Is data democratisation really a good idea?

It was Clive Hum­by, the math­e­mati­cian and data sci­en­tist behind Tesco’s Club­card, who coined the phrase “data is the new oil”. It has since become some­thing of a cor­po­rate cliché, uttered by every con­sul­tant and CEO who wants to be seen to under­stand the dig­i­tal econ­o­my. 

For Ved Sen, head of busi­ness inno­va­tion in the UK and Ire­land for Tata Con­sul­tan­cy Ser­vices, this max­im doesn’t quite ring true. He would instead describe data as “the new plas­tic”, because “we cre­ate a lot, we strug­gle to know what to do with it and it tends to turn up in the wrong places. And, for all the talk about the democ­ra­ti­sa­tion of data, busi­ness is not yet cul­tur­al­ly geared up to han­dle this. There’s a lot to do.”

Cer­tain­ly, there are ben­e­fits to be had from data democ­ra­ti­sa­tion: the rel­a­tive­ly new school of thought that data should be made read­i­ly avail­able through­out an organ­i­sa­tion instead of being kept in silos. The idea is that this pro­vides a basis for more informed deci­sions through­out the organ­i­sa­tion, as well as pro­mot­ing inno­va­tion. Some even claim that empow­er­ing employ­ees by “digi­tis­ing” a com­pa­ny, as West­ern Union calls it, will be a game-chang­er. 

Yet only 27% of exec­u­tives sur­veyed by MicroS­trat­e­gy last year said they felt that they’d built the right organ­i­sa­tion­al cul­ture to sup­port a data democ­ra­cy. Indeed, the many hur­dles to be sur­mount­ed in depos­ing a data dic­ta­tor­ship – aside from obvi­ous ones, such as con­cerns about secu­ri­ty and, for SMEs, sim­ply find­ing the time – are only start­ing to become clear.  

Trust and transparency

Matthew O’Kane is glob­al head of AI solu­tions at tech ser­vices com­pa­ny Cog­nizant. He believes that com­pa­nies need to han­dle the democ­ra­ti­sa­tion process with great care. In part, this is because of the poten­tial inter­de­part­men­tal sen­si­tiv­i­ties about data own­er­ship, espe­cial­ly in larg­er organ­i­sa­tions. A recent mul­ti-indus­try study by McK­in­sey found that, for teams request­ing access to inter­nal data beyond their depart­men­tal remit, the response time could be mea­sured in months in 53% of cas­es.

O’Kane cites one retail bank’s board-lev­el dik­tat that ordered the cen­tral­i­sa­tion of all data overnight but also gave reas­sur­ances that no mate­r­i­al would be used with­out the con­sent of the team that orig­i­nat­ed it. 

Once a ful­ly acces­si­ble cen­tralised data store has been estab­lished, pro­mot­ing it as such can be help­ful, he says. For instance, AT&T rebrand­ed its inter­nal online mar­ket­place, Amp, as a com­pa­ny-wide data hub last year. This sort of “per­son­i­fi­ca­tion can cre­ate trust in the data”, accord­ing to O’Kane. 

And trust is vital. When employ­ees are offered greater access to data, they don’t nec­es­sar­i­ly believe what they’re pre­sent­ed with. Respon­dents to a 2018 sur­vey by Exper­ian said that they con­sid­ered 30% of the data held by their com­pa­nies to be inac­cu­rate on aver­age.

This is not the only issue of trust that affects data democ­ra­ti­sa­tion, accord­ing to Kevin Hane­gan, a found­ing part­ner of the Data Lit­er­a­cy Project and chair of its advi­so­ry board.

“The num­ber-one thing I hear from the many CEOs and CIOs I talk to is that they don’t feel ready for the democ­ra­ti­sa­tion of data, because they don’t trust employ­ees to make the right deci­sions using it,” he says. “There’s a lot of talk of soft­ware being the solu­tion, but tech­nol­o­gy is the least part of this. Any­one giv­en access to data needs to be able to inter­pret it. Until then. it’s like giv­ing some­one a black box that says ‘the answer is A’, to which their imme­di­ate reac­tion might be ‘why is it A? How would I know?’. There isn’t a tool to bring about this change. It won’t hap­pen overnight.”

Reading the numbers

The lack of trust is part of a big­ger weak­ness with the democ­ra­ti­sa­tion con­cept: the fact that most peo­ple strug­gle to deal with sta­tis­tics. A recent US study found that 46% of high-school grad­u­ates were unable to esti­mate how many times a flipped coin would prob­a­bly come up heads in 1,000 toss­es. The most com­mon wrong answers were 25, 50 and 250 times. 

“The prob­lem is that a lot of sta­tis­tics are coun­ter­in­tu­itive and full of sur­pris­es, which became clear dur­ing the Covid cri­sis. That’s espe­cial­ly the case with brains like ours, which are just not built to be good cal­cu­la­tors of, say, prob­a­bil­i­ty.” 

So says Stian West­lake, chief exec­u­tive of the Roy­al Sta­tis­ti­cal Soci­ety, which has found a lucra­tive side­line as a train­ing provider to firms in indus­tries such as phar­ma and petro­chem­i­cals. He adds: “Our brains are very good at see­ing pat­terns, but strug­gle to see ran­dom­ness when it’s there. And it’s often the case that [when it comes to data analy­sis] peo­ple don’t know that they don’t know.”

This applies even in pro­fes­sion­al cir­cles where a good under­stand­ing of data might be con­sid­ered cru­cial. A study of 492 physi­cians for the Jour­nal of the Amer­i­can Med­ical Asso­ci­a­tion this year, for instance, found that their assess­ments of pre-test data led them to over­es­ti­mate the like­li­hood of breast can­cer in a patient by 976%. 

Small won­der, then, that a 2020 study by data ana­lyt­ics com­pa­ny Qlik found that only 17% of lay employ­ees con­sid­ered them­selves con­fi­dent in han­dling data. More than two-thirds (67%) of respon­dents admit­ted that they felt over­whelmed by the num­bers, while 19% said they had gone so far as to find oth­er ways of com­plet­ing a task with­out using data.

The value of data

The dan­ger here is that there will be an expec­ta­tion that lay employ­ees pos­sess the same under­stand­ing of data as its tra­di­tion­al keep­ers, the ana­lysts. It’s why O’Kane sees a grow­ing role for arti­fi­cial intel­li­gence sys­tems in fil­ter­ing data into more com­pre­hen­si­ble pack­ages and also in check­ing all resul­tant deci­sions. 

“After all”, he notes, “a bank man­ag­er doesn’t decide whether or not to approve your loan any­more. A com­put­er does.”

This is also why train­ing is impor­tant, accord­ing to Sen, whose com­pa­ny gives its employ­ees manda­to­ry train­ing in cer­tain aspects of data analy­sis. 

“Giv­en the vol­ume of data that’s out there, it’s fun­da­men­tal that any busi­ness seek­ing to democ­ra­tise data should also edu­cate its peo­ple in using that data,” he says. “With­out learn­ing how to han­dle data objec­tive­ly, we all have bias­es and will lean into our own expe­ri­ences when pre­sent­ed with mate­r­i­al that chal­lenges us. The train­ing is not suf­fi­cient at most com­pa­nies, but the smart ones are real­ly invest­ing in it.”

But there is an argu­ment that, even before a busi­ness address­es data lit­er­a­cy, it must address its data depen­den­cy: the belief that all data is valu­able, even if this doesn’t help solve any known prob­lem. 

“It’s impor­tant for any busi­ness to recog­nise the dynam­ics of com­plex­i­ty in data. And the fact that zoom­ing in on real­ly crit­i­cal infor­ma­tion can some­times be high­ly pre­dic­tive, bet­ter so than when using all the oth­er data that might be added,” says Flo­ri­an Artinger, pro­fes­sor of dig­i­tal busi­ness at Berlin Inter­na­tion­al Uni­ver­si­ty of Applied Sci­ences. 

He cites the wild­ly fluc­tu­at­ing prices of air­line tick­ets over the pan­dem­ic – a prod­uct of the data on which the industry’s pric­ing mod­el is based. 

“Yes, mak­ing data more avail­able can empow­er an employee’s exper­tise, but we shouldn’t be lured into think­ing that an idea which can be backed by data is nec­es­sar­i­ly bet­ter than one that comes from intu­ition or expe­ri­ence. After all, data can also be used sim­ply to ‘cov­er your ass’ and jus­ti­fy inac­tion or a bad deci­sion,” Artinger says. “What we need first is a busi­ness cul­ture that knows not only how but when to use this democ­ra­tised data – and when not to.”