Developing trustworthy AI algorithms is Keen’s business. We’re helping marketers manage billions of dollars in marketing investments. We endeavor for our users to leverage our AI to its fullest potential. In our expansive vision for the future, we have many designs for evermore trustworthy algorithms.
Trusting AI is not just a matter of time and it’s a joint effort with users. Users should hold AIs to the same principles they hold people. In the eBook, “A User-Oriented Framework for Evaluating Trust in AI,” we explained and developed a framework based on that premise. The framework helps users demystify AI, take control of their relationship to AI, and ensure that their interests are aligned.
Let’s look at the Keen Platform through the lens of that framework.
Understanding Keen’s commitment to trustworthy AI
Context: The Keen Platform is designed to provide recommendations for investment decisions according to your decision frame. Therefore, its fundamentally important for building trust that as a user you define the frame. Think of the frame in terms of five questions:
- What is my objective? Are you trying to maximize the value of marketing defined by the value of your total sales?
- How would you describe the decisions you’re trying to make: All tactics? Single product line or portfolio of products?
- What are decisions that are being made by others or in other contexts that are necessary to account for as drivers of your objective? Pricing? Product? Distribution?
- What do you know or can reasonably forecast to happen in your environment: Seasonality? Persistent shift in consumer trends?
- What are things that could happen, but that you cannot predict in the future? This helps draw a boundary on your frame. These factors should be left out of the frame and left to random chance.
Accountability: The Keen Platform is designed to forecast your business based on whatever action you take. As a user you should hold the algorithms accountable to that. However, we also know that neither the past nor future can be explained perfectly, neither by people nor algorithms. Three important features of the Keen Platform allow you to hold it accountable:
- Holdout tests: Dividing the data between training and evaluation sets allows you to build your model and preemptively develop and understanding of how well your forecast might perform. This prevents overfitting; avoids false promises of accuracy. It gives you a helpful pretense for just how accountable it can be.
- Decompositions and Forecast Actualization: All forecasts are prone to error, and so all forecasts fail to an extent. However, to build trust you want to know why. Using the decomposition and forecast actualization when the future unfolds, you can understand what changed, when sales were different, and why.
- Continuous learning: This also falls under intelligence. When errors occur and you know why, to the extent that those reasons are predictable you will want the algorithm to learn to use those signals in the future. Models in Keen Platform can be updated and learn over time.
Availability: The Keen Platform is a SaaS system. It’s designed and rigorously tested to ensure that all the features are available when you need them.
Speed: The Keen Platform is designed for speed, and not just computational speed. Its also designed to increase your speed to decisions. We know that businesses are run against timelines and that things happen. When they do it’s important to be able to make those decision quickly. One such design decision is in the data. Keen Platform simplifies data requirements and makes the loading process easier. It also makes quality assurance easier. And the data provisioning is segmented from the decisioning. So, when you’re working on providing new data, you can still make decisions based on the current state of information. This means that when you’re using the Keen Platform you can make decisions quickly.
Intelligence: The Keen Platform is intelligent because it learns from multiple sources. As it does it continuously adapts and provides new recommendations and forecasts based on that information. The algorithm uses a Bayesian approach to incorporate information from three sources:
| Source | Description |
| Keen’s Marketing Elasticity Engine | This is patent-pending technology that learns from multiple public and private sources, combined with the particulars of your business to provide a prior assessment of the general effectiveness of your marketing tactics. |
| Your own previous experience, including studies of marketing effectiveness | This encompasses a wide range of information from marketing mix studies, in-market tests, attribution studies, platform metrics, and your own experience. |
| Time-series data | Data gets the last word in the Bayesian approach. Observations provide evidence based on how the business changes based on changes in your decisions and the environment. The models learn from the available signal in the data and weigh that against the other sources of information. |
Transparency: The Keen Platform provides a holistic framework for making marketing investment decisions. When you’re receiving a recommendation, ultimately you need to know why. Transparency can be thought of in terms of inputs and outputs. The Keen Platform provides transparency to your inputs including sales, activity and financial data, and prior information. In the Keen Platform users can both understand the implications of prior information in terms of ROI ranges and provide prior information in those terms. ROIs transmit information in terms that are easily understood and transparent. The Keen Platform also provides transparency to recommendations and the reasons behind those recommendations. Results are coupled with a range of metrics, most importantly marginal ROIs and response curves. This helps users understand the whys behind the recommendations. Finally, in terms of outputs, Keen Platform provides a forecast. As discussed above, this allows you to hold the platform accountable.
Confidentiality: The Keen Platform maintains users’ and clients’ confidentiality. All data is stored in a deidentified manner, no data is transmitted in an identified manner, and none stored in any way outside Keen’s control. The only interaction on an identified basis is by authenticated users via their browser. There are no opportunities for identifiable information to be provided to another user.
Shared Values: The Keen Platform aligns with users’ values. First and foremost, Keen’s only business is to help our clients and users make better marketing investment decisions. We’re not selling advertising or any other service to anyone else. The algorithms have no other incentives than to provide for the best decision. Importantly, users can explicitly state their objectives and the recommended investment and forecast will be consistent with those objectives. Users can maximize long-term value, optimize a budget, or hit a forecast. Within the stated context, users’ values are aligned with our AI.
Keen Platform: Building trust through innovation and transparency
We value the trust that we’ve created with our clients and users. They entrust us with billions of dollars of marketing investment. It’s our desire to maintain and strengthen that trust over time to help make decisions about billions more in marketing investments. It’s our desire that with this framework, you can identify and relate of the sources of your trust in us. Furthermore, as with any relationship you’re able to hold us accountable to build and foster trust. We welcome your feedback and endeavor to make Keen the best it can be.
Take a tour of our platform today to learn more about our solution.
