1. What shifts are you observing in ecommerce traffic and consumer behaviour, and how do these differ from the traditional search journey?

The clearest visible shift is that ecommerce site visits are growing faster than unique visitors. Over the last twelve months, worldwide web visits grew 6.8% against 5.8% for visitors. The year before, the pattern ran the other way (+1.3% visits, +4.5% visitors). The effect is sharper in some markets. In the UK, visits grew while unique visitors fell; fewer people are arriving, but they are browsing more intently.

AI is not the sole cause, as consumers are doing more research before committing generally, but it feeds directly into that behaviour. Nearly half of US consumers (49.1%) say AI has changed how they browse, and AI’s presence in journeys that end at a transactional retail site has risen from 4.5% to 11.4% in under two years.

The difference from a traditional search journey is friction. AI lets a shopper shortlist, compare and get tailored advice in one conversation, in what feels like a neutral environment. Previously that meant going back and forth between retailer sites, buying guides and affiliate content, where the impartiality of the advice was often in question.

2.  What does shorter discovery-to-purchase mean for retailers that relied on multiple visits, product-page engagement and retargeting?

AI may shorten the path, but it does not reduce engagement overall. Journeys that combine AI and Search span an average of 14.9 sites, against 10.1 for Search alone. AI-only journeys are shorter at 7.3 sites, but they account for just 11% of journeys that use AI. What changes is where the engagement happens. Shoppers may reach a retailer’s site directly only at the final purchase, which moves much of the relationship away from the retailer.

That does not make retargeting ineffective. The shopper may simply use AI to research the very products you retarget them with. The retailer no longer owns the whole journey, but it still owns the product details, pricing, fulfilment options and any existing customer relationship that drive the final purchase and repeat business.

3 & 4. How should ecommerce businesses understand the relationship between AI discovery and search, and what does it mean for SEO and paid strategy?

They should not treat them as separate channels. In 89% of journeys where AI appears, Search is present too. Consumers are stacking tools, not switching between them. AI sits disproportionately in the middle of the journey, appearing as a middle touchpoint in 76% of the journeys where it is used, against 23% as the first touchpoint and 18% as the last. Search is spread far more evenly, appearing at the start, middle and end of at least half of the journeys where it is used. The most complex journeys, those using both, convert best. AI is a new touchpoint, not a replacement destination.

Established SEO and paid practices remain effective. Each channel still needs its own focus, but they should feed one strategy. The question is no longer “how are we performing in SEO and paid?” but “what is the combined result of our efforts across AI visibility, search and paid?” Retailers should rethink the following:

●     Product data: detailed enough for an AI tool to reason over, with descriptions built for both humans and LLMs.

●     Brand consistency: the same story across paid social, Google and the retailer’s own site, so it holds whichever tool a shopper reaches for.

●     Marketplace-native assistants: conversational assistants such as Amazon’s Alexa for Shopping are growing faster than external AI referrals. Traffic to Amazon product pages from the assistant was up 537% year on year in June 2026, against +203% for external Gen AI referrals to ecommerce sites. They apply their own rules on who gets seen, favouring sponsored placement and purchase history, which traditional SEO logic doesn’t reach.

5. How does AI influence before the site visit complicate attribution, and are existing models equipped for it?

No, they are not. Traditional attribution significantly undercounts AI’s impact, and indirect influence has always been an issue. A product seen on social that leads to a branded Google search shows up as a search referral, for example. The misattribution is greatest for AI because shoppers use it to discover, evaluate and research, but rarely to navigate. They tend to take what they learned and then search for the brand or type the URL.

The scale of the gap is visible in our data. Gen AI referrals account for under 1% of worldwide ecommerce web visits, yet AI appears in 11.4% of US desktop journeys that end at a transactional site. Looking only at referrals from AI platforms measures the smallest and least representative part of AI’s role, and is the wrong metric.

6. What data should ecommerce leaders combine to understand the impact of AI shopping?

No single dataset will answer this, so leaders should combine four views:

●     AI visibility: how often and how prominently the brand appears in AI responses for relevant prompts, and which sources those responses cite.

●     Prompt intent and category influence: which types of prompts (research, comparison, transactional) and which topics within your category AI influences most (think Skin Care within Beauty).

●     Referral and traffic data: AI referrals,where the growth rate is a better gauge of adoption than the share, alongside direct and branded search trends that may reflect AI-influenced demand.

●     Journey-level conversion: how conversion differs between AI-only, Search-only and combined journeys.

Together these show not just how many visitors AI sends, but where it shapes the decision.

7. Does AI discovery translate into higher purchase intent, or does it just shift where consumers research?

Purchase intent may be the wrong question. It’s unlikely any channel or research tool creates intent. What AI does is increase confidence in a decision, and that is its greatest impact. It shows most clearly in complex journeys involving both AI and Search, where

23.0% of sessions that include a transactional site end in conversion, compared with 17.3% for Search alone and 12.5% for AI alone.

This is an association rather than proof of causation, but it fits the confidence argument: shoppers who combine AI and Search are cross-checking sources before they commit. To separate browsing from meaningful buying intent, retailers should look at conversion by journey type, not just AI-referred traffic volume.

8. What should retailers change first: product information, pricing, onsite experience or acquisition strategy?

Product information first, then pricing.

●     Product information: complete, clear data is what lets AI models make full comparisons, and it is the one lever the retailer fully controls.

●     Pricing: a shopper may ask a final question about where to buy at the best price. If a competitor is cheaper, the sale could be lost after all the preceding work.

●     Acquisition strategy: this isn’t a step away from acquisition. It is a strategy for acquiring AI-assisted shoppers.

●     Onsite experience: it perhaps matters least of the four, because if an LLM takes the shopper straight to the product, the wider site had little to do. The exception is checkout. After all the groundwork to get a shopper to that point, a single friction point can undo it.

9. What is the biggest assumption marketing leaders will need to revisit by 2027, and what data should inform it?

The biggest assumption is that the channels you can measure are the channels that matter. Many leaders still treat AI as a separate channel, judged by referral volume, and set it against search as a competing channel while the evidence points the other way. AI shows up in more than one in ten journeys to a transactional site, almost always alongside Search, and mainly works through the middle of the path. Its role is also broadening: its share as a middle touchpoint has fallen from around 81% to 76% since July 2024 as it appears more often at the start and end of journeys. Discovery is also moving to places traditional measurement doesn’t reach, such as marketplace-native assistants that apply their own ranking rules. Agentic checkout protocols are starting to let shoppers complete the whole purchase inside a single AI experience, which will widen the measurement gap further.

Leaders should revisit this assumption using:

●     Journey-level data: how often AI and Search co-occur, and how conversion differs between journey types.

●     AI visibility and citation data: where and how the brand appears across AI platforms. AI recommendations also carry beyond the session. In the report’s Finance, Travel and Beauty examples, users were consistently more likely to visit the AI-recommended brand than its competitor within seven days.

●     Marketplace-native growth: the growth of assistants inside marketplaces compared with external AI referrals.

●     Category data: AI’s role varies by category (14.1% of journeys in Consumer Electronics and 13.2% in Beauty, against 10.4% in Home & Garden). Growth is concentrated where innovation is becoming accessible, such as wearables, robot vacuums and K-Beauty, which win attention out of proportion to their shelf space by bringing premium technology to a mainstream price point.

The planning unit should shift from the channel to the whole journey, covering product data, paid media, on-platform AI and pricing together. Consumers do still want control over the final decision, but full agentic commerce is on the horizon. How quickly it will be adopted is still unclear, but it could change the game once again.

1. What are retailers getting wrong about turning holiday shoppers into long-term customers, and how can personalized follow-up help extend the relationship beyond the first purchase?

Retailers put a lot of focus on the holiday shopping rush, but they can miss the opportunity to build a relationship with customers who may be shopping with them for the first time. During the holidays, associates learn what customers are looking for, what they like and what matters to them. That gives retailers a starting point for future conversations, but those details often go unused once the holiday rush is over.

A personalized follow-up gives customers a reason to think about the brand again, whether an associate checks in about a purchase or reaches out with a product that fits their interests. Endear’s research found that 63% of shoppers are more likely to return to a store when an associate follows up with them personally, and more than half have made an additional purchase as a result. A customer who came in for a holiday gift might not have planned to shop with that brand again, but a thoughtful message can give them a reason to return.

2. How can retailers use customer data and AI to help associates deliver more personalized outreach at scale without making those interactions feel automated?

Retailers have a lot of information about their customers, but associates don’t always have the time to sort through it and decide who to contact. When an associate is helping customers on the floor and managing hundreds of relationships, it can be difficult to remember each person’s interests and find the right reason to reach out.

AI can help associates prioritize customers and suggest messages based on their purchase history and previous interactions. Associates can then review and personalize those suggestions before sending them, using what they know about each customer to make the outreach feel relevant. We found that retailers using AI for clienteling increased associate-led outreach sixfold in six weeks without adding headcount. This shows how AI can help associates reach more customers while keeping the personal touch that makes clienteling effective.

3. What role should AI play in clienteling, and how can retailers make sure associates remain at the center of the customer relationship?

AI can help an associate understand what a customer has purchased and suggest what might be relevant to them next. But purchase history doesn’t tell the whole story. An associate might know that a customer was shopping for a special occasion, was deciding between two items or had a specific concern that never showed up in their purchase history. That context can make the difference between a message that feels personal and one that feels like any other marketing email.

Retailers should use AI to support an associate’s judgment, not replace it. Associates can decide whether a suggestion makes sense, how to approach the customer and what they want to say. The goal is for customers to feel that the brand remembers them and understands their preferences, even when they aren’t in the store. AI can make that kind of personal attention possible across more customer relationships.

4. How do you see AI changing the role of the store associate as retailers look for new ways to build repeat customer relationships?

AI gives store associates more ways to stay connected with customers after they leave the store. Our analysis of more than 100 million SMS messages and 2.2 million clienteling interactions during Black Friday week 2025 found that customers who received tailored outreach from store associates were 30 times more likely to purchase than those who received traditional marketing emails or SMS blasts. Clienteling-driven revenue also grew 22% year over year.

These results show that the relationship between an associate and a customer can continue to influence shopping decisions long after a store visit. AI can make it easier for associates to keep in touch with more customers, share relevant recommendations and give people a reason to return. That expands the associate’s role beyond helping with the immediate purchase to building relationships that bring customers back over time.

1. Can you tell us about Underscore and the challenges biopharma marketers turn to your team to solve? How does your focus on rare disease and oncology shape your approach?

Underscore is a specialist media agency built for complex biopharma, particularly rare disease and oncology. Clients bring us the problems that are hardest to solve: finding small patient populations, reaching the right healthcare professionals, connecting fragmented data and preparing specialty brands to launch without wasting time or investment.
Those challenges shape how we approach the work. In rare disease, the patient population may be very small and diagnosis can take years. In oncology, treatment decisions can involve multiple stakeholders. We have to understand those realities before deciding how to reach people, what the data is telling us and where media investment can make a difference.
Our team has worked across more than 50 specialty launches. That experience helps us recognize patterns and ask better questions early, especially when a brand has very little room for wasted spend. We bring that understanding into the decisions we make with clients, so the media plan reflects the business and the people it needs to reach.

2. As AI makes sophisticated marketing tools more widely available, where does deep category expertise still change the quality of a decision in biopharma?

Technology can help teams analyze information and move faster and more precisely, but the value of AI depends on how that information is interpreted and applied. Biopharma recommendations must account for small patient populations, long paths to diagnosis and treatment decisions that may involve multiple stakeholders. Those conditions change how you plan media and evaluate performance.
Deep category experience changes the quality and speed of the decision: it helps a team recognize when a signal matters, when it is misleading and when the data simply does not know enough. AI can accelerate the work, but it cannot manufacture judgment or perspective that isn’t there. I believe that experts + carefully curated AI drive the best results. Being expert enables you to spot when AI is connecting signals that don’t have realistic potential or are rooted in unstable proof points.

3. In rare disease and oncology, patient populations can be small and the path to treatment complex. How should those realities shape the way marketers interpret performance signals and decide when to act?

Start by defining what the data can actually tell you and where its limitations are. A signal may look clear in a report but become much less useful once you consider the patient journey, the concentration of relevant healthcare professionals, access, or the commercial constraints facing the brand. For example, driving traffic to” brand.com” or “diseaseEd.com” may aid awareness if the people landing there are the right people, but without qualifying the visitors, you may be wasting resources without realizing it. I think it’s important to have everyone aligned on what success looks like and moving in unison with the same understanding of how the key metrics and KPIs drive action from the beginning.
This is also where category experience matters. Teams need to ask whether the data is thin, biased, delayed or easy to misread. They also need to understand what would change the recommendation. More data does not remove the need for those questions.
In a specialty launch, there isn’t room for wasted spend. The team interpreting performance needs to understand the business well enough to challenge an apparent success or recognize an early warning. That judgment should influence how the campaign is optimized and where the next dollar goes.

4. Agencies often describe AI in terms of efficiency. What should clients look for to understand whether those efficiencies are translating into better strategic thinking and more valuable work?

Clients should expect to see the value AI creates. It may show up as faster analysis, more frequent optimization or greater output. But the more important test is whether experienced people are spending less time producing the work and more time improving the decisions behind it.
I do not think lower fees are the automatic answer. The more useful question is whether AI is making the agency's contribution more valuable. Is the team getting to a useful answer faster? Does it have more time to pressure-test a recommendation? Can the client point to a change in how the work gets done?

An agency saying it uses AI is not enough. If the client cannot identify anything that is faster or more useful, it should ask harder questions about what the technology is delivering.

5. Much of an agency planner’s judgment has traditionally developed through hands-on research and analysis. As AI takes on more of that work, how can agencies build the experience their next generation of leaders will need?

We have to redesign apprenticeship intentionally. The risk is that agencies automate junior work, celebrate the efficiency and then discover a few years later that they have stopped developing senior thinkers.
I do not believe we should preserve manual work for its own sake. If a task can be responsibly automated and does not require human judgment, we should automate it. But junior people still need to learn how to frame a problem, evaluate evidence and understand why a recommendation was made. 
That means bringing them closer to the decisions sooner. What question are we really answering? What does the data not know? Where is the commercial risk? In healthcare, AI can synthesize information quickly, but teams still need to understand the reality of a patient journey and the constraints on a healthcare professional's attention. AI should help junior people become better thinkers sooner. We can accelerate how their focus on diving into more specificity rather than sorting and sifting more widely.

6. Underscore’s focus has narrowed over time toward complex biopharma. What did that evolution teach you about the difference between capabilities that travel across industries and expertise that becomes more valuable through specialization?

Our path was not perfectly linear. We started more broadly and had capabilities that translated across verticals. Diversification created growth opportunities, but it also helped us understand where our experience made the greatest difference to the quality of the work.
That became clearest in complex biopharma, including rare disease, oncology and specialty launches where the margin for wasted learning is small. We could ask better questions, see risk earlier and connect media decisions to commercial outcomes with more precision. Narrowing our focus was a decision to concentrate on the work where we had earned the right to lead.
There is also a personal dimension for me. Rare disease has affected my own family, so I understand why smaller patient populations deserve the same depth of attention, expertise and investment as much larger ones.

7. Category credentials and promises of senior involvement are common in agency pitches. How can a CMO determine whether that expertise will meaningfully influence the decisions made after the business is won?

A credentials deck full of category logos proves exposure. To understand the depth behind it, a CMO needs to ask how much of the agency works in the category, how long the people leading the account have been doing this work and what they have learned from situations like the client's.
The expertise should be evident in the conversation. Can the team explain where conventional wisdom breaks down? Does it know which data limitations matter? Can it describe what it would do differently because it has seen a similar pattern before?

Clients should also test the promise of senior involvement. Ask who will be working with you every week and what decisions senior leaders will personally help you make after the first 90 days. Then speak with existing clients about who shows up when the business gets difficult. That will tell you more than a staffing slide.

8. When a CMO reviews an agency roster, how can they look beyond satisfaction with today’s work to determine whether their partners are equipped for the commercial challenges ahead?

 I would start by asking where each partner is being asked to solve problems it is not built to solve. A roster review does not have to mean replacing everyone. It should clarify where each agency creates value and where the business has outgrown the original arrangement.

Look at the hardest commercial problems coming over the next two or three years and work backward. Those might include launch complexity, patient finding, payer dynamics or data limitations. Does the roster reflect where the business is going, or where it was when those agencies were hired?

There are also situations where a holding company is the better choice. A global CMO who needs creative, media, commerce, production and technology under one operating model may reasonably prioritize geographic scale and integration. A specialist has to earn its place by helping the client make better decisions with measurable impact. The right choice depends on the problem the client needs to solve.

1. You've identified an "insight gap" — the disconnect between how much marketers value customer feedback and how they collect it. From your position overseeing both communications and marketing, what's the organizational reality behind that gap?

The insight gap is secretly an execution gap. Our research found that 83% of business owners say customer feedback helps drive decisions, but only 54% collect it more than once or twice a year. That’s pretty telling. People value feedback, they just don’t always have the time or processes to collect it consistently. 
I get why. Marketing teams make decisions all day long, and many can’t invest in lengthy research projects to gather customer feedback for every single one of them. When listening feels separate from the work they’re already doing, it’s easy to just keep moving without it. The problem with that is assumptions can start to fill the space where evidence should be. And we all know that today, people’s thoughts, feelings, and preferences are changing as quickly as the news cycle.
We need to stop treating customer research as an occasional project and instead treat listening as part of the work itself. The best time to hear from customers is before the decision is made, while there’s still time to change the work.

2. The insight gap isn't new, marketers have always known they should be collecting more feedback than they do. What has changed about the consequence of ignoring it now versus three years ago?

Three years ago, there was more time between having an idea and putting it into the world. Today, AI can turn an idea into something that looks finished in less than a minute. That speed can be tricky because the end product looks great, but at best it’s a replication of what’s come before and at worst it’s the result of a hallucination. There’s no good AI replacement for asking customers what they want or think right now.
We’re seeing this all play out at work every day. Our 2026 State of Curiosity Report found that half of workers have had to redo work because the right questions weren’t asked at the start, and 46% have seen time or money wasted because assumptions went unchallenged.
The bottom line is: Listening needs to happen more frequently when execution gets faster.

3. "Not slowing existing operations or needing to hire a research firm" is a specific promise. What does a marketing team that has relied on either a research agency or a quarterly survey cycle looks like six months after embedding continuous feedback into their workflow?

Six months in, I’d expect customer feedback to show up in a lot more decisions. Instead of saving their questions for a quarterly study, the team can ask as questions come up: Does this message resonate? Which idea is stronger? Are customer expectations changing? Which brand name or tagline should we use?
AI makes all of this practical by taking on more of the mechanics of building surveys and analyzing responses, while benchmarks give teams immediate context for what they’re seeing. You don’t need a major research project every time you need an answer, but you do need a consistent groundswell of feedback, so it feels like your team is just swimming in the customer’s point of view all the time.
To me, that’s the exciting part. Curiosity becomes actionable. “I wonder” quickly becomes “let’s find out.” The answer comes back while you can still do something with it. 

4. AI survey creation removes the technical barrier to feedback collection. But survey design quality directly affects the reliability of the data. How does SurveyMonkey's AI handle that tradeoff and what does the system do to ensure speed doesn't produce low-quality signal?

Speed is only helpful if you can trust what comes back. A badly designed survey is still a badly designed survey. It won’t produce reliable results.
With Today’s SurveyMonkey, a user can describe what they’re trying to learn in plain language, and SurveyMonkey AI helps build the survey, adjust questions, apply logic, and refine it. Our proprietary AI makes the survey science and research expertise we’ve built in our platform over the last 27+ years more accessible and easier to implement for the business user.
AI can also analyze patterns across responses for you. That gives marketers more time to focus on what they’re actually trying to understand, whose perspective they may be missing, and what they should do differently because of what they heard. That’s where I want AI doing the work. Let it handle more of the mechanics so marketers can spend more time being creative, crafting the questions, and connecting with the humans behind the answers.

5. “Listening faster" and "acting with confidence" are outcomes that require the AI to surface the right signal at the right moment rather than just generating data. What does that signal-to-noise problem look like in practice and how does your approach address it?

Most marketers already have plenty of data. We don’t need more for the sake of having more. We need to know what deserves our attention and what it means for the decisions in front of us.
When SurveyMonkey revisited our company values recently, AI helped us analyze qualitative employee comments and spot gaps between intended values and employees’ actual experiences. Those insights gave us something concrete to act on and helped shape the values we landed on.
The same principle applies to quantitative feedback. A score becomes much more useful when you know how to interpret it. Our upcoming Benchmarks experience puts results in the context of hundreds of thousands of other company data points, helping teams understand what “good” looks like relative to their industry and peers.
That’s a great job for AI. Help me make sense of what we already know, so I can spend my time with my team deciding what needs attention and what to do about it.

6. Integrations with Claude, ChatGPT, and Workday mean feedback collection can live inside the workflows where decisions happen. In practice, what does that look like and what decision does it improve?

We firmly believe human input still matters for every business on the planet. Our goal is to make it easier to get real human input when you need it. If a marketer is already using Claude or ChatGPT, they can create a survey, analyze responses, and access customer insights without leaving that environment. Workday does something similar for employee feedback. Data such as department, location, and tenure can flow directly into SurveyMonkey, so teams can see where experiences differ across a team without manually connecting the dots in the data themselves.
When teams can ask for and analyze feedback inside the tools they already use, it’s much easier to make listening part of the work. A marketer deciding between two messages shouldn’t have to start a separate research project, come back days or weeks later, or download and combine multiple Excel docs into one. They should be able to ask, hear from real people, and use that input while there’s still time for it to shape their decisions.

7. The pre-launch survey has been the default feedback checkpoint for most marketing teams before a campaign goes live. You're arguing that model is insufficient. What does a marketing team lose by treating listening as a checkpoint rather than a continuous system?

A lot can change after a campaign launches. Customer expectations evolve, competitors respond, and the market keeps moving. The feedback you gathered before launch may not reflect what customers think a month or six months later. When a pre-launch survey becomes your only listening checkpoint, you lose the chance to catch those shifts and respond to them.
Another problem I see all the time in marketing is that we ask too late. Teams do the work, get attached to the idea, then ask customers what they think. By then, everyone has an opinion or a favorite, and changing course gets a whole lot harder and more expensive. That’s exactly when you wish you’d asked earlier. 

8. Customer listening embedded in workflow only creates value if the insights reach the people making the decisions in time to change them. What does the organizational infrastructure need to look like for that to happen?

Technology is part of the infrastructure, but how the team operates matters just as much. Insights need to reach the right people early enough to influence the work. Then comes the hardest part: people have to be willing to change their minds when the evidence tells them they’re wrong. 
Our State of Curiosity research found that while 95% of workers say they’re at least somewhat curious, only 30% say their workplace strongly rewards curiosity. That willingness to ask another question, challenge an assumption, and change course based on what you hear is what turns customer feedback into better decisions.
You can build the perfect listening system, but it won’t do much good if your culture punishes the person who asks the hard questions.

9. The market right now is full of AI tools that promise to tell marketing teams what customers want without asking them. You're taking a different position that AI should amplify human feedback, not substitute for it. What's the specific failure mode you're trying to prevent?

I worry about what happens when an AI-generated assumption starts looking enough like an insight that we stop checking whether or not it’s true.
AI is incredibly good at giving us a plausible answer based on patterns in what already exists. But if I want to know whether our customers understand a new message, what they think about a product idea, or why their expectations are changing, I still have to ask them.
AI can help me ask better questions, listen at scale, and make sense of what I hear. But I don’t ever want to lose the human connection that has driven marketing for decades. I still want to deeply know our customers, because that’s where all the great storytelling begins. There’s just no substitute for listening to humans.

1. Most sales organizations know their CRM and technology stack is creating friction. What does the Seller Experience Audit surface that internal teams have missed or avoided naming?  

Most sales organizations know something is slowing reps down, but they tend to misdiagnose it as individual rep performance, or lack of discipline. The Seller Experience Audit gives us a clear, evidence-based view of exactly where the CRM and sales tech stack are creating friction, not just anecdotally, but mapped against actual rep behavior and performance data. 

It's rarely one broken tool. What we usually find is a pile of workarounds reps built over time because the CRM stopped reflecting reality. Things like shadow spreadsheets, manual reports, private Slack messages, and managers who quietly stopped trusting the pipeline numbers. Critical deal context gets trapped in silos and disconnected tool stacks. Leadership may think poor data quality or low system usage as a performance problem when it's really a design and adoption problem. The audit gives organizations clear evidence to see that gap and a roadmap that separates the quick wins from the deeper fixes.

2. A seamless buyer experience is due to a well-functioning seller experience. When you're making that case to a sales leader focused on productivity, how do you connect internal friction to external outcomes that changes the level of fixing it?   

To change the level of urgency, we highlight how internal system friction directly translates into buyer pain. We show them where the rep's time actually goes and quantify the exact revenue capacity leaking out of their pipeline. If a rep spends twenty minutes rebuilding account context before a call because the CRM data is stale, that shows up in the call itself. Buyers can tell when someone hasn't done the homework, or worse, when they reference something wrong because the system gave them bad information. A clean seller experience is really what makes a seamless buyer experience possible.

3. The data layer affects marketing, customer success, and forecasting as much as it affects sales. How do you build the organizational case for treating the data layer as shared infrastructure rather than a sales team tool?   

The data layer is the foundation the entire revenue organization builds on. When CRM adoption is low or intelligence is fragmented across spreadsheets, it creates blind spots for everyone downstream. Marketing can't personalize because customer data is incomplete, or outdated, causing customer signals to go to waste. Customer success can't proactively engage accounts because they don't have visibility into what happened pre-close, and forecasting becomes guesswork because the underlying data was never clean to begin with. 

We saw this play out firsthand with Little Caesars Fundraising, where we completed an extensive Seller Experience Audit and Salesforce CRM optimization. Andrea Fenton, their Director of Marketing, put it simply, that the CRM optimization gave them efficiency, planning, and transparency they hadn't had in years. Once leadership sees that cost stacking up across every function, the conversation shifts from whose budget this is to who's accountable for the shared asset.

4. The RAND study found that 87% of AI pilot failures are due to poor data foundations. What does an AI implementation built on a fragmented data layer look like when it starts to fail?   

Every sales organization wants to make its team more efficient, accelerate cycles, and drive growth with AI, but if optimizing the data layer isn't a priority while you're implementing those high-impact use cases, the results are going to be severely limited. If AI is implemented without the right data foundation, it just accelerates sales teams to work from bad data, leading to downstream impacts Generative outreach tools can impact brand credibility by drafting "personalized" pitches based on duplicate contacts or outdated job titles, resulting in reps pitching to former employees or offering products an account already owns. Furthermore, when reps enter fake data to bypass validation rules, AI forecasting engines generate wildly inaccurate revenue predictions and deal scores. Without clean, context-rich data inputs, AI summaries and recommendations ultimately yield generic, low-value advice that reps quickly learn to ignore.

Layering AI on top of a cluttered CRM "junk drawer" amplifies underlying data issues, leading to misaligned conversations with buyers and starts eroding their trust.

5. If a sales organization has already deployed AI on top of a data layer that isn't ready, what does the remediation path look like, and how do you address the confidence problem in the team? 

Remediation starts with refocusing on the use case the AI was actually meant to solve, then mapping out exactly what data and context that use case needs to work properly. From there, we run a focused audit on the data feeding that specific use case, sorting out what's clean, what's missing, and what needs to be fixed before re-exposing to the AI.. Rather than trying to repair the entire data layer at once, we fix what that one use case depends on, get it working reliably, then expand from there.

Rebuilding confidence is just as important as fixing the data. It takes real transparency about what changed, a visible before and after on data quality, and a low-stakes environment for reps to test the tool again before trusting it in front of a buyer.

6. The promise of Headless 360 is that reps manage workflow from wherever they already work while a single CRM backend maintains the record of truth. How does that get addressed at the product and process level?    

On the product side, it comes down to exposing the CRM’s data, logic, and workflows through APIs and MCP-style integration layers so teams can act directly from where they spend their workday - whether that's Slack, Teams, email, calendar, or call platforms - without ever logging into the CRM. This isn't point-to-point plumbing per tool. It's standardized services any channel can call, so business logic isn't rebuilt five different ways. Critically, permissions, validations, and approval logic travel with the API call, so a rep updating a deal from Slack is governed by the same rules as in the console, which is what keeps the backend consistent across channels.

On the process side, workflows built around a clicking through specific CRM screens are redesigned into event-driven processes.  Pipeline stages auto-advance on triggers, like a proposal generated or a follow-up confirmed, rather than manual field updates. Rules and paper trail apply no matter where the action starts, ensuring the CRM backend remains the trusted single source of truth no matter where the action originates.

7. Work-in-place selling changes what reps spend time and spend more time in sales conversations. How do you measure that shift in a way that a revenue leader can hold up as evidence the investment in this architecture is producing returns?  

We measure ROI by tracking time reallocation, how much time reps used to spend on CRM admin versus how much of that now converts into active selling time through seamless work-in-place tools. This directly drives core performance metrics like pipeline growth, deal velocity, and average deal size, while also streamlining adjacent processes such as lead qualification, onboarding, and customer satisfaction (CSAT), as seen when Little Caesars Fundraising cut their lead management stages from 24 to 5, boosting conversion by 8%

8. Dreamforce '26 is expected to shift the conversation from AI chat features to the fully agentic enterprise. For a sales organization evaluating its readiness, what is the most honest self-assessment they should run before that becomes the expected standard?  

Honest self-assessment isn't about which agentic tools to buy. It's asking whether your data and workflows are clean and consistent enough that an agent could execute a multi-step process on its own without human intervention.. Reps often mask poor CRM hygiene by filling in missing context on the fly, or inconsistent processes with manual judgement calls. Audit your sales data, like open pipeline records to verify that deal stages, next steps, and ownership are clean enough, and if workflows are genuinely rule-based versus judgment-based to support unsupervised AI and automation. 

9. The emphasis on Data 360 grounding AI agents is the argument that connects everything you've described. As a Product Director, how do you bring those pieces into an implementation roadmap for an organization that is trying to do all of it at once without doing any of it well?    

The roadmap has to be sequential. Start with use case ideation and prioritization to help define that roadmap, not parallel workstreams. Generate use cases that will solve for challenges and areas of friction today. Evaluate those use cases on value versus complexity to implement, including the data readiness, the systems, the processes and the people impact. Prioritize low-complexity, high-value candidates; flag high-value/high-complexity for later discovery. That prioritized list becomes the roadmap. For each use case, align on KPIs and how you will measure before any build starts so there is a clear definition of success, and define a change enablement plan to ensure team adoption. A short list done well, sets the foundation that builds momentum rather than everything at once.

1. Private markets fraud isn't new, but the AI layer is. Was there a specific incident or pattern you encountered that convinced you this was the moment to act? 

Two things caught my attention. The first was that wire fraud kept growing faster than the security tools and protocols built for private markets. The second came out of our discovery process. Every firm we interviewed executed fund flows in pretty much the same way, and AUM made no difference. It was spreadsheets, human review of PII, callbacks, bank portal uploads, bank templates, and emails back and forth. Sensitive data ended up sitting in multiple places, and every one of those places is an attack vector. Once AI made impersonation cheap, a process that looked identical at every firm became something an attacker could learn once and use everywhere.

2. Walk us through what an AI-driven fraud attempt looks like today. What does a convincing voice-cloning or deepfake attack look like in practice? 

It rarely looks dramatic. The head of a company has a phone call with someone who reproduced his/her boss's voice, accent, and cadence. Nothing he heard gave him a reason to doubt it. He or she then wires hundreds of thousands of dollars. Attackers are building entire video calls of familiar colleagues. What makes these attacks convincing is preparation. Attackers study email threads, communication patterns, and travel schedules before they strike, so the request arrives with the timing, tone, and formatting of every legitimate one before it.

3. Business email compromise has been around for years, but you've said it's now moving faster than compliance frameworks can track. What's changed on the attacker side that's outpacing the defenses? 

The grammatical red flags compliance teams trained employees to spot are gone. Compliance frameworks are built around known controls like callbacks, PDF instructions, and inbox approvals, and attackers now go after those. They can alter wire instructions that are in transit or clone the voice of someone who is on the other end of the callback. One of our clients got a call from an LP asking to confirm capital call instructions, which is routine in a manual process. The GP had never issued a capital call. The LP had received a spoofed email convincing enough that they were about to wire funds to a fraudster, and the GP only found out because the LP happened to pick up the phone. The LP’s problem quickly became the GP’s problem too. 

4. Private market transactions often rely on relationships and trust built over years. How does that trust become a liability when someone can impersonate a known counterparty? 

Familiarity is what social engineering is built to exploit. The more confident a team is that they'd recognize their fund administrator's voice, or their LP's writing style, the less likely they are to double-check. That’s what the attacker counts on. 

Trust built over years is valuable, but private markets need to keep the relationship, and verify identity and instructions independently.

5. Are certain players more exposed than institutional players with dedicated compliance teams? Where's the weakest link right now? 

Yes. About half of family offices have been hit by a cyberattack in the last 2 years, and 63% of them carry no cyber insurance at all. Firm size is only part of it, though, because exposure runs along the whole chain. A GP can have excellent internal controls and still lose capital because an LP's family office, a wealth manager, or the law firm coordinating a closing didn't. Attackers look for the weakest link and work from there. That's why separation of duties and verification have to reach everyone who touches the money, beyond your own four walls.

6. Fraud prevention often creates friction, extra verification steps, delays, additional approvals. How do you build protection into capital call and distribution workflows without slowing down transactions that need to move fast? 

You stop treating verification as a step and start treating it as a layer that runs underneath the transaction. On 6lock, for example, identity and banking instructions are verified before any money moves. In the background, 6lock escalates signals that don’t fit that pattern to the GP or fund administrator: a new device, an unusual location, or a banking change in the 72 hours before a capital event. When an LP updates their own banking details directly on the platform, verified before it takes effect, that's actually less friction than the callback it replaces, not more. Speed and verification stop competing once verification isn't a manual task sitting in someone's inbox.

7. What's the hardest technical problem you've had to solve so far, detecting a cloned voice, verifying a video call is real, catching a compromised email thread, or something else entirely?  

Honestly, the hardest problem sits underneath all three: running continuous verification at scale without creating false alarms that train people to ignore the system. The bigger question turned out to be whether we could verify intent without relying on whether the media itself is real, because media authenticity is a battle that keeps moving.

8. Does AI-driven fraud detection ever get ahead of AI-driven fraud generation, or is this a permanent game of catch-up? 

If the goal is detecting fakes, it's a permanent arms race. Generation will keep improving, and detection will always be reacting to it. That's why we built 6lock around verifying identity and intent in a way that never depends on judging whether a call or an email is authentic. Verify the person and the instruction through an independent system of record, and it stops mattering how convincing the deepfake on the phone is, because the phone call was never the control.

9. If you're talking to someone at a fund who thinks "this won't happen to us," what's the one thing you'd want them to know?  

"Our process has worked so far" is the most expensive sentence in private markets someone could say about money movement. A clean run on manual callbacks, emailed instructions, template uploads, and spreadsheets only tells you nobody has come for you yet. Voice cloning, compromised inboxes, and lookalike wires don't care that your last hundred capital events went fine. Companies have lost millions to emails sent from a lookalike domain, forcing them to suspend operations. The day it happens to you, you're explaining to LPs why you kept running fiduciary fund flows on an exposed system. That conversation is getting harder as LP operational due diligence deepens and verifiable controls over how capital moves become part of what fiduciary care means. You will answer for those controls in a diligence questionnaire or after a loss, and the questionnaire is cheaper.

1. Core banking platforms have been in production at many credit unions for decades. The data structures they created weren't designed for modern querying, and replacing them isn't a realistic option for most institutions. What does that mean for any analytics tool trying to sit on top of that infrastructure?

Given that core platforms are sticky and any migration takes years, a modern analytics tool needs to work with where the data already is: the core databases, the on-prem data warehouses, or even the cloud data warehouses. Furthermore, core data is often connected to other data sources that track member journeys, e.g., CRMs, CDPs, web analytics, and so on. This requires linking all these sources together, interpreting the complex schemas and data structures of each, performing in-situ data transformations to make them queryable, understanding the domain-specific semantics needed to serve business users, and making the relevant pieces of data accessible to stakeholders securely and in natural language — all of which is fortunately possible today with AI.

2.  The argument for specialization is the industry context which has to be built in, not bolted on. How much of Tursio's value sits in that embedded context versus the underlying technical architecture and how long did it take to build that knowledge layer?

Tursio's primary value sits in the embedded context, not the underlying technical architecture; the architecture is what makes that context fast to build and keep current, but the context itself is the differentiator. In fact, industry context is critical for any AI to work. Unfortunately, there is no one-size-fits-all: each organization has its own nuances, even within the same industry, and all of them must be incorporated for the AI to behave just like that organization. Tursio’s architecture assists that by ingesting database schemas, BI reports, and query histories to bootstrap a context graph that mimics the current organizational understanding — a well-defined process that completes within a few weeks. In addition, many of the nuances remain undocumented and constitute the bespoke knowledge in people’s heads. Tursio records feedback from all interactions, from users and agents, and attaches it as annotations to relevant portions of the graph. This part of the process is ongoing, and it treats context as a living data structure that continually evolves over time.

3. Trust is the word that comes up in every regulated industry conversation. What does it mean for an institution to trust an AI system that is making queries against their most sensitive data and how does the deployment model address that rather than just the contractual terms?

Trusting an AI system is a leap of faith, and it is our job as a vendor to facilitate that. Tursio does this by deploying on-premises and connecting to data sources within the credit union environment, so that all responses come from their data and are not hallucinated from external knowledge. On-premises deployment further keeps the access boundaries within the credit union, protecting member data from any accidental leakage. Also, Tursio uses LLMs from credit unions’ own Azure subscription, wherein all prompts are fully private and guaranteed not to be used for any model training anywhere. By securing data sources, data access, and LLM prompts, Tursio puts the credit union fully in control of their AI deployment and helps build trust from the get-go.

4.  The "your subscription, your tokens" model you've described changes the vendor incentive structure. When the vendor's revenue doesn't scale with token consumption, why does that matter specifically in an AI/BI context where query efficiency has a direct cost implication? 

Many AI applications are LLM wrappers that meter usage with tokens; essentially, they resell tokens and rely on growing token usage over time. This is a problem for AI/BI since the number of queries can grow very quickly and vendors have little incentive to optimize them. Tursio flips this model by allowing credit unions to plug in their own Azure subscription with their choice of Azure-compliant models, while Tursio's revenue grows with the size of the context (data sources and organizational nuances) that it helped the credit union build. This means Tursio is focused on optimizing token usage, rather than bloating it, since that will lead to more adoption and hence more context to be built. At the same time, the credit unions leverage the enterprise-grade governance and control provided by Azure across their entire AI estate.

5. The goal of natural language database search is that a non-technical executive can ask a question and get an answer without knowing how the data is structured. How close is that to working reliably in production and where does it still require human expertise in the loop? 

The reliability of natural language database search depends on the quality of the context. For Tursio, everything captured by currently documented semantics (schemas, query histories, Power BI reports) can already be answered reliably in production. However, human expertise is still needed to help capture the bespoke knowledge that is not yet documented or that keeps evolving over time. Tursio provides the tools and processes to continue capturing and vetting that additional context with expert assistance. In terms of analysis types, Tursio’s data search can be layered with agents like Copilot, Claude, or others on top, via its Model Context Protocol (MCP) server, for rich analysis ranging from descriptive all the way to prescriptive.

6. The data governance requirements that protect sensitive information also tend to restrict who can access it and how. How do you design an analytics system that is open to non-technical users while maintaining the access controls and audit trails that a regulator would require?  

Managing data access controls is one of the first questions when opening sensitive data to querying by non-technical users. Given the shared context graph that Tursio operates, it is uniquely positioned to enforce access controls centrally, in one place. Specifically, Tursio integrates with existing enterprise authentication (e.g., Microsoft Entra ID) to enforce access control at a fine-grained level. Admins can create one or more logical datasets that define the querying scope (tables, columns, etc.) from each database and decide which users (Entra IDs) or user groups (Entra groups) have access to which datasets. They can further apply table-, row-, or column-level access controls (TLS, RLS, CLS) on a given dataset, again via their Entra ID. Likewise, Tursio’s MCP server gatekeeps all access via users’ Entra IDs, so they only get to search what they have access to. In addition to Entra ID, the Tursio search portal lets admins organize users into roles such as admins, owners, users, and viewers, and personalize the search experience and visibility for each role — e.g., viewers cannot download any data artifacts. For enterprises managing data governance via Microsoft Purview, Tursio can connect and sync definitions and access controls across all data products. All data controls, as well as data access, are logged along with the identifying Entra IDs and roles for audit purposes. These logs are further compressed in-situ as they grow, or users can connect them to other log-management systems, such as Splunk or Grafana, for observability.

7. You've written about natural language becoming the interface layer for data not as a feature on top of existing BI tools, but as a replacement for the query-and-dashboard model. How far away is that as the production reality for organizations, and what has to change at the infrastructure level before it gets there?

The query-and-dashboard model has one fundamental problem: the questions come from business users, but they must wait through long dashboarding cycles before getting the answers. Since these questions are in natural language anyway, LLMs can now connect business users directly to enterprise data, cutting down layers of inefficiency and unleashing new levels of productivity. Still, the gap remains in the context layer that is not yet built, which is actually an infrastructure problem. Building and managing context that can be harnessed to serve business users is the change organizations must go through. Tursio is one such effort in this direction, but the trend of opening highly valuable business data to business users and agents is likely to amplify in the near future. That also means rewiring the data and BI teams into context teams that build reliable context with proper security and access controls built in to make it ready for agents. Most organizations are already thinking in that direction, but it is still early days, with efforts being contained and siloed to avoid any big disruption. I expect organizations to build more confidence and move far more quickly as the context infra matures over the next 6-12 months.

8. As CEO, you're making product bets today on where data access is in three to five years. What are you building toward that the market isn't asking for yet and what does the current state of the technology tell you about where the problems still are?   

Today, data is fragmented across systems and processes; it is dirty, with incomplete or even incorrect values; and it is ambiguous, with lots of loaded semantics needed to interpret it — in short, it is very hard to use. Making AI work on such a data estate is highly challenging yet sorely needed — data migration and integration are tediously long, data cleaning is painfully hard, and data semantics are excruciatingly slow to capture. At Tursio, we are building a disambiguation layer that abstracts clean, unambiguous data — usable by AI — from the organization-wide messy data underneath. Automatically inferring this disambiguation layer, refining it over time, enabling sophisticated data access over it, and keeping a human in the loop throughout are some of the challenges we are working on. As LLMs become increasingly more capable, we believe all of these challenges are going to be addressable.

Q1: Why is metadata-based social listening no longer sufficient in a short-form, AI-driven ecosystem?

Traditional social listening was created for a time when text, hashtags, and manual labels were the norm. Short-form video has changed that landscape. Today, products are often shown as part of daily life instead of being tagged or captioned. The things that spark viewer responses, like tone, cultural context, and genuineness, are found within the video, not in the surrounding labels. Metadata cannot capture these key details.

AI is moving toward actually watching videos. Rather than just using captions, hashtags, or transcripts, these systems review the entire video with models that recognize what’s happening on screen. This change enables services like Archive to answer questions that older tools cannot, such as which creators are showing your product without tagging you, which content is working, and why. Most of the meaningful insights come from this untagged content, which metadata alone cannot reveal.

Q2: What does “agent-first” actually mean for marketing operations teams?

Agent-first means moving beyond basic automation to systems where agents manage entire tasks from start to finish. For marketing teams, this evolution means less manual work finding and checking creators. These systems can also constantly scan user-generated content, highlight posts that mention a brand, and identify promising creators. They handle more content than any human team could, so marketers can quickly focus on the best posts instead of sorting through thousands by hand. Marketers can then focus less on manual work and more on leading strategy and keeping up with cultural trends.

Q3: Has the content bottleneck in creator marketing shifted from production to operations?

Yes, and to put it another way, creator marketing is really a labor-intensive business that looks like a marketing channel. The main bottleneck is often the amount of human time required.

Typically, one coordinator manages, let’s say, about 50 creators, and a single product launch can lead to hundreds of TikToks and Reels in just two days. Brands want to take advantage of this drive but often cannot, because there are not enough people to collect the content, get usage rights, connect posts to audiences, and act before the opportunity passes. The content is there, but the system cannot keep up.

That is why the next phase of growth will come from teams that break through the limits of manual work. Using operational intelligence, finding content, managing rights, and tracking performance in real time, can turn scattered user-generated content into an impressive growth engine.

Q4: How are brands using AI to generate near real-time cultural intelligence around major events like the Super Bowl or Fashion Week?

In the past, brands waited until an event was over to analyze it, but by then, the moment had already passed. AI changes this by speeding up the process. AI-powered platforms can review thousands of videos and posts in minutes, showing trends while the event is still happening. This lets brands see in real time which formats are working, which creators are leading the conversation, and which stories audiences are following, often the behind-the-scenes and creator-led content that does better than official coverage.

During big cultural events like New York’s Fashion Week, AI-powered platforms can rapidly process thousands of videos and posts, disclosing trends that would have been missed before. For example, this year’s event saw a “Global Idol Takeover” that led to over 350 million views. There was also a trend called “survival styling” because of the cold weather, with creators sharing how they dressed, protected their hair, or adapted their style. Street-level content, like behind-the-scenes videos and vlogs about walking in the cold, outperformed runway clips, showing that audiences preferred real, creator-led stories over official coverage. Finding great trends early is another way marketers can leverage AI.

Q5: How should marketing leaders think about governance, brand safety, and control in an agent-driven environment?

Governance should be seen as part of a team’s core strategy, especially when it comes to the way the market is headed. As AI speeds up content creation and makes operations more complex, safety and brand alignment must be built into the process from the beginning.

The main priority should be large-scale transparency. Leaders need systems that help them track and measure creator content using clear brand and data rules, with people involved in important decisions. Content is growing faster than manual quality checks can handle, so keeping an eye on governance, brand safety and more is more important than ever.