Customer support is hard

Before we talk about Help Scout No Surveys, let’s be honest. The art of looking after customers is as old as the hills, but despite this many companies struggle.

We’re a small business with hundreds of customers spread over many countries. All of our support is managed via email by a small remote outsourced team split between Australia and the Philippines.

Our biggest challenge is achieving consistency as we grow. Our support partner is superb, but as the business owner I worry that a poor agent could undo our great reputation, or unnecessarily drive churn.

Surveys don’t cut it

The traditional method of monitoring support is to survey customers. Trouble is, very few of our customers will even press on a simple smiley. This means that results are extremely general. And surveys are always backwards looking. You know afterwards that you had a problem.

It would be so much better to know in the moment that a specific customer is not satisfied. That would allow remedial action before it’s too late.

Could we use Artificial Intelligence (AI) to automatically scan conversations and alert problematic ones?

There is value in Artificial Intelligence (AI) for small companies too

AI has been hyped as the ‘next best thing’ for some time now, but we assumed that it is out of reach for any company that does not have both a huge data lake and a team of geniuses.

We challenged that assumption by commissioning a computer science research project at the University of Sydney. That successfully demonstrated that AI can be applied within our own small business for commercial benefit.

Spurred on by this exciting conclusion we took the next step of conducting sentiment analysis on our customer support conversations.

What is Help Scout No Surveys?

Sentiment Analysis is the analysis of the ‘sentiment’ (or emotion) of written text. It is a branch of AI and Natural Language Processing (NLP).

Because it is fully automatic, the content of all messages can be scanned and their sentiment immediately identified. Sentiment is measured numerically as 4 values – positive, negative, mixed and neutral.

The algorithms are very complex and sophisticated. In summary they ingest written language and return numerical sentiment scores. We fed algorithms with real customer conversations in our support ticketing system on Help Scout.

How we display Help Scout No Surveys results

We display customer sentiment on a bar chart with the conversations on the horizontal axis and sentiment on the vertical. Up is happy (green coloured) and down unhappy (red coloured).

Here is a real example:

There’s a lot of red in this conversation so clearly the customer is unhappy. More concerning is that for the first several messages the negativity is getting worse. When we review the underlying conversation this makes sense:

1. Customer has a pre-sale question. Agent replies “yes”.
2. Customer replies “OK but…” and agent replies with more details.
3. Customer replies “I’ve now purchased but…”.
4. Before the agent replies the customer sends another message “Let me show you…”. Agent replies “I suggest…”
5. Customer replies “Now I’m stuck here…”. Agent replies “do this…”.
6. Customer replies “Now I’m stuck here…”.
7. Before the agent replies the customer sends another message “I’m also stuck here…”

Message 4 is the low point in the conversation and from then on the sentiment improves. It’s not until this point that the agent starts delivering valuable support to the customer.

Trend is as import as colour

In this next real example there is much more green so the customer appears to be happier. In practice we find that’s the direction of the sentiment trend that is more indicative of the conversation health than colour alone.

Again, here’s the conversation actual content:

1. Customer reports 2 problems. The agent replies seeking more details.
2. Customer makes contact again and asks “what’s happening?”. They appear not to have seen the agent’s previous reply. The agent repeats their original response.
3. Customer supplies more details. The agent seeks login details from the customer.
4. Customer supplies the login details.
5. Customer sends more information. The agent solves the problem.
6. Customer is delighted with the solution.

Message 3 is the low point. As with the earlier example, the change in direction occurs once the agent starts delivering valuable support to the customer.

In our experience, the chart trend after 3 exchanges is a good indicator of support relationship health.

No Surveys

Business owners seek rapid oversight of all customer conversations. They want to know whether and where there are lurking problems, and potential drivers of cost or churn. These were key conclusions from discussions with business owners about our findings.

So No Surveys was born!

This is a stack of bar charts, with filtering, each augmented with additional information to allow the business owner to rapidly dive deeper as required. A fully operational demo is here.

How valuable is it?

Lessons that I’ve learned from 10 months using No Surveys internally are:

• I save time and sleep easier – rapid and effective oversight of customer conversations without reading them is possible.
• My fears of hidden problems (in my business) are overblown.
• Some agents can close issues with customers much faster than others. I’m looking forward to unpacking that further.

I haven’t been looking for customer testimonials during the last 10 months, but I’d like to think that it would help me find some. That said, a quick scan through the last 30 days of very green conversations just now did not reveal any, but it’s a great energy boost to see so many happy customers!

It’s only in a minority of conversations – c.20% – where any positive or negative sentiment is identified. I believe this is because the majority of our conversations are short transactional exchanges, with relatively few complex issues to resolve. In almost every case where sentiment is reported, it is better at the end of the conversation than at the start.

So can No Surveys transform customer support?

Not for me, but I sleep easier – I already have great people doing great work. If I had ordinary people doing ordinary work then I expect it would enable me to spot issues, react, and target coaching support. It’s a pain killer not a vitamin!

Curious to trial Help Scout No Surveys in your business?

We have now made No Surveys available for any business who uses Help Scout.