It looks as if ByteDance is finally divesting its US TikTok operations under political pressure. A $14 billion deal led by Oracle and investors raises a much bigger query than ownership: Who will control the algorithmic intelligence that powers the platform’s unique engagement engine?
In some respects, the headline event is a technology story. Yet for marketers, TikTok’s position as possibly the most engaging social platform so far could mean changes to firms’ strategy consistent with the change on top of things. TikTok’s content-recommendation model has set recent standards for predictive engagement, and has created creator economies, real-time audience segmentation, and levels of engagement that other platforms have tried – with various degrees of success – to emulate.
A unique algorithm behind the platform could reshape how billions of consumer interactions are measured and monetised.
At present, it’s not clear if the TikTok ‘gold dust’ code will probably be re-engineered or retrained. And when the algorithm is the product, the consequences could mean brands which have invested in the social platform will feel any changes hardest.
Business impact: when the algorithm is the product
TikTok’s “For You” page represents an indication of machine learning at scale. The system refines recommendations from signals like watch time, comments, and replays to predict what each user wants next. If TikTok’s US algorithm is retrained from scratch, as the White House has indicated, the effect might be immediate. Machine learning models are a product of their training data, so a change in that data or in training parameters will indubitably alter how content is chosen and surfaced.
How’s this going to affect creators’ reach and advertisers’ ROI? It’s unclear at the moment, although hopes are high that the algorithm’s recent owners, Oracle, won’t rock the boat. Speaking to Business Insider, comedian Winta Zesu said: “Whatever they do, I just hope it’s still the same. What we love about TikTok is the algorithm and the way you only find exactly what you would like.”
Technologically speaking, that seems unlikely, and given the strength of political influence over big technology firms at present, the likelihood of the same TikTok operating in the US as before is vanishingly thin. Likely not immediately – but marketing professionals would do well to begin rethinking their short-form video content strategies sooner not later.
The shift in the underlying suggestion dynamics could mean re-calibrating entire campaign strategies. One enterprise marketing lead described the risk succinctly: “When the algorithm changes, our performance model changes overnight.”
Implementation and operational challenges
Recreating TikTok’s intelligence in a siloed US environment poses formidable operational hurdles, some arguably insurmountable. ByteDance’s system shouldn’t be a single codebase but a multi-layered network of AI models continually retrained on vast datasets. Replicating this without the original engineers or data infrastructure may significantly alter performance. The details of the deal between ByteDance and Oracle are murky, so it’s unclear how much migration from one company to the other will happen.
Experts suggest retraining could take years. “Even if the code were copied line for line, without the people and the learning data behind it, the results would diverge quickly,” said Paul Resnick, a University of Michigan researcher specialising in suggestion systems.
The recent owners might want to comply with data localisation rules and maintain suggestion quality, a balance that would rely upon further investment in infrastructure, data sets, and folks. Those investments carry industrial implications: every layer of difference between old and recent adds cost and complexity.
From a governance standpoint, ByteDance’s limited involvement post-sale will satisfy lawmakers but the deal removes a the foremost source of institutional knowledge. Pure-play technology leaders know that AI infrastructure isn’t transferable or replicable: Models reflect training data and ongoing data ingestion, plus Oracle will add its own culture and processes.
Takeaways for marketing leaders
The TikTok AI sale underscores a broader truth about the martech landscape: algorithmic differentiation is now an enterprise asset class. As generative and predictive AI develop into more necessary to customers and digital platform vendors, data independence will define long-term market value.
CMOs should draw three lessons from this unfolding story. First, marketing performance increasingly relies on how, and where, algorithms are trained. Second, technology migration is as much an organisational challenge as a technical one, and third, when the technology in query is AI, it doesn’t as much migrate as morph because it moves.
As TikTok’s US entity begins its algorithmic retraining journey, marketers will probably be watching to see what happens to their For You feed, and people of their audiences.
(Image source: “Tiktok” by TheBetterDay is licensed under CC BY-ND 2.0.)
Find out more about the Digital Marketing World Forum series and register here.
Read the full article here
