Artificial intelligence (or AI, as it’s commonly known) has power to change work in the short and long terms. Advances in techniques like machine learning (ML) and natural language processing (NLP) offer huge ways to help small businesses cut costs and differentiate themselves.

Gartner predicts that by 2021, brands that redesign their websites to support voice and visual search will increase digital commerce revenue by 30 percent. In the meantime, small business owners use AI to solve several problems – including interruptions.

Cogito, an AI-first behavioral analytics firm,  uses AI to analyze and improve human interactions in real time. By tracking speakers’ nonverbal signals  (including emphasis and mimicry) Cogito helps users improve their communications skills. When Eli Orkin, Cogito’s marketing manager, kept cutting one of his colleagues off, the tool sent him a “Frequent overlaps” notification that helped him refocus.

Despite this potential, AI is still its early days. In a report available to clients, Gartner predicts that just five percent of organizations will gain value from AI through 2019. That’s largely because most teams lack the talent needed to manage AI projects.

It’s also tough to predict concrete benefits from artificial intelligence projects. Without knowing the benefits upfront — like time or money saved via AI — it’s understandable why small and midsize businesses (SMBs) worry that now’s not the time to invest in the unknown of AI.

Luckily, artificial intelligence projects don’t have to be all or nothing. These six tips will keep your AI expectations in check— and help you learn from others’ past mistakes:

  • Adjust your definition of “AI project”
  • Aim for “soft” outcomes
  • Use AI to augment employees’ work instead of replacing them
  • Know that early AI projects often fail
  • Include AI in your software search
  • Use open source tools to learn about AI

Adjust your definition of “AI project”

It’s tempting to think of AI as its own entity. This technology is so vast and complex that it can be hard to consider in relation to your work.

But “AI projects” don’t exist in a vacuum. Instead, AI enhances work that you’re already doing.

For example, take NLP. This AI technique lets computer programs understand spoken human language. It’s often used in call center chatbots to help businesses lower costs or differentiate themselves. When chatbots are used to answer common customer questions, employees get more time to do strategic work.

GetApp’s advice

Start by confirming which problems your small business needs to solve most (like cutting costs or giving your team more time to work on high value projects). Once you know what your business must do next in order to grow, you can learn which AI techniques (if any) are best equipped to solve those problems.

Aim for “soft” outcomes

AI is touted as a transformational tool, so it’s not surprising that global spending is expected to eclipse $19.1 billion this year. But if you ask business leaders what they want AI to achieve, the answer is often unrealistic.

Gartner’s survey of CIOs, Research Circle members, and executives responsible for developing AI found that most folks expected AI to improve their decision making and process efficiency. They also expected instant financial gains — and are thus bound to be let down.

Many executives said that senior management in their organizations prioritized AI “only in terms of immediate financial impact.” Gartner’s advice for folks looking to implement AI? Don’t use short-term financial gain as your success metric.

GetApp’s advice

Don’t measure AI’s value strictly based on finances. If you can use AI to boost team productivity or customer satisfaction, you’re way ahead of the game.

For example, one executive in Gartner’s survey used AI analytics tools to decrease error rates in his company’s financial forecasts. The outcomes they used to boost performance decreased their error rate from 20 percent down to one percent.

Small business owners who want to see the same success with AI can start by trying to solve one specific business problem (like reducing time spent in emails, thereby increasing productivity). Even if you don’t get the results you want, the lessons learned will help you run subsequent trials.

Augment employees’ work instead of replacing them

Many forecasts about AI focus on projected job loss. It’s true that some industries (like manufacturing) will take a likely hit as AI keeps automating routine tasks that it can easily learn. But this is just one side of the story.

AI currently works best when humans and machines collaborate on tasks. That’s because machines are better than humans at certain tasks (like efficiently mining large amounts of documents to find data), while humans outperform machines at other tasks (like reading social cues).

GetApp’s advice

If you’re considering how to use AI, revisit the business problems you need to solve most. Then, ask yourself how colleagues within your small business can use AI to improve the work they’re already doing to solve those problems.

Gartner cites Udacity as one example. The education company built its own chatbot based on past sales conversations and transcripts. The bot used this data to differentiate which sets of words led to successful sales transactions compared to sets of words that yielded unsuccessful transactions.

Then, Udacity’s engineers designed the bot to recommend successful dialogue to its salespeople. This tactic improved sales by 50 percent.

Know that early AI projects often fail

I know what you’re thinking: “You just told me how to use AI successfully! Now you’re telling me that what I do will most likely fail?!”

I am. Using any new technology means you’ll have to surmount inevitable hills. For several reasons, this is especially true with AI.

Gartner predicts that bias in data, algorithms, and the team members who choose the data for AI systems will cause 85 percent of AI projects to have errors through 2022. And that’s if you’ll be able to hire AI talent at all. Businesses like Udacity are the exception, not the rule.

Most organizations — regardless of size — don’t have the talent to support AI. That’s due to a severe shortage of engineers who can build AI algorithms and data scientists who can comb them for bias.

AI systems need constant retraining so they can learn from subsequent data and produce better results. These results can’t be achieved if you don’t have employees to clean, input, and retrain the data.

According to Element AI, only 10,000 people in the entire world are qualified to do serious AI research. Until this talent pool widens, AI’s use cases will be limited.

GetApp’s advice

Adjust your expectations for AI by acknowledging today’s talent restrictions. One executive told Gartner that they tried using AI to analyze past product failures, then predict which of its current products would fail as well.

But since colleagues didn’t give the machine enough details about why product failures were predicted, engineers couldn’t trust the system. As a result, they didn’t reduce their rate of product failure.

Silicon Valley’s “fail fast” approach applies tenfold to AI. Not every use of AI will work wonders for your business – and that’s okay. Learning what won’t work is just as valuable as learning what will work. The key is to be realistic from the start.

Include AI in your software search

In case you’re not thrilled by the thought of failure, I have good news: Planning ahead for project failure doesn’t mean that you can’t start using AI to benefit your business. Finding cost-effective ways to test AI doesn’t have to be as hard as it sounds.

Many cloud software tools — including popular collaboration apps like Slack — build AI-powered features into their product roadmaps. And since these vendors work with larger data sets that they train and maintain, you don’t need to worry about having in-house AI talent to manage your data: Vendors can do the heavy lifting for you.

For example, one of Slack’s goals is to help users be more productive and avoid information overload. To do that, its “Highlights” feature uses AI to find messages that it thinks are most important for you to see.

Slack also uses a Spark pipeline to find the amount of time users spend reading and writing in specific channels. Then, it recommends new channels for users to join based on where Slack’s AI predicts you’ll be most active.

GetApp’s advice

The next time you shop for software, ask vendors if they offer AI-powered features. Knowing which questions you should ask vendors will help you answer two key questions before you invest in a software purchase:

  1. Does this software really use AI?
  2. Will it work with the data and systems my small business uses now?

Gartner predicts that by 2021, more than 70 percent of business software users will have cloud-based tools at their disposal. Most of these software will use everyday AI (the use of AI techniques within commonly used apps), which differs from general AI (tools that attempt human-like thinking, perception, and action).

Investing in software with everyday AI is a low-risk way to introduce your small business to AI. As more vendors incorporate AI-powered features into their products, you’ll have an advantage the earlier you learn.

Use open source tools to learn about AI

If you’re still worried that AI sounds impossible, fear not: There are an increasing number of ways to learn about AI via open source platforms. Earlier this year, Mazin Gilbert — VP of advanced technology at AT&T — told Quartz that AI’s real breakthrough in recent years is open source technology.

Gilbert said that fundamental AI algorithms haven’t changed — but the release of open source algorithms reduced barrier to entry. In theory, that gives a Kansas-based accounting firm the same power as a tech giant like Google.

GetApp’s advice

Several open source AI frameworks and APIs are available online. This presents another low-risk way of introducing your small business to AI.

Using AI-powered cloud software and giving your team the freedom to learn new algorithms presents a win-win. It lets you use AI from proven vendors to your small business’ advantage. It also sets the groundwork for training in-house talent that can find new ways for AI to grow your business. This removes some pressure while giving your employees room to grow.

Want to learn more about AI for project management?