Zapier Expert Walkthrough: Akkio Integration by Luhhu

At Luhhu, we love the chance to play with new tools, and AI is something we’re quite excited with. When the team at Akkio reached out to invite us to experiment with their new Zapier integration, we jumped at the chance.

Akkio makes it quick and easy to build AI decision models and thus leverage them in your automated workflows. Using it to estimate whether a new inquiry is likely to convert to a paying customer is one obvious use case.

Keen to explore other ways of using it - we struck on an idea. Our founder spends a good deal of time on Twitter talking about and generally promoting automation. Some of those tweets are planned in advance. 

So, what if we could use Akkio + Zapier to analyse them before posting - identify ones likely to do well and make tweaks where needed. That would be cool!‍

Building the AI model in Akkio

Getting started with Akkio is really simple. Sign up, log in, and you’ll be presented with some pre-build models or the option to build new ones.

In the top left, click “Create Flow”. Next, you’ll be prompted to select your input source. Direct links to Airtable and Google Sheets are in the works, but for now you can upload a CSV, Excel or JSON file - this is what we chose to do.

All AI models need data to learn with. In our case, that’s historical tweets alongside engagement stats so Akkio can get to know what’s worked before.

Luckily, Twitter is happy to hand that right over, just check out the Analytics section and hit the “Export Data” button top right (and select ‘by Tweet’).

It takes a few moments, but you’ll eventually end up with something like this. At this point feel free to look through the dataset, trimming column and irrelevant tweets as necessary. You’re looking for a clean dataset, so removing short replies which weren’t intended for wider engagement is a good idea. 

Back over in Akkio we uploaded our tidied up dataset - which Akkio will then ingest and do its best to categorize.

Next, we add a new step to our Akkio workflow, this time selecting a “Flow -> Predict” step. This is where the magic is going to happen!

We tell Akkio which fields we’d like it to predict when it receives a new data entry and using the existing dataset upload, it will build a model to do just that.

Give it 30 seconds or so, and you’ll be presented back with some fascinating insights. In a nutshell Akkio will tell you what elements of your dataset are most important at making predictions and then run a subset of your dataset through the model and see how good it was at predicting compared to the actual data.

At this point it’s worth saying that there are two things that will make the predictive model more useful. a) data - the more and varied the better and b) time spent tweaking the model - both in terms of which fields to predict and which to ignore.

Building your zap

We’ll come back and fiddle with our dataset later, but for now, we’re excited to get a working zap!

First you need to accept this invite to add Akkio to your Zapier account. 

In our zap, the trigger is a new record in the Airtable where we plan new Twitter posts - you can see an example above.

Then we need to add our Akkio action step.

There is only one option right now ‘Make Prediction’ - just what we need.

We need to select the flow we build in Akkio and then map the tweet text from Airtable.

With that done, click through to test your action step.

And voila! We have our prediction. According to everything Akkio can surmise about our prior tweeting habits, it thinks this tweet will get 373 impressions, 12 engagements with an engagement rate of 0.05. Not bad!

So what next? In this case, we’re going to have Zapier send us an email, but only if the predicted number of impressions is above 300.

One of the great features of Zapier, is that it lets you build dynamic filters. In this case the zap will only continue running past this step if the impressions value from Akkio is > 300. For our test it was.

In the final step of the zap, we’re sending an email to ourselves, with the tweet text in the subject line and a note of likely impression # in the body. This acts as a notice to our team to put said tweet on the approved list.

Lots of possibilities

Once you start to get familiar with Akkio, and comfortable with curating and labelling data to train the AI with, you’ll realise the sort of power that comes with the ability to classify inputs with machine learning models.

Sales, customer service, feedback analysis - even content planning as we’ve shown in our example - can all benefit, especially when Akkio can be seamlessly integrated with your other apps using Zapier.

Need help getting set up? Please reach out to Luhhu.