When to Reach for Custom Artificial Intelligence Solutions vs. Off-the-Shelf
We’re in the midst of a time of great excitement and enthusiasm for AI and the possibilities it brings. You may find yourself at a company wondering how exactly can AI help us? Off-the-shelf tools may not feel like a good fit. What you do is unique, and you may be considering building your own custom artificial intelligence solution.
Before you set down the path of making a bespoke AI solution, I encourage you to step back and reconsider. Do you need custom software, or will an off-the-shelf solution suffice? Below are the general steps for deciding if you need a custom AI solution:
- Identify the business problem you’re trying to solve
- Recognize how AI can help
- Record Why Off-the-Shelf Solutions Don’t Work for You
- Estimate the Cost of a Custom AI Solution
- Weigh Tradeoffs
- Reach for Custom Artificial Intelligence Solutions
Define Your Problem & Solution
Software is a powerful tool for solving problems. AI-enabled software even more so. It’s so powerful that I find it gives rise to the following problem:
We’re so confident that AI can solve any problem that we don’t thoroughly define the problem to be solved.
While that faith in AI might make me feel good as a practitioner, it’s concerning as a professional. If we can’t agree on the problem at hand, we’ll be unable to agree on what success looks like. This sort of misalignment can doom a project before it’s even started.
Knowing what you want to achieve is critical in a custom software build, and doubly so for AI. Defining the problem in detail lets us say what sorts of AI can help us solve it and what that help might look like. It also lets us scope the problem and the cost of solving it, a critical step in all custom builds.
Record Why Off-the-Shelf Solutions Don’t Work for You
Deciding whether to build custom AI compounds the problem. We can’t say where exactly off-the-shelf AI is insufficient if we don’t have a problem statement. In software, misalignment isn’t exactly a binary thing. Depending on how the off-the-shelf solution doesn’t fit, we may be able to adapt it for ourselves.
Regulatory & Security Misalignment
Some misalignments can’t be worked around. If I need AI in an air-gapped network, I won’t be able to use externally hosted solutions. If the IP posture of a generative system isn’t aligned with our corporate or client needs, I won’t be able to use that tool. Security, IP, and regulatory concerns are the largest reasons that off-the-shelf is a non-starter.
AI Approach Mismatch
Some mismatches are inherent in the need or approach. If you know that generative AI, machine learning, and computer vision aren’t fit, you may be hard-pressed to find off-the-shelf options. In those situations, you may want to adapt your problem to use one of these approaches. While GenAI, vision, or machine learning may not be able to solve your whole problem, maybe they can solve a part of it and lighten the load. This is a cost to consider when comparing this against a bespoke solution if they can. If they can’t, you may need a custom AI solution.
Some mismatches can be overcome with training, more tools, or adapting our expectations. In these cases, something about our interaction with the off-the-shelf solutions is less than ideal. The input format isn’t aligned with our data at rest. The output from the system is prose, but we’d prefer structured text or vice versa. The accuracy of the AI system isn’t what we need it to be.
None of the above problems are immediate deal-breakers. Instead, they each increase the cost of interacting with the off-the-shelf AI system. Needing to transform your data at rest for AI could add development time in service of building a tool. It might add a per-interaction cost for a user to manually translate the input. Issues of system accuracy can be handled in a similar fashion. We can build extra tools to filter bad output before it gets to the user, or we could train the user to be vigilant for bad output.
Weigh Costs & Benefits of Custom Artificial Intelligence Solutions
The important thing is to notice when an off-the-shelf tool doesn’t meet our needs and record that mismatch. Ultimately, no tool, not even a custom-built one, will be perfect. Development time and capital are finite. Somewhere, we must make compromises to arrive on time and on budget.
As we noted above, the off-the-shelf tools also need time and money investments to be fit to purpose. They need adapters, your users need training, and so on. The question of when to reach for custom AI solutions is really one of cost-benefit analysis. Will it cost us more to use an off-the-shelf tool as is, to build improvements that make it acceptable, or to build a custom system from scratch? This decision should be easier to make in light of the information we recorded while evaluating off-the-shelf solutions.
So, Should You Use Custom AI Solutions?
Off-the-rack AI solutions won’t fit every company and every need. Unfortunately, that’s also true of custom AI solutions. While a bespoke solution is going to fit just right, it takes additional time and money to produce. Before you set down that path, you need to compare your options. Think hard about how much the mismatch between commercial AI and your needs costs in time and in money.
If you need help understanding the cost and possibilities with custom AI solutions, or you’re having trouble identifying where commercial offerings may not serve your needs, please reach out to us! We’d love to talk to you about that.
Unlock the Power of AI Engineering
From optimizing manufacturing materials to analyzing and predicting equipment maintenance schedules, see how we’re applying custom AI software solutions.
You Might Also Like
