Context is king. 5 takeaways from the Times Tech Summit and what they mean for marketers

Written by Simon Spyer | Nov 19, 2026, 3:21:24 PM

44% of UK SMEs and mid-market businesses now use AI, according to NatWest's first AI Adoption Report, a survey of 1,400 firms published in September.

Yet only 6% say AI is reshaping their operations, decision-making and customer experience.

Scott Marcar, Group CIO at NatWest, brought that gap to the stage at the Times Tech Summit. It was the most useful figure I heard all day. Most of the sessions that followed explained some part of why the gap exists.

My reading of the day for marketing leaders is this: in 2027 the brands that gain most from AI will be the ones with clean customer data, brand rules documented and consent that an agent can read. Marcar said the AI available today is already good enough to reinvent most industries. For most marketing teams, the work that remains is in their own data and operating model.Most businesses use AI without changing how they work

The NatWest report also found that 95% of firms in its transforming group report higher revenue, against fewer than 3 in 10 of those at the earliest stage of adoption.

Rob Hornby, Global Co-CEO of AlixPartners, said fewer than 1 in 10 companies have an agent doing consequential work in live service. Many are keeping their experiments inside HR and finance for now.

Pip White, who leads Anthropic across the UK, Ireland, Northern Europe and Israel, said the enterprise customers she works with are moving from using AI as a chatbot to rebuilding core processes. Her example was Barclays, which uses Claude to handle around 120,000 inbound customer emails a day.

For a marketing team, a stack of licences and a few enthusiasts puts you in the 44%. Getting into the 6% means changing a process end to end, such as how a segment is built, how a campaign is briefed or how performance is read.

Clean data and written-down brand rules decide who moves fastest

Marcar said that without good, well-managed, well-controlled data the AI is "pretty useless", producing hallucinations and wrong outcomes. He added that around 80% of the automation in NatWest's process work has turned out to be ordinary deterministic software, with large language models used for the steps that need them.

Nathan Hancock, who runs Adobe across the UK, Ireland, Middle East and Africa, applied the same point to marketing. The enterprises he sees taking off fastest have spent time getting their data right and codifying their brand guidelines, strategy and narrative. He said much of this often lives as tribal knowledge in the team's heads. An agent can't use it and nobody can check the agent's output against it. Once that work is done, Hancock expects the technology to become "the easy part".

This is where most marketing teams can move first. Your data already contains the answers. For one ecommerce brand we built a single customer view in 48 hours and a growth plan in 72, identifying around £3.9M of growth potential from data the business already held.

Personalisation at agent speed makes consent the hard limit

Hancock described the stakes with what he called the Friday afternoon barbecue. The forecast says 15 degrees and rain, then the weather breaks and it's 20 degrees and sunny. A retailer with 20 million loyalty members has about 30 minutes to respond.

Done well, that means billions of engagements across every channel, each matched to what the customer eats, where they live and how they like to be contacted. Hancock put the upside at tens of millions in that window. The downside is breaching trust or consent at the same scale, with the regulatory and brand damage that follows.

An agent can only respect the rules it can read. In 2027, consent records, contact preferences and suppression lists become campaign infrastructure. If they sit in 3 systems that disagree, an agent can't act inside a 30-minute window without risking a breach.

On the AI safety panel, the advice for businesses was to give an agent tightly scoped access, define the outcome it's working towards and make everything it does visible. An agent writing subject lines has no need to see payment data.

Measurement has to keep up once agents start buying

Evan Spiegel said around 77% of Snap's advertising business is direct response. He expects the platforms that win to be the ones that can prove a conversion happened, whatever share of traffic comes from AI agents.

That matters to anyone who signs off media budgets. Attribution models were built around people clicking through journeys. As agents start researching and buying on a customer's behalf, more of the path to purchase will be invisible to those models.

The practical step for 2027 is to treat incrementality as a standing capability. Hold-out groups, geo tests and media mix models measure outcomes without depending on a clean clickstream. A team that can show what its spend caused will keep its budget when agent traffic makes click-based reporting less reliable.

Smaller, more senior marketing teams are coming

Hancock said Adobe is already hiring smaller, more senior teams. He expects "100x marketers": people who know the tools well enough to deliver 100 times what they could before. He wants managers to set the outcome and leave the method to the team.

Spiegel gave a figure from his own business. Merged pull requests at Snap doubled year on year while the team shrank by about 30%.

Marcar said NatWest hired more UK graduates last year than ever before, plus 450 engineering graduates worldwide. It's also spending heavily on retraining the people it already has. White expects junior staff to manage fleets of agents much earlier in their careers than they'd have managed people.

For a CMO, the 2027 budget question is how many of your team can direct an agent, check its output against the brand rules and judge the result.

The work starts before you choose a model

Every session pointed back to work that sits with the marketing team: clean data, written-down brand rules, consent an agent can read and a way to measure what's working. None of it depends on which model you choose. All of it can start this quarter.

DiagnosticAI is where we start that work with clients. It's a one-week, fixed-fee audit of your customer data, tech stack and channel performance. AI agents run it on your live data and it ends in a clear action roadmap.

Want to see how Data Agents can help you do more with less? Book a quick strategy call.