As enterprises embrace AI and work towards integrating it increasingly into their business processes, one of the key decisions they are all having to make is whether to buy or build the AI components that will drive their enterprises into the future. More than 61 percent of businesses said they have already implemented AI, demonstrating that adoption is on the rise [Narrative Science, 2018].

Cost, time-to-market, ROI, criticality to business success, and quality of solution are common factors that must be considered between the buying versus building decisions, and they apply to AI initiatives as well.

The challenge is how to make this decision definitely, and the best way is through considering the capabilities that are uniquely needed to succeed in AI.

Here is a guide through these capabilities to reconcile before making the buy versus build decision:

  1. Capability to gain access to quality data scientists.

    The scarcity of data scientists is well-known, but most enterprises still feel the need to first hire a bunch of data scientists before embarking on AI projects. Although enterprises may eventually achieve the goal of building a strong core in-house team, most AI initiatives can rely on vendors in the meantime.

  2. Capability to translate business problems into AI building blocks.

    Even with a strong in-house data science team, one of the core reasons for AI initiative failures is the difficulty companies face in breaking down business problems into the right AI building blocks. These building blocks must subsequently integrate the solutions seamlessly into the workflow.

    Usually, this is due to the emergence of data science teams or the sheer lack of ability of the data science team to communicate with business functions. Hiring a few data scientists does not necessarily give an enterprise the ability to build and integrate quality AI solutions at scale. The sooner an enterprise understands this, the lesser start-stops they will face in the beginning.

  3. Capability to measure and improve data quality.

    An audit of the available data, quality and structure of the data, and readiness of the data for various AI models is a prerequisite for accurate AI models. We have observed several enterprises embarking on large scale AI projects, struggling to get the desired accuracy and precision levels due to lack of quality training data.

    Also, sometimes, the model may require data from other external sources that the enterprise may not have readily available. In other cases, before any AI initiatives are embarked upon, considerable effort needs to be invested in cleaning the data as well us structuring unstructured datasets in the form of images, audio, video and text, into structured forms that an AI model can be trained on.

    Without quality data structured and feature-engineered for AI models, AI initiatives are sure to be doomed.

  4. Capability to experiment rapidly.

    Let’s say you have access to data scientists, either in-house or through vendors, the solution architects that can understand your business challenges and convert those into AI initiatives, and that you also have clean, structured product data.

    You now need the ability to perform several rapid experiments. AI is part art and part science. Any AI modeling exercise relies on several assumptions. Only through rapid experimentation and trying out different modeling techniques can one decide on which models are the most accurate and seem to be consistent in their predictions.

    These experiments can take a long time unless done in parallel. However, conducting these experiments is critical and will cost considerably less than scaling an AI model that hasn’t been rigorously compared with other approaches, and then fails after full-scale company-wide deployment.

  5. Capability to scale and maintain the solution.

    Once you’ve uncovered and tuned the right AI models for your requirements, you now need to build the infrastructure needed to integrate the AI models with your existing IT systems seamlessly and quickly.

As more and more enterprises join the AI bandwagon, they will all need to consider the five capabilities listed above and decide on the best way to acquire those capabilities, either through building or buying the technology.

The good news is that there are a broad landscape of vendors today that can work with clients to deliver all or some of these capabilities, becoming their AI team or an extension of their AI team. Enterprises should view the decision, not as “build or buy,” but as “build and buy,” where some of the required capabilities are in-house, and others are acquired through external agencies.

Over time, internal capabilities can be strengthened, and some key AI initiatives or core parts of those initiatives can be moved in-house reducing dependency on external vendors.

Divyabh Mishra

Divyabh Mishra

Divyabh Mishra is the founder and CEO of CrowdANALYTIX, a platform for building high precision custom AI solutions. For the past seven years, Divyabh has focused on growing the company's global footprint, expanding its portfolio of Fortune 100 customers. Prior to founding CrowdANALYTIX, Divyabh was the Director of Global Marketing & Branding at Aditi Technologies, and the Director of Marketing at Lionbridge. He was also the Group Head of Marketing with Symphony Services, and previously the Director of Marketing & Strategy at Genser Aerospace.