The Enterprise in 2030: What IBM's research is telling us — and what we would add

IBM's The Enterprise in 2030 report lands on an uncomfortable truth: most enterprises today are bolting AI onto existing processes to automate tasks and optimise workflows. It produces incremental gains and gives the CIO a clean ROI slide — but it won't deliver durable advantage.

A few themes from the report stood out to me, and I've added my perspective wherever applicable.

1. Generic AI won't create differentiation. The winners will build tailored AI — agents, models, and data that encode the organisation's specific business logic, purpose, and culture. The moat isn't model power or headcount cuts; it's AI that knows the specific business, reflects its values, and amplifies its people. Generic Copilots or COTS agents don't differentiate; orchestrated agents tuned to a firm's DNA do.

2. Reframe the human-AI question. IBM suggests leaders stop asking "where can AI replace people?" and start asking two sharper questions: where does AI augment people, and where do people augment AI? The second half of that pair is the one most leaders are skipping.

3. Governance is a scaling lever, not a brake. This one I'd underline. Speed and responsibility aren't a trade-off — weak governance is precisely what stalls enterprise AI at the pilot stage.

4. The future is multi-model. Per the report, 82% of executives expect their AI stack by 2030 to combine LLMs for general reasoning with SLMs tuned for specific functions.  

What this means in practice:

The report's prescription is pragmatic: pick the 3–5 capabilities where AI can meaningfully shift customer relationships, specialised knowledge, or operational economics; get your proprietary data AI-ready; and map the models and assets that genuinely amplify those few capabilities. Add to that a neutral orchestration layer, governance built in from day one, rapid experimentation loops, and AI fluency at the senior leadership level — and you have an executable agenda.

I've seen the front end of this play out at a bank that chose hyper-personalised marketing as its priority capability. The sequence:

Data unification and enrichment: Customer data sat siloed across deposits, loans, cards, mutual funds, and daily transactions. They consolidated it, then enriched each transaction with merchant detail, location, date/time, amount and transaction type.

Pattern intelligence: From the enriched base, they surfaced individual-level patterns — spending, income, savings, investment behaviour, channel preference, and risk markers like overdrafts and near-defaults.

Life-stage detection: Those patterns translated into life-stage signals: a parent likely to need an education loan next year, a frequent overseas traveller, a newly married salaried customer entering home-buying range.

Offer personalisation: Outbound offers were then tuned to the detected life stage rather than broadcast by segment.

The programme is still in flight, so too early to claim numbers — but the direction is clearly positive, and the build itself tells you why. The AI is grounded in the bank's own data, its own customer logic, and its own definition of a good outcome — exactly Key Idea #1. It also reflects Key Idea #2: the model surfaces signals, but relationship managers decide which conversations to actually have.

To borrow the spirit of the report's conclusion, the companies that win in 2030 won't be the ones who deployed the most AI. They'll be the ones whose AI carries their fingerprint.

Source: 'The Enterprise in 2030', IBM Institute for Business Value.