Independent advertising demand-side platform (DSP) Viant recently launched a new generative AI (genAI) tool that aims to address programmatic complexity by making digital media planning and execution easier, more efficient, and eventually near-autonomous – but for now with human oversight.
Dubbed ViantAI, the new tools promises to provide answers on what, where and how much to allocate of ad budgets to meet advertiser-specific KPIs and goals, with eventual plans for automatic decisioning capabilities. As of now, it starts with just four simple inputs to create a fully data-driven programmatic media plan through a conversational, genAI prompt-driven interface, executed within minutes.
The launch could also serve as significant step for Viant to grow its base of customers, as Wall Street analyst Laura Martin of Needham told StreamTV Insider she sees it expanding the total addressable market (TAM) for the DSP to smaller advertisers and brands looking for self-serve capabilities.
StreamTV Insider got a demo of the tool (which you can also see in action in this separate video released by Viant co-founders) during an interview with Viant CMO Jon Schulz.
Distilling 98 trillion combinations
As a programmatic DSP, Viant offers automated ad buying across CTV, mobile, desktop web, streaming audio, and digital out-of-home.
The ad buying landscape is fraught with complexity, where audiences and media are splintered across channels, platforms, and services, and advertisers need to deal with disparate data sources and ad tech partners as they weigh decisions on where and how best to ad allocate dollars across various channels.
To that point, when considering all the different ways to slice and dice budgets among various channels, publishers, and different formats within each, among other choices to make on the Viant DSP, it amounts to approximately 98 trillion possible combinations – more than a human can possibly compute, according to Schulz.
The model is trained on all of those combinations and sorting through to quickly distill down 98 trillion combinations into a digestible, actionable media plan based on advertiser goals and what they’re trying to achieve is one of the aims of ViantAI, he noted.
ViantAI is coming to market amid a fast-moving and often changing digital advertising space (such as this year when Amazon Prime Video introduced advertising as the default on its platform and earlier ad-supported entries of Netflix and Disney+ brought new inventory to market).
The platform serves as a jumping off point for programmatic traders to plan and execute media buys, with an emphasis on outcome-driven outputs. It also is a step towards Viant’s larger vision for so-called autonomous advertising or advertising-as-a-service, where AI-powered automation provides capabilities across campaign planning, bidding, optimization, and execution.
How it works
Here’s a high-level rundown of how ViantAI works:
- ViantAI can create a full media plan using four inputs: brand or advertiser, timeframe of campaign, budget, and objective or KPI of advertiser (such as brand awareness, driving conversions or increasing leads).
- The video demo with co-founders used specialty grocer Sprouts as an example. If Sprouts has $5 million to spend in Q4 and is primarily looking to drive in-store visits. ViantAI will generate target audiences (in this case, a demo of age 25-54, predominately female, with behavioral and interest-based targeting like health conscious, and geographic details about where Sprout stores are located, as well as competitors)
- Then it does ‘smart channel selection’ – suggesting which channels to allocate, such as CTV, streaming audio or other and how much of the budget (i.e. 30% to CTV, 15% on streaming audio etc.), along with a justification as to why those channels should hit client goals. This is an area Schulz said is a particular challenge for clients, citing the notion of “choice overload.”
- ViantAI drills down further into timing in terms of days and times – such as evening for CTV between 6-10 pm, or audio streaming only during commuting hours during the weekday – and determines frequency cap.
- Then it puts together a plan with line items publisher by publisher (for example: allocate $X dollars to Hulu on CTV with specified audience segment, frequency cap of 3 times per week, time of day, maximum CPM bid and expected impressions).
Schulz categorized the summarized media plan as a first pass, where media planners can then modify and utilize their own expertise to tailor further. “It’s meant to be very flexible,” he added.
If planners or traders don’t like the recommendations, they can tell the tool to eliminate certain channels or publishers and allocate the budget elsewhere.
Humans can come into the mix to refine with prompts such as: "We don’t have audio assets, where would you reallocate that budget?." Users can expect to ask detailed questions and receive specific answers that provide insight into the rationale. For example, if a planner asks ViantAI, “If I remove streaming audio will it hurt campaign performance?” the tool will generate pros and cons.
After ViantAI generates a plan, the platform offers up interactive and adaptive elements. A user could ask, “How should I measure this campaign?” and the platform will generate ways to measure success and potential partners, alongside reasons why that partner would be good and the ability to show additional or alternate options.
The speed of analysis behind each variable in creating a media plan is part of the appeal. Viant emphasized the tool doesn’t just automate mundane processes but gets oversight and feedback from humans with their own expertise and knowledge. While it envisions a fully autonomous future, the company is promoting fewer errors and greater efficiencies thanks to genAI. This lets marketers focus more time and attention on strategy while ViantAI handles execution, Shulz said.
LLM trained on campaigns
With the tool debuting in recent weeks, Schulz said there's been a strong response from users.
While some users tried similar prompts in the publicly available ChatGPT, he said ChatGPT's results helped illustrate the power of ViantAI. ChatGPT uses internet-sourced information. Comparatively, ViantAI is trained on historical data from thousands of campaigns that have run through the DSP.
“Because [ChatGPT]’s just scanning the broad internet, it’s not really dialed-in, and it doesn’t have the rules and guidelines and the expertise for advertising,” Schulz said. Whereas ViantAI can deliver recommendations for allocations of budget by channel, day parting for different channels, frequency capping and more.
Going back to the quickly changing landscape, he suggested a media planner that might simply look to reuse an earlier media plan wouldn’t be as up to speed because of new inventory and services coming to market.
“So that’s all net new opportunity,” Schulz said, adding that ViantAI is “constantly learning and evolving as things are changing” and enabling planners to stay current.
“What this is doing is doing a lot of the heavy lifting…and it’s goal-driven, rather than biased or preferenced, or based on just the history for this one account,” he continued. “It’s saying, if you have this much money and this is your goal, this is the best away to approach it. And it’s trained on thousands and thousands of campaigns.”
Future data integrations
While historical data is a really good input, the CMO acknowledged it’s not always the best judge of the future.
To that end, Viant plans to introduce third-party datasets including data related to reach numbers, prevailing CPMs, and media mix modeling so that external data can continue to train the model to make it better. Later this year the company expects to integrate first-party data and campaign performance metrics, enabling instant analysis and campaign insights. It will also allow custom data, whether that’s a mix of first-party or partner data to create audiences, which can be ingested or uploaded into the platform.
The data element might not come into play for all advertisers, Schulz noted, as some might already have their own datasets. For example, if a political advertiser wants to target constituents in North Carolina and already knows their target audience but also wants to to understand which channels would be best to reach that audience, ViantAI can do that for them. ViantAI is assumes a baseline, and then the more sophisticated the planner, the more prompts they can tailor to drive outcomes.
Looking ahead, Viant intends to integrate media plans into the DSP directly via APIs. This means that once ViantAI builds a media campaign, it’s automatically created in the DSP in a fully automated process with no need for manual uploads.
Simplifying the learning curve
While not a revenue-generator on its own, the ViantAI offering could help attract and more easily onboard clients and their budgets to the DSP.
The DSP, which went public and started trading on the Nasdaq in early 2021, has seen recent success with CTV growth in particular. In Q2 Viant reported record advertiser spend for both CTV and streaming audio, a quarter that generated $65.8 million in total revenue for the company. CTV spend specifically grew more than 40% YoY on the Viant DSP in Q2, driven by its direct-access program and household ID technology.
Viant has roots in digital and streaming TV, including as founder of the Xumo FAST that was eventually acquired by Comcast and now operates under a joint venture with Charter.
However, Schulz described conversations, where prospective clients had showed interest in shifting over to the Viant DSP but expressed hesitation over figuring out new or bespoke portals and systems, likening it to having to relearn riding a bike. He acknowledged that the need to get trained on Viant’s platform had been a hurdle in the past in terms of new customers interacting with the DSP. Compared to the old way of doing things, ViantAI, which works straight out of the box, is “so much easier,” he said.
With the prompt-based interface it “simplifies the learning curve,” as not much training is required and the DSP provides instructions and user guides to help with prompts.
“Adoption will be super simple, the transitions will be super simple. It really does take that barrier down, and that’s something we’ve been challenged with,” he commented.
Addressable market expansion
Viant’s not the only one that sees the tool’s simplicity as creating opportunity to onboard new DSP customers.
Viant counts three primary segments of customers including agency holding companies, independent and mid-market agencies (where Schulz said the DSP does particularly well) and direct-programmatic clients. He emphasized the genAI model is scalable up or down and meant for advertisers with budgets of any size, rather than trying to cater to only one type of customer.
That said, Needham senior analyst Laura Martin believes the tool expands Viant's addressable market.
Martin said ViantAI seems to represent “a new total addressable market” (TAM) for the company, opening the door to a potential larger pool of smaller advertisers and the opportunity to pull share from walled garden digital giants like Meta and others.
Martin, who covers Viant as a buyside analyst, doesn’t think the company is going after big ad agency clients with ViantAI – where a large role agencies play is media planning – but instead sees the tool as targeted toward smaller advertisers and “mom and pops who’ve never had an agency do a media plan.”
Unlike big-budget traditional TV advertisers, those types of smaller digital-first advertisers “don’t know how to figure out where to put money” as they’ve always gone with an easy-to-use platform like Facebook.
For example, if an advertiser always went to Facebook or YouTube directly, with ViantAI they could also see potential combinations to allocate budgets to different channels or publishers.
“My gut feel is this is trying to pull money out of walled gardens,” Martin commented, noting she feels it’s targeting advertisers that are currently using a single solution to entice them to try other venues with an easy prompt-based media plan, and then use the Viant DSP to execute those buys.
“To me this is TAM-expanding,” she said. The opportunity is aimed at smaller advertisers that haven't used an agency in the past or only gone through previously mentioned walled gardens with single-point solutions. Martin acknowledged there’s potential an agency could use ViantAI as a starting point for media planning, but expressed doubt they would do so, suggesting it wouldn’t make as much sense for a large or mid-market ad agency whose functions include building media plans.
Still, the expanded TAM opportunity for Viant with small business advertisers is, “huge, huge,” Martin said, without putting a figure to it, but noting that Facebook has millions of advertisers. A small business support page by Meta, for example, claims that more than 10 million advertisers – most small and medium businesses – use Meta’s personalized ad tools.
Viant could stand to take share from walled gardens like Facebook and Google search and pull in some of those advertisers that go directly to self-serve platforms. Martin noted SMBs like the simplicity offered by those platforms but now have a way to build media plans and execute across channels outside of those.
“Advertisers can do this on their own, and [Viant’s] simplifying the on ramp” for them into the open internet, Martin said.
Humans trusting humans?
Schulz did acknowledge a little bit of fear among some people when they saw just how powerful the ViantAI tool is. But the CMO suggested hesitations revolve more around concerns of tech replacing the work of humans, where worries may be a bit more self-serving rather than forward-thinking.
For its part, Viant is pitching the tool as enhancing programmatic traders’ ability to plan and execute campaigns – encouraging those piloting to test and challenge it and where human expertise can always override suggestions made by the AI.
The CMO noted he wasn’t aware of any similar competitive products on the market, but expects plenty of fast followers now that the lid has come off and details disclosed.
That said, Martin doesn’t anticipate similar genAI planning tools from major publicly traded ad tech players, namely The Trade Desk, which has a different customer mix than Viant.
According to Martin, The Trade Desk could do something similar overnight it they wanted, but the analyst doesn’t think it would be valuable as that DSP counts all of the largest ad agencies among customers. She doesn’t believe large ad agencies or individual mega brands would use a genAI at this stage in the game for media planning. Part of that is because it’s a big aspect of agency work, but also because the analyst sees genAI and these types of algorithms as too nascent and not yet fully reliable or transparent, whereas agency clients prefer to lean on decades of human experience.
She noted the inability to get to the back of genAI algorithms in terms of why they come to certain conclusions, as well as concerns over the veracity of accuracy. “So you’re taking a risk,” Martin noted, adding humans tend to trust humans.
“I think this is for people that don’t have a choice, that they don’t have a media plan in the past…and they don’t have a human they can go to,” Martin said of ViantAI.