Five AI services (you can afford) to help your business now.
Once you remove the varnish, buzzwords, future-hype, mission statements and niceties, your business offers something to someone in exchange for money with the goal of making a profit. Most of the time, that’s not easy to do. In the current crisis, it’s extremely difficult.
Cue the headlines on social media that talk of how important technology is to innovate and transform your business. We agree, and one of our founding values is “Possibility”, but we also know that not all businesses are able to simply overhaul and hope for the best. So, this series explains five technologies that are quick to implement, do not require a huge budget, and can directly help you get more business or retain more business than your competitors:
- Recommendation Engines
- Enhanced Marketing Automation
- ML-based forecasting
- Sentiment Analysis
We’ll start with an introduction to each of these technologies before we explore each in more detail in subsequent articles.
Chatbot is probably the worst thing to call a chatbot – they should be named conversational agents. If you’re unfamiliar with them, they will be most recognisable to you as the little pop-up on webpages that say things like “Hi, I’m Rory the Robot, ask me anything”.
The key behind them is that they are like talking with a human using natural language. For example, you don’t have to ask “Show me information on product xyz”, you can simply ask, “Hey, is this thing available in red?” or, “What colours have you got?” etc. They can work out variations in the way you ask questions or use different words for the same thing – plus they learn and get better over time. This is done mostly through something called NLP and it’s not limited to the web-chat I’ve already mentioned: It can be used to drive a two-way sms interaction, a mobile app chat and even a voice-based conversation.
They can be poorly implemented (we’ll show you some of those soon), but with some simple steps they can be very helpful to do things like:
- Provide answers to questions 24/7.
- Make recommendations on what options there are, alternatives if there’s no stock etc.
- Provide answers to transactional questions, like where’s my order or what’s my account balance
- Kick off lead generation flows with a highly contextual starting point.
We’ll dig deeper into these in the detailed chatbot article; There are many simple yet powerful ways to use this!
2. Recommendation Engines
When you’re browsing for products and click on one for more detail, you’ll often see additional products that appear under titles such as “You may also like…”, or “People who bought this also bought…..”. These are driven by recommendation engines and can be highly effective: Amazon retail makes 35% of their revenue from recommendations. A recent study by Accenture also found that 91% of consumers will shop with brands that “know them well” i.e. that remember them, what they like, and provide them with relevant offers.
Recommendation engines are not just for online shopping sites. They can be used to suggest additional content (like pointing to another article or blog post), highlight similar job opportunities a candidate may be interested in on a job website, list items that are popular right now with similar clients or suggest which items could still fit in a box before it ships.
Surely you need a team of data scientists, a pile of data and a huge inventory of products to see benefit? Actually not – Amazon’s models are available to use out-the-box, and you can build, train, tune and deploy your own personalised recommendations in a couple of weeks.
3. Sentiment Analysis
Anthropologist Ray Birdwhistell estimated that we can make and recognise round 250,000 facial expressions. Somaybe it’s not that impressive that we’ve taught computers to recognise 7 core emotions. What is impressive is that it can be done at low cost, by anyone, very quickly. And although not directly related to emotion or sentiment, you can also identify objects, content, extract text, who the person is, do comparisons and whatever else you want to train the service to identify.
The same goes for analysing text – finding key phrases, words and brands, where people or places are named, medical terms and, of course, sentiment.
Again, many people see this as being too complicated or expensive for their business, but the reality is that it’s extremely accessible and affordable. In fact, getting a dashboard up to monitor what’s being said about your company or brand on Twitter is particularly quick and inexpensive, and with thousands of tweets generated per second, it’s a great place to start. Just think about how important it could be to respond quickly to negative sentiment to avoid reputational damage, or how much mileage you could get from turning people who post positive reviews into ambassadors for your business
4. Enhanced Marketing Automation
How many promotional emails, messages, loyalty program offers and similar attacks on your sanity do you get every day? How many of them work? Not many, I’m sure – except for those that bring you the right offer, at the right time, using the right delivery method. So how do you make sure it’s your message that meets these criteria? By having a system that can:
- Segment your audience based on something they’ve done, purchased, clicked on, asked for or some other characteristics you can define.
- Decide what to do at each touchpoint – splitting the customer journey depending on what their reaction was to a message or activity. A personalised recommendation engine can provide the next best step here too, so this is not a prescribed action you have to think of in advance. I wonder where we can get one of those….?
- Use multiple channels to deliver the offer or message – sms may work well for younger people and where the offer is based on time (pizza for dinner) or location (at the airport). Sending an email on a Sunday with a special offer to visit the climbing centre would be poorly timed and not many people read emails like that on the weekend. The channels that work best get used more, automatically
Marketing automation and workflows can get complex and confused, but they don’t have to be. You can start with a simple journey, analyse the results using the built-in analytics tools, get the data into personalisation engines and run it again, all without having to commit to expensive software licensing, messaging systems, hardware or a raft of gurus
So many things are forecast. The weather. Sales revenues. Resource demand. What Kanye West will say next. OK, maybe not that last one, but most forecasts have one thing in common – they rely almost entirely on the past i.e. historical/time-series data. There’s huge value in that, but accuracy is low when trends are not regular, or where other factors that will not be repeated have affected the patterns you’re seeing.
Using machine learning, you can improve forecasts up to 50% by combining other sources of data (not just time-series data) to discover their effect. For example, forecasts for foot-traffic through a shopping centre may look different when adding local event calendars.
Again, you don’t need a team of scientists and actuaries to take advantage of this technology – give the service your historical data, link the related data, and Forecast can select the right model to use, as well as train and optimise it. One company indicated some despair that a 2-week project they ran could have been completed in 10 minutes using Forecast.
There you have it – five technologies you can use in your business without having to completely reinvent yourself or sell the farm to access. If you’re wondering what inexpensive means, get in touch with us to discuss your needs and we’ll give you some numbers. We’d be delighted to help you with some ideas too, with no requirement to commit.