business Tools Singapore series, where we focus on practical AI adoption for marketing, operations, and customer engagement.

The Lilly–Insilico partnership is a clean example of the direction the market is moving in 2026: AI is becoming a supply chain for decisions. The winners won’t be the firms with the most tools. They’ll be the firms with the clearest workflows, the strongest measurement habits, and the smartest partnerships.

If you’re exploring AI for your Singapore business, don’t start by asking, “Which model is best?” Start here: “Where do we lose time every week—and what would it be worth to get it back?”