How media teams can thoughtfully balance liquidity and control

A campaign is more liquid when it has fewer constraints. That's because limiting restrictions means that the delivery system is able to select the best opportunities to serve ads from a broader pool, potentially turning up valuable opportunities that would otherwise have been excluded.

When you have clear goals and signals – pieces of consumer behavioural data – it's easy to tell the system what you want and to set it free to find the optimal approach. But if your goals and signals are more opaque, you need to put in some guardrails to guide the system and make sure it doesn't go too far off track.

While adding these constraints delivers more control, the trade-off is less liquidity. That can result in higher costs and lower efficiency, as reducing the pool of opportunities limits the system's ability to find the most valuable ones.

Adding control does not trigger a complete loss of liquidity. Rather, there is a spectrum of liquidity, and liquidity products can empower advertisers to achieve the level of control needed while still maintaining efficiency.

Specifically, different products can work together to allow advertisers to gain control across the four main dimensions of liquidity: placement, audience, budget and creative.

By thoughtfully adding constraints in these four areas, it's possible to give machine learning systems more information about what you value and to help the algorithms understand how to better weigh different opportunities.

When thinking about implementing constraints, it's important to note that the dimensions of liquidity are interdependent. Putting a constraint on one dimension will limit the opportunities available to the system overall, even if the other dimensions are fully liquid.

How can you determine whether you should exert more control on ad delivery systems? Some common reasons for needing to deploy constraints include:

• You may value opportunities within the same campaign differently – for example, Instagram users versus Facebook users – and not want the system to treat them equally.
• You may be using the same channel for different strategies – for example, prospecting and retargeting – and want to communicate different cost goals to the system.
• You may have data from sources that the ad system can't access – for example, from offline channels – and want to express this information to help evaluate opportunities.
• You may have business restrictions – for example, brand safety requirements or regional budget allocations – and want the system to account for these complexities.
• You may want to hone your campaign goals over time – for example, you may be willing to increase your target spend to get more volume – and want the system to be flexible.
• You may have multifaceted goals – for example, wanting to drive Page engagement while also needing to keep costs down – and need to express these to the system.

While it can be tempting to be heavy-handed, the key is to gain the control you need in the lightest touch way possible, as implementing fewer constraints gives the system more freedom to find valuable opportunities.

Moreover, the different dimensions of liquidity – placement, audience, budget and creative – are interdependent, so carefully placing constraints on one to two areas can often deliver the desired amount of control.

Ultimately, when seeking to navigate liquidity in a complex world, the goal should be to add only the constraints absolutely necessary in order to provide machine learning systems maximum leeway to deliver liquidity.