Filtering Notifications FOR HUMANS

Hi there,

My name is Murray and I’m a final year product design student based in the UK, I came across you guys as part of my research for my major project which initially began with our society’s issue with work-life balance.

Having found that it is the phone and its notifications that are are causing us to work at home as well as deal with home issues at work, my project developed into me designing devices that filter out unnecessary notifications and provide a much more humane method of notifying users in both offices and homes. The devices are hubs at which we can leave our smartphones, whilst giving us the reassurance that if important notifications occur, we will be notified. The user can pick and choose by who, and what, they are notified by, giving the user control over their lives again.

This project is about creating a physical separation between ourselves and the smartphone.

I find this whole topic fascinating and would love to discuss it the people on here, as I’m sure there is knowledge that surpasses my own.

Thanks and I look forward to hearing about some fantastic insights - M

1 Like

Hey Murray!

Loving the sound of your project. My name is Kieran, I’m a PhD student in Dublin, Ireland - my research revolves around mobile notifications too and mediating their delivery on behalf of the receiver as opposed the sender in an intelligent manner with an emphasis on scrutrable and transparent decision making so the user can understand exactly what goes on under the hood (and have input to how the technology operates).

The design of your proejct sounds intriguing - so by leaving the device at the hub, would the important notifications that get past your filter be forwarded to another device of the user for consumption? Or possibly alert them to return to the hub? I love the idea of leaving the phone out of reach - from my readings, self-checking your phone even when you don’t have a notification is a huge problem - I’ve notificed it myself (to the point where I’m almost positive I’ve seen the notification signal flash even though it hasn’t causing me to turn on the screen and check, just to make sure!). Simon Sinek makes some excellent points about leaving the phone ‘out of reach’ too. I could post some research papers I’ve come across in this area if you’re interested? I’d love to hear more about your project too - have you any papers on it yet?

My research to date has been about building a framework/agorithms which can identify important from non important notifications (based on the user’s context) - a paper I wrote on it if you’re interested. However, I’m getting ready to release another experiment which also attempts to gauge the perspective of the user on whether or not they consider their action of opening or dismissing a notification as ‘good’ or ‘bad’ behaviour. My end goal is to identify notifications which generate tasks that could be completed by an agent on behalf of the user as opposed to disrupting the user.

Would love to hear your (and the community’s) thoughts on this too - I’m open to collaborations too. I actually just stumbled upon this website and ‘Time well spent’ yesterday. it’s been so motivating to read all about the fantastic effort going into this.


Hi Kieran,

My apologies for the late reply, I’ve had a busy couple of weeks!

Essentially you can control through an app who and by what is able to notify you through the device. The device itself lights up depending on the type of notification e.g. pulsating for an incoming call or increased light intensity for an important email. The idea is that the user has the choice to say if the device lights up for certain notifications or calls from certain people e.g. the light pulsating if you have a call from your mum but not if your work colleague calls you. The idea is that by giving the user this sense of control they are able to feel reassured and leave their phone knowing that if an important notification arises they will know about it and not be constantly bombarded with meaningless ones as they are physically away from their phone. It’s basically quality over quantity of notifications (there are also cool additional features, such as wireless charging to further support leaving your phone at the hub).

Your research sounds amazing, I would love to hear about it and seems genuinely aligned with the goals I am trying to reach. A piece of research I came across that you may be interested in is Barry Schwartz’s talk ‘The Paradox of Choice’ ( He discusses how at home we now have a choice to work or socialise, something we have never been able to do before. This talk ultimately ended up being one of my key insights and validations for my project.

Let me know your thoughts!


Hi Murray,

I have an idea for a project similar to yours that I’m eager to prototype. My idea was to organise notifications around two variables: priority and time/urgency. High, mid and low priorities are defined by the user (e.g. based on a specific communication channel, person or keyword) and then notifications are withheld in batches based on their priority. High priority notifications would arrive within 20 minutes (or whatever the user prefers, medium within 90 minutes, and low priority every 3 hours.

I don’t have any research to back this up but my intuition tells me that this is would be a pleasant set of rules for users. I have been looking at implementing it using a linux app called RamBox to aggregate all my notifications and using a python script to capture the notifications over dbus.

1 Like

Fast forward to 1:20 in this demo video to see how we let Siempo users batch notifications to desired intervals. Naturally we can expand this feature in a variety of contextually intelligent or ways.

CHT recommends turning off all notifications except from humans, but the notifications problem is much more nuanced. We actually conducted a national survey (using mTurk) of 500+ and found of those who frequently or very frequently feel guilty/unhappy after using some of the apps on their phone, 53% feel frequently or very frequently overwhelmed by notifications from humans.

These stats suggest that even if notifications trigger overuse, filtering out non-human notifications won’t necessarily help.

Hey Murray (and group),

My turn to apologise! I’ve been out of office for the past few weeks, just getting back into the swing of things.

I like how, in the project, you’re giving the user this ‘sense of control’. It’s a big part of what I’ve been researching to date too - trying to enable Human-in-the-loop intelligence. Like you say, it gives the user reassurance. There’s too many black-box algorithms at the moment where the user doesn’t have any input into the decisions made on their behalf, which is terrible from a UX point of view. Then again, (after watching the video you recommended - so fascinating, thanks!) too much input wouldn’t be positive either, as it defeats the purpose of easing cognitive effort and would definitley lower satisfaction. So finding the correct balance between control and automation would be an important goal.

I take it you allow the user to set a whitelist of who is able to contact them, via contact lists or such? What would you think about inferring whether or not a person is important, in a given context? I’m running an experiment in the coming weeks which attempts to identify important contacts in a given context based on accepted/dismissed notifications - I’d love to hear your thoughts on this!

Forgive me if I’m wrong, but we seem to be tackling the same problem but from slightly different angles - you maximize user control (allowing the user to set contacts which can interrupt etc.), I attempt (emphasis on attempt!) to infer contacts, subject etc. which can interrupt. I think this is really cool, because, ideally - we should meet in the middle whereby there’s a balance between giving the user control over what’s important and inferring what’s important (beyond the scope of the rules which the user set - as we can’t expect the user to express every possible situation through settings).

If you’re interested in my experiment, you can check it out here. Feel free to add your email to the ‘interest’ page and you can take part (or share with colleagues who may be interested)!

1 Like

Hey Ross,

That’s a great idea - there isn’t a huge amount of research in the area of identifying the best delivery time for a notification (maximizing it’s contextual value). For example, a social notification that arrives during work hours can wait until 1. My next break (lunch perhaps), 2. I finish work (if it’s not urgent and doesn’t require a reply such as, yes I’ll meet for dinner) or 3. Next Tuesday (when I usually walk pasta the bookstore that’s offering this sale).

Similar to your timings, I did an experiment a few years ago which mediated delivery depending on ‘importance’ which was derived from the sender, the app and the subject of the notification. Based on these variables and a set of rules set by the user (e.g. if sender is VIP and subject is VIP and app is IMP then deliver at NEXT BREAK), a Fuzzy Inference System would deliver notifications 1. immediately, 2. next break in activity (based on google calendar), 3. next free period (2 hour gap of free time in user’s calendar), 4. later 5. much later.

I’d love to hear your thoughts on how much input you think the user would need, and how much you think might be acceptable to infer on their behalf? and also, the types of rule system that a user may enjoy playing with. I know myself that f I was required to enter lots of if-else-then rules I may be turned off from using it as it would require too much upkeep for my changing contexts, tastes, preference, goals etc. User control is essential though I think for something like notification delivery because there are such subtle ‘nuances’ (as mentioned by Andrew already), that a machine couldn’t possibly infer.

1 Like

Hey Andrew,

Really like the sound of what you guys are doing with Siempo and can’t wait to see what you do with intelligent notification delivery - this much aligns with my current research.

Currently, I’m attempting to infer importance/relevance/value etc. using the user’s current context, such as their location, current task, level of hope for current goals, and the incoming notification variables such as sender, subject, app etc. as well as other variables. I’m hoping to nail down some consistency in terms of what the motivation is for interacting with notifications as well as understanding the user’s perception of whether they could condone their interactions as ‘positive’ behaviour in a given context (e.g. opening a social notification while in a work setting may be seen as ‘bad’ behaviour).

I’m wondering if you have published the findings from your mTurk survey? I’d love to see what you unearthed.

You might be interested too in this paper from Martin Pielot and Luz Rello (if you haven’t come across it already) - they found that without notifications (complete blocking), some participants of their study became anxious as they felt disconnected from their social groups. It’s interesting as it seems that there’s definitely a tentative balancing act which needs to take place whereby the user

  1. isn’t getting too many notifications which could cause them to feel overwhelmed
  2. isn’t getting too few notifications which could cause them to feel anxious or disconnected
  3. doesn’t receive any notifications which are not relevant and adding value to their current task/emotional state/well-being
1 Like

Thank you for the encouraging words and for the work you are doing! Context awareness will take things to the next level. Our phones can be smart enough to know when to interrupt us.

We haven’t published the results, but DM me your email address and I will happily send over the raw data.

Balance is the name of the game :slight_smile:

My 14 year-old son and I are working on a web app/SaaS product called Outrigger ( It’s a smart email autoresponder that responds to important emails and includes a link that when clicked, will trigger a non-email notification (i.e. via SMS, Telegram, Slack, etc).

It’s very similar to Slack’s do not disturb feature or iOS’s do not disturb while driving, except for email.

Outrigger currently works with Gmail/G Suite (Office 365 planned) and has user configurable options to only auto-reply to messages marked “Important” by Gmail, never auto-reply to CC’s or BCC’s, and only reply to a new message from the same sender after X amount of time (after an initial auto-reply has been sent).

We are extending Outrigger to make it calendar aware so it can turn itself on and off based on a set schedule and also by observing a user’s calendar for events marked as “busy.” Outrigger will also eventually be able to batch its notifications for delivery after a user frees up.

Also in the roadmap is machine learning to help Outrigger determine which emails and senders to auto-reply to.

I have been using the current version when I have all day or back-to-back meetings or when I need to go “heads down” to work on a project and it’s been extremely liberating.

My son runs the VR club at his high school and guest presenters have used Outrigger to alert him to scheduling or equipment issues that he otherwise wouldn’t have known about (since he isn’t checking his email during the school day).

While Outrigger creates a slight imposition on the sender, feedback has indicated it’s generally a welcome one. Senders appreciate knowing the recipient is away from email while having the option to trigger a notification if their issue is or becomes urgent.

If anyone is interested in trying Outrigger or learning more. Please let me know.



@kfraser what you did here is very similar to what we’re thinking for Outrigger (see post above).

Emerging research shows that batching smartphone notifications can improve mental health outcomes.

Compared to those in the control, participants whose notifications were batched three-times-a-day experienced significantly improved attention, stress, perceived productivity, happiness, felt interruptions, and control over their phone.