Why Businesses Need A Month Or More To Deploy Ml Models And What You Can Do

Why Businesses Need A Month Or More To Deploy Ml Models And What You Can Do

Why Businesses Need A Month Or More To Deploy Ml Models And What You Can Do
Why Businesses Need A Month Or More To Deploy Ml Models And What You Can Do

We are excited to return Transform 2022 in person on July 19 and virtually July 20-28. Join AI and data leaders for insightful conversations and exciting networking opportunities. Sign up today!!

Machine learning (ML) is an invaluable asset to today’s business worldwide. However, when it comes to ML models, B2C and B2B companies face the problem of delaying time to market. According to Algorithmthe vast majority of companies take at least a month or more to first develop and then deploy their ML model.

The reason for this is a complex and often very expensive two-stage process. Developing an ML model can be a lengthy and potentially costly process in itself. But what many companies often don’t realize at an early stage is that the initial stage should be followed by another, perhaps more difficult, stage – deployment. This second stage involves moving the finished model into production, testing and fine-tuning it, and then scaling it accordingly.

Only about 10% of all businesses are estimated to have sufficient experience, financial resources and technical expertise to deploy a new ML model in production within a week of its completion. Many struggle for a year, with at least 30% of all companies occupying at least three months after deployment. Exactly how long this will take depends largely on which of the three popular model types the company chooses.

Constant, non-standard and adaptive models

Of the ML models currently available on the market, there are the following: generic models, custom models, and custom adaptive models.

Common and custom models are mostly polar opposites. The difference is that common models are there low cost and accuracy, while custom models are characterized by high cost and accuracy. This is because common models are designed to suit virtually every business in the industry. They are usually based on ResNet, BERT / GPT and similar standard technologies. As a result, these models are affordable and reliable, but they are also far from ideal.

In contrast, individual models are always adapted to the task and are therefore much more accurate. However, they also come with a much higher price due to their high development and maintenance costs. Those who start with an overall solution and then try to improve their ML model often go beyond the basic architecture of the model. What they end up with is an individual model. A custom model that can be immediately adapted to broad business needs and abandon most of the long-term fine-tuning after deployment is an individual adaptive model.

Thus, an adaptive model is a type of custom model with some advantages that common models offer. Like all other custom models, adaptive models are tailored to the specific needs of the business. For this reason, they are very accurate. At the same time, they do not require the company to understand MLops after the initial stage of development. As a result, they in a sense work as common models in the deployment and post-deployment phases, with relatively low maintenance costs and increased time to market.

Choice of ML model

Which model is required for your business – that is, whether it is worth the surcharge – depends on your specific situation. Your business may need something simple, such as sending online orders to different warehouses depending on their location. In this case, a regular ML model can just do the trick, especially if you’re a small business.

On the other hand, if it’s something specific, like content moderation an online community of physicians when discussing medical equipment, a custom model will work better. The fact that the general ML model can be considered as inappropriate language – for example, mentions of the genitals – is not only relevant but also necessary in the context of medical discussion. The training model in this case must be adapted to the individual needs of the company. And this individual model can be both adaptive and not.

Let’s consider the pros and cons of each model:

Comparison of ML model types. Image by the author

Individual adaptive models

Custom ML models are expensive due to often unforeseen costs before and after deployment. Because of these typically high start-up costs, some companies tend to shy away from the individual option, instead choosing a less accurate but also less expensive overall track. How expensive the learning model is depends on a number of factors, including the chosen one data labeling methodology, which is manifested in the flexibility of the model or its absence.

The following case illustrates a custom adaptive model based on crowdsourcing in action, i.e. adaptive model based on human marking in the loop:

One well-known company that offers a technical editing environment wanted to increase the accuracy of its software and reduce the cost of learning the model. The team of engineers had to come up with a more effective solution to correct the sentences in English. Any solution must be in accordance with the fully manual conveyor labeling that has already been in place.

The final solution involved the use of an existing user model for linguistic processing, which was adapted to the needs of the client. Third-party AutoML was used to classify the text in the target sentences. Subsequently, the accuracy of checking phrases increased by 6% – from 76% to 82%. This, in turn, reduced the cost of model training by 3%. In addition, the customer did not need to make additional investments – financial or otherwise – in the infrastructure of the model, as is usually the case with most custom models.

Highlights to keep in mind

Choosing the right ML model for your business can be a daunting task. Here is a summary of what you need to take into account to make an informed decision:

  • Consider how specific your needs are: the more specific your needs, the further away from the overall model you should move away, as a rule.
  • Always consider the possibility of scaling – if you know you need it, think about paying extra for something created just for you.
  • If you don’t need high accuracy, but need a quick deployment, consider choosing a shared route.
  • If accuracy is important to you, think about how much time to market you can save.
  • If you have little time and high accuracy is required, consider using a custom adaptive route; otherwise any custom solution can potentially meet your needs.
  • In terms of total cost, a shared route is the cheapest of all – followed by an individual adaptive route that bypasses most MLops costs – and finally, all other custom solutions, the cost of which can increase significantly after deployment (the exact figures vary greatly from case to case).
  • Think about whether you have your own data scientists and MLEs – if so, you can take advantage of traditional customized options developed internally; if not – consider the other two (general or custom adaptive).
  • When choosing between custom and individual adaptive options, consider how accurate and specific to your customer’s needs the ML model should be. The higher the accuracy and adaptability, the higher the cost and the longer the waiting period for model preparation and maintenance.

Fedor Zhdanov is the head of the ML products department Toloka AI.


Welcome to the VentureBeat community!

DataDecisionMakers is a place where experts, including technical people working with data, can share their thoughts and innovations related to data.

If you want to read about cutting-edge ideas and current information, best practices and the future of data and data technology, join us at DataDecisionMakers.

You might even think contribution of the article your own!

Learn more from DataDecisionMakers