SXSW 2016: Messaging Bots

My most recent post on LinkedIn cross-posted here.

————————————

This past weekend, I had the pleasure of moderating the “Testing Your (Aritificial) Intelligence” panel at SXSW. On the panel, we had Dror Oren from Kasisto (vertical messaging assistant for banking), Alex Lebrun from Facebook M (horizontal messaging assistant) and Dimitra Vergryi from SRI (runs the speech research lab). Much of our early discussion was regarding the future of assistants in-part drawing on some of my experiences from SRI and Tempo AI but we quickly moved to the hot topic of messaging bots and the role of AI.

If you’re not familiar with messaging bots, I encourage you to read “The Bot Paradigm” from The Information as well as Jonathan Libov’s super-aggregation of messaging UX design. In short, messaging (communication) represents our primary workflow (day-to-day) and as exemplified by applications like WeChat, messaging can be used to facilitate other experiences (eg shopping, sending money, customer support).

This new modality has the disruptive potential of replacing all app experiences and if so, would represent an opportunity as significant as the iPhone App Store was to publishers.

Some panel takeaways:

  1. Horizontal (general purpose) assistants are hard! Users do not know what to ask nor remember what primary use case to use the assistant for. In addition, the technical complexity of the system is exponentially greater in that you have to deal with out of context and extreme queries. For example, Alex mentioned they no longer intend to support “pet delivery” in FB M since that’s a use case out of their machine automation wheelhouse.
  2. Standalone messaging apps don’t stand a chance. I found this feedback interesting in that they are advocating that messaging bot experiences need to be built upon the existing large players (eg WhatsApp, Slack, Facebook Messenger, Google Hangouts etc). This intuitively makes sense since the App Store is no longer a great growth channel.
  3. App bot discovery and the ultimate app store for messaging bots is still unclear. Having played with the app store within FB Messenger, I found the workflow to be sub-par. Alex and Dror suggested displaying related app bots as you type within Messenger (in-context). This certainly would be an improvement at  smaller scale but will remain a challenge if we have 100s of K of messaging bots. The ultimate design of the APIs that the messaging apps provide will be critical.
  4. Personas are necessary for messaging bots to. I found this interesting in that personas can certainly lighten the mood and set expectations but can also be tiring when you want expedience. See slides I presented on mobile AI/UX design at SXSW a previous year (wish I had a video of the presentation with the voice-over).
  5. Machine-powered conversational NLP is a very long-way out. Dror felt you must go vertical and thus their focus on the banking sector. Alex said that the FB research team described conversational NLP as level 10 technical difficulty and AlphaGo as level 1 (my FB post on the AlphaGo AI milestone). All of the panelists also acknowledged that building a data corpus (collection of queries/conversation) for training is the real challenge in any AI application.
  6. Key initial industries to be disrupted by messaging bots were financial services, customer service, commerce and travel. Not surprisingly, some of these categories are identical to what we saw with early vertical Siri-clones.

In summary, messaging bots represent the next growth channel and will spawn new billion dollar opportunities; excited to see it happen!

Reflecting on the Productivity Category

My most recent post on LinkedIn cross-posted here.

————————————

Having spent the last 4 years working in the productivity category, I wanted to share some learnings from the space. Note, everything below is slanted towards an early-stage startup ultimately selling prosumer to SMB.

1/ Individually vs Group Useful
Sometimes called single-player vs multi-player mode but the premise is the same. Is your app useful to the individual or useful to a group (eg workplace).

For example, Asana is useful in a group but not so much as an individual. Evernote is useful to the individual but not so much in a group. Rare apps like Dropbox are both but this is not a requirement. Given that productivity apps typically touch personal data, they are not generally viral. Brainstorm ways to be both individually and group useful.

2/ Direct-to-Prosumer vs Top-Down
Is your app B2C or B2B – the trend is to drive prosumer adoption and then to sell to the SMB. Yammer pioneered this model; Xobni extended this model and in some instances it can even backfire such as when Microsoft asked their employees to uninstall the Xobni plugin.

If your app is direct-to-prosumer, you need to think about whether you can really get to 100s of M of users or 10s of M at a really high price-point. Evernote is probably the best example and is still struggling with accelerating freemium growth.

With SMB deployment, you can charge a higher SAAS price-point plus drive more seats. Ideally, you can achieve this without a sales team like Slack and Github have demonstrated but realistically, you will need an inside sales team and ignoring this reality is why I’ve seen many productivity app developers fail.

3/ Replacing In a Category vs Creating a Category
Are you an app the SMB already pays for or something they don’t yet pay for? Slack, as an example is a new category in that SMBs did not previously pay for chat. However, if are a “Todo” app, you are probably competing with an already existing tool such a Jira that the SMB pays for. Getting a company to switch tools is hard and thus why I recommend targeting super-SMBs when replacing in a category.

4/ Create Lock-In
Most productivity apps aggregate some cloud data (files, CRM etc). In the PIM (email, calendar, address book) category, we aggregated email accounts (Google, Exchange, iCloud). Being an aggregator means we are the presentation layer. But without any content exclusively stored in our system, the user can switch presentation layers without penalty.

To create lock-in, some options:
(a) Store some content exclusively
(b) Require upfront customization such as Salesforce
(c) Introduce paid; this will be the best thing you can ever do and will improve all of your metrics
(d) Achieve network effect. Hard to pull of but if you get it, hold on to it!

5/ Be Pervasive
Productivity is a lifestyle and is integrated across personal and work. Although mobile-first is my recommendation, don’t discount laptop usage at the workplace. Apps must be avail on all screens otherwise you are destined to fail.

6/ Have a Macro Thesis
Product management in productivity is hard – there is no 80/20. Workflows are unique to each individual and attempting to change and behave like a coach is exceptionally hard except when managed down (eg everyone has to use Expensify). It’s best to embrace the existing workflow and improve while also maintaining a macro product thesis. Without a thesis, your resultant app will look like the Settings dialogue in Office.

7/ Don’t Be Too Smart
As I’ve written previously, I believe predictive intelligence will be a new layer on all apps. But in the productivity, trust is always the #1 feature. Failing to sync an email will be an immediate deal-breaker. Optimize on precision (accuracy) vs recall (# of results) and ask the user when unsure versus being too smart. Although most contact mergers are pretty good, the 1 out of 10 times they fail is why many don’t embrace.