With the rise of dig­i­tal ana­lyt­ics, KPIs have been a focus for mar­keters and unfor­tu­nate­ly it is com­mon for met­rics to be report­ed up the chain to the board, with­out suf­fi­cient expla­na­tion or seg­men­ta­tion.

Specif­i­cal­ly, aggre­gat­ed site wide met­rics such as bounce rate, con­ver­sion rate, and time on site can have exces­sive val­ue placed on them, which in turn skew opin­ions and actions to the detri­ment of over­all site goals and per­for­mance.

(Note: It is prob­a­bly worth just explain­ing what ‘rate’ means, and in most ana­lyt­ics pack­ages it is per ses­sion (or vis­it) rather than per user, this means that with­in a 30 minute time­out win­dow, whether they com­plet­ed an action and whether they did it over sep­a­rate ses­sions.)

Bounce rate is bad if not segmented

My favourite exam­ple for a dys­func­tion­al sitewide KPI is bounce rate – high bounce rate is bad, low bounce rate is good right? This engage­ment met­ric is typ­i­cal­ly cal­cu­lat­ed by some­one only see­ing one page, on most sites you want users to vis­it more than one page when they vis­it?

A great exam­ple of where a bounce rate becomes a bad met­ric is on an infor­ma­tion­al page – imag­ine you search for pass­port infor­ma­tion, you land on a page with the cor­rect infor­ma­tion, you read it, you are unlike­ly to spend more time on the site brows­ing help­ful gov­ern­ment infor­ma­tion, there are oth­er exam­ples such as bag­gage allowance on planes and store open­ing hours.

Crit­i­cal­ly if the infor­ma­tion should be found in a sin­gle page vis­it then a high bounce rate is a almost a pos­i­tive thing, but you still want to add in relat­ed links and infor­ma­tion that helps them nav­i­gate fur­ther – is a bounce rate bor­der­line use­less in these cas­es?

As Avinash Kaushik, Dig­i­tal Mar­ket­ing Evan­ge­list for says…

 

Per­son­al­ly I would say seg­ment and test, nev­er look at a site wide rates and don’t com­pare your­self to aggre­gat­ed site wide aver­ages with­out con­text – you can look for areas to improve from the 404 page right through to prod­uct land­ing pages to spot issues, but the moment you start look­ing at a site wide bounce rate is the moment it becomes a use­less met­ric.

And don’t try to improve it, instead improve expe­ri­ences and facil­i­tate the user jour­ney.

It is also worth address­ing that long time myth that your bounce rate has an impact on your rank­ings, your Google Ana­lyt­ics bounce rate has no rela­tion to your rank­ings. Whilst this is not a direct sig­nal, how­ev­er users who ‘pogo stick’ (a tech­ni­cal term for users being on your site for only a few sec­onds) might be a sig­nal.

So, bounce rate – good for find­ing prob­lem pages, improv­ing PPC land­ing pages, opti­mis­ing your 404 page; bad for report­ing at a site lev­el or mak­ing mean­ing­ful indus­try com­par­isons.

Time on Site

It is also worth not­ing that ‘time on site’ is also a ter­ri­ble met­ric – users are no longer on dialup, they can leave tabs open in their brows­er and go make cof­fee – this, as a sitewide met­ric has so lit­tle val­ue now. If you have a user replay tool on the site you will see just how often tabs are left open, users today who wan­der off or browse on anoth­er tab aren’t real­ly spend­ing time on your site or pages. Addi­tion­al­ly the way it is cal­cu­lat­ed is lim­it­ed and sin­gle page vis­its are exclud­ed.

Conversion and Intent

The oth­er pop­u­lar sitewide met­ric is ecom­merce con­ver­sion rate. Real­is­ti­cal­ly, stock sta­tus and pric­ing has more impact than almost any oth­er ele­ment on con­ver­sion rate (if you don’t have it, or it is over­priced they it will be chal­leng­ing to encour­age users to part with their mon­ey) — this is anoth­er met­ric which if con­sid­ered sitewide los­es all mean­ing.

On a typ­i­cal ecom­merce site activ­i­ty should be direct­ed at improv­ing the num­ber and qual­i­ty of con­ver­sions which may be a dif­fer­ent out­come to sell­ing to every user on every vis­it — users may not be ready to pur­chase but are ready to signup for a newslet­ter or engage social­ly. Like bounce rate, there are times when as a met­ric it works, and this is where mul­ti­vari­ate (or A/B) test­ing should come in, to tru­ly answer if the actions you are imple­ment­ing are improv­ing per­for­mance.

Going back to intent, it may be that many users are using your site for some­thing oth­er than to pur­chase (e.g. the jobs sec­tion). It’s also worth address­ing if peo­ple return­ing for prod­uct infor­ma­tion (such as dri­vers or prod­uct instruc­tions). There is an excel­lent arti­cle on Con­ver­sion Rate as a met­ric here.

Matthew Cur­ry from Love­Honey explains it well, in a video and it is well worth watch­ing (although if at work be aware this imagery comes from their sex toy prod­uct range). Matt argues that you should look at user behav­iours over KPIs. He also explains how some users go to his site for the pic­tures rather than the prod­ucts which can vast­ly cor­rupt the data if these users are fac­tored into con­ver­sion rate cal­cu­la­tions.

Stop reporting sitewide KPIs that don’t ultimately inform

Ulti­mate­ly the ques­tion should be what changes can I put in place to improve vis­i­tor per­for­mance, which isn’t always about improv­ing the num­ber of con­ver­sions, bounce rate, or dwell time, but about improv­ing prof­it (tak­ing into account fac­tors such as cost per sale and aver­age order val­ue).

So the key is to under­stand your users’ intent (i.e. what did they come to the site for?). We should be think­ing about whether their expe­ri­ences reduce the chance of them being able to pur­chase again, and would their jour­neys be sat­is­fied with a sin­gle page view. From this point we can start to move the jour­ney fur­ther.

If we aren’t using the data effec­tive­ly it is ulti­mate­ly use­less, not every user comes to your site to pur­chase, not every user record­ed in ana­lyt­ics is even human (there has been for a few years an increas­ing num­ber of bots delib­er­ate­ly spam­ming ana­lyt­ics accounts), and it is chal­leng­ing to seg­ment out all users who are not there to pur­chase.

Met­rics are crit­i­cal, but tar­get­ing users by intent gen­uine means improve­ments can be made.

Segment users, analyse, and investigate your data

With­out seg­ment­ing the data it is sim­ply not action­able – report­ing on this data sitewide is like report­ing on the weath­er world­wide. So stop imme­di­ate­ly, seg­ment, analyse and inves­ti­gate. Almost all dig­i­tal ana­lyt­ics plat­forms now allow you to seg­ment users (at least on one device) so you can start to see whether the activ­i­ty you do in one chan­nel is pay­ing off.

KPIs for measuring user intent

Not every­thing can be mea­sured with con­ven­tion­al ana­lyt­ics. User expe­ri­ences often need to be addressed at by talk­ing, lis­ten­ing, and con­duct­ing research with real prod­uct users (ide­al­ly both yours and your com­peti­tors). Hit lev­el data com­bined with machine learn­ing can also pro­vide insights into behav­iour and access to this lev­el of anla­y­sis is becom­ing increas­ing­ly acces­si­ble.

Of course offline behav­iour is hard­er to cap­ture but should not be ignored. If you aren’t track­ing phone con­ver­sions with call track­ing, or able to cor­re­late onsite behav­iour with store activ­i­ty you can only ever see a small part of the jour­ney, how­ev­er the scale of this issue may be known and quan­ti­fied.

Increas­ing­ly we can start to spot hints towards the users true intent, right now we have access to their loca­tion, time of day, device and all the actions they com­plete with­in a ses­sion (and if we are lucky in oth­er ses­sions), we maybe able to com­bine this with much more rel­e­vant cus­tomer infor­ma­tion and we can start to build a much clos­er pic­ture leav­ing to us being able to answer the ques­tion.

Then we can real­ly start to move beyond the vis­it based met­rics and start to real­ly look at the vis­i­tor based met­rics that ulti­mate­ly lead us to improv­ing per­for­mance at a much greater lev­el where we can mea­sure user suc­cess and poten­tial­ly in the future move towards that pic­ture of users rather than pages.

Key questions:

UX folks often talk about “delight­ing users”, maybe one day we can have a ‘Delight­ed Rate’?

Are we now mov­ing towards user cen­tric report­ing rather than web­site-cen­tric report­ing?

What kind of met­rics should mar­keters be report­ing on instead?