There is no doubt that the poten­tial of AI is excit­ing. Whether it’s dri­ver­less cars or faster drug dis­cov­ery, AI is chang­ing our world and how we live in it.

In the cor­po­rate world, finan­cial ana­lysts are able to use AI to improve the qual­i­ty of their mar­ket analy­sis. Using AI, organ­i­sa­tions can pull togeth­er and analyse mil­lions of pieces of data from com­pa­ny tweets to reg­u­la­to­ry fil­ings and even media appear­ances, to iden­ti­fy pat­terns of lan­guage and behav­iour. In oth­er words, AI can tell you whether a company’s good news is real­ly good news.

Still, despite the excit­ing and futur­is­tic appli­ca­tions of AI tech­nol­o­gy, many ear­ly adopters have strug­gled to deliv­er any ROI or busi­ness val­ue out­side of small, exper­i­men­tal projects. Gart­ner reports that, in 2022, 48% of CIOs have already deployed or plan to deploy AI and machine learn­ing tech­nolo­gies in the next 12 months. How­ev­er, they also report that many of these organ­i­sa­tions won’t be reli­ably using AI in pro­duc­tion until 2025.

The prob­lem isn’t usu­al­ly with AI itself, but rather with deliv­er­ing AI at scale, says James Coomer, vice-pres­i­dent for prod­uct man­age­ment at DDN, which spe­cialis­es in stor­age sys­tems for AI. Hav­ing solved the scal­a­bil­i­ty chal­lenge for some of the largest glob­al AI projects, DDN now helps organ­i­sa­tions plan their AI jour­ney to pro­duc­tion. These chal­lenges are more often linked to the large-scale imple­men­ta­tion and deploy­ment of AI, as organ­i­sa­tions look to deliv­er greater val­ue and inno­va­tion to their cus­tomers and share­hold­ers.

For exam­ple, some use cas­es like online retail and con­sumer pref­er­ence analy­sis can work well in mod­est deploy­ments, using datasets that are not too large in size. On the oth­er hand, nat­ur­al lan­guage pro­cess­ing and real-time video analy­sis need datasets that are well beyond the expe­ri­ence of tra­di­tion­al IT teams.

“Tru­ly deliv­er­ing the ben­e­fits of AI means deal­ing with vol­umes of data that are orders of mag­ni­tude high­er than any­thing we’ve seen before,” says Coomer. “Not only is the vol­ume of data high­er, but much of this data is rich, unstruc­tured infor­ma­tion like video, audio or images.”

For some organ­i­sa­tions, the abil­i­ty to gath­er and process very large vol­umes of data can unlock the next lev­el of inno­va­tion. For exam­ple, DDN recent­ly worked with Recur­sion Phar­ma­ceu­ti­cals to use AI to accel­er­ate the process of drug dis­cov­ery. “Using AI rather than con­ven­tion­al meth­ods can mas­sive­ly speed up our abil­i­ty to learn and devel­op new vac­cines or med­ica­tions,” says Coomer.

DDN has 20 years’ expe­ri­ence sup­port­ing cus­tomers’ AI projects across sec­tors such as finan­cial ser­vices, auto­mo­tive and life sci­ences. It doesn’t sur­prise Coomer that some experts report that 87% of AI projects nev­er make it past the pro­to­type stage. While projects are small, data vol­umes can be eas­i­ly man­aged. But once a mod­el is suc­cess­ful, it will accel­er­ate, and the vol­ume of data and rate of data inges­tion will both increase rapid­ly. “It’s some­thing that often takes peo­ple by sur­prise,” he says.

Despite this, Coomer is opti­mistic that AI is on the verge of throw­ing off its rep­u­ta­tion for being com­pli­cat­ed, cost­ly and fre­quent­ly unsuc­cess­ful. “What we are now see­ing is that the mar­ket is becom­ing more mature, the tools are more mature, and it’s often pos­si­ble to buy AI-opti­mised sys­tems that make it much eas­i­er for organ­i­sa­tions to run AI projects at scale,” he says.

Tru­ly deliv­er­ing the ben­e­fits of AI means deal­ing with vol­umes of data that are orders of mag­ni­tude high­er than any­thing we’ve seen before

At the same time, we are build­ing a bet­ter under­stand­ing of how to run AI projects to the same stan­dards as oth­er types of devel­op­ment, Coomer adds. DDN advis­es clients to con­sid­er adopt­ing the ‘AI fac­to­ry’ approach. This is an approach to AI projects that allows organ­i­sa­tions to build a con­sis­tent frame­work of process­es around peo­ple, data, process­es and pro­duc­tion that can be repli­cat­ed, mak­ing it eas­i­er to scale AI projects in future.

If a tra­di­tion­al fac­to­ry builds a large num­ber of phys­i­cal prod­ucts in a cost-effec­tive and reli­able man­ner, then the AI fac­to­ry does the same with AI solu­tions. An AI fac­to­ry pro­vides a frame­work for organ­i­sa­tions to build large num­bers of AI mod­els reli­ably, and at scale. It is more than just MLOps (machine learn­ing oper­a­tions) – it com­bines this approach with DataOps, Mod­e­lOps and DevOps to cre­ate a con­sis­tent approach to accel­er­at­ing
AI devel­op­ment.

An AI fac­to­ry frame­work might define agreed process­es and approach­es to things like how secu­ri­ty is applied to data, how the data plat­form is man­aged, or how silos are reduced to improve over­all per­for­mance. “It’s a ref­er­ence archi­tec­ture that reduces the risk and work involved in build­ing AI at scale,” explains Coomer.

By adopt­ing stan­dard­ised process­es and gov­er­nance, the risks asso­ci­at­ed with AI are mas­sive­ly reduced. For organ­i­sa­tions that are new to AI, Coomer sug­gests think­ing about AI projects in terms of four
essen­tial require­ments:

• First, you need to be able to iden­ti­fy what data is need­ed; what is the right data source and how will you cap­ture and organ­ise that data?

• Sec­ond, do you have appro­pri­ate gov­er­nance in place? If you’re build­ing an AI mod­el that will make rec­om­men­da­tions about insur­ance cov­er, how do you prove that the mod­el is trained and using data reli­ably?

• Third, it’s impor­tant to have the right peo­ple with the right tools. To build knowl­edge, make it eas­i­er for data sci­en­tists and data teams to share knowl­edge and col­lab­o­rate.

• Final­ly, ensure you have the right plat­form to process data effi­cient­ly. Many projects fail because of data bot­tle­necks or because your net­work and devices can­not process data at scale.

Accord­ing to Coomer, many AI sys­tems drown in data as soon as they move out into the real world because tra­di­tion­al approach­es can’t man­age the huge vol­umes need­ed to feed large-scale AI mod­els. “AI sys­tems need to ingest data at incred­i­bly high rates, and you need to make that data acces­si­ble and secure to enable the kind of deep learn­ing need­ed to make crit­i­cal deci­sions.”

When it comes to the fourth require­ment, Coomer advis­es com­pa­nies to think care­ful­ly about the capac­i­ty of their stor­age sys­tems. “In our expe­ri­ence, peo­ple think about AI projects in terms of GPU and com­put­ing require­ments, but don’t con­sid­er the data inges­tion, and the vol­ume of data that will be involved once an AI mod­el goes into pro­duc­tion,” says Coomer. “You could have the best data sci­en­tists and mod­els, but if your stor­age can’t ingest the 250 petabytes of data that’s sup­port­ing your AI mod­el, then your project will fail.”

Busi­ness­es should pre­pare for much larg­er data require­ments by adopt­ing AI-opti­mised sys­tems and process­es. These are sys­tems devel­oped with AI in mind, so that adjust­ments are made at every stage of the data jour­ney to improve the speed and effi­cien­cy of data inges­tion and trans­fer. In its State of AI in the Enter­prise report, Deloitte found that the top pri­or­i­ty for IT lead­ers invest­ing in AI was mod­ernising the data infra­struc­ture for AI.

These days, AI does not nec­es­sar­i­ly need to be the com­plex, painful learn­ing curve that ear­ly adopters have trav­elled, says Coomer. Com­pa­nies embark­ing on AI projects today have access to an increas­ing­ly large num­ber of AI-opti­mised tools and plat­forms that make it eas­i­er and faster to deploy AI at scale. Deloitte found that 80% of tech­nol­o­gy lead­ers are buy­ing at least as much AI tech­nol­o­gy as they build, and the longer a com­pa­ny has used AI, the more like­ly they are to buy com­mer­cial plat­forms and tools.

To achieve quick wins from AI, organ­i­sa­tions must invest in all four pil­lars of AI – data, peo­ple, process and plat­form. “They’re all impor­tant, and what we focus on at DDN is solv­ing the plat­form part of the issue, whether you need a twen­ti­eth of a sin­gle rack or some­thing that can han­dle petabytes of data with­out increas­ing the com­plex­i­ty,” says Coomer. “The key thing to remem­ber is not to fall into the trap of doing what you’ve always done, and rather start with some­thing that’s built to solve the prob­lem in front of you.”

---

Q&A: Unlocking the promise of AI

Alex Bouzari, CEO of DDN, talks about how enter­pris­es can get their AI projects right

Who is DDN, and how do you help organ­i­sa­tions on their AI jour­ney?

DDN is a mar­ket-lead­ing glob­al organ­i­sa­tion which pro­vides pow­er­ful data stor­age solu­tions, and facil­i­tates the AI- and data-enabled dig­i­tal trans­for­ma­tion jour­ney for enter­pris­es and research facil­i­ties. With over 10,000 cus­tomers and thou­sands of deploy­ments world­wide, DDN is mak­ing it eas­i­er, safer and faster to imple­ment AI and reap mas­sive rewards in health­care, autonomous dri­ving, finance, retail and many oth­er sec­tors.

We de-risk the cus­tomer jour­ney to a suc­cess­ful AI trans­for­ma­tion: AI sys­tems need enor­mous amounts of data to unlock mean­ing­ful busi­ness val­ue: at DDN, we accel­er­ate their dig­i­tal trans­for­ma­tion jour­ney, by deliv­er­ing the fastest path to imple­men­ta­tion of AI, help­ing them move seam­less­ly from pro­to­type to pro­duc­tion. This means that organ­i­sa­tions can focus on busi­ness val­ue, instead of man­ag­ing tech­ni­cal com­plex­i­ty and hav­ing to make com­pro­mis­es on data and scale.

Many busi­ness lead­ers see AI as a daunt­ing, com­pli­cat­ed under­tak­ing. Is that a fair assess­ment, and how are things chang­ing?

I think when any new tech­nol­o­gy comes along that has the poten­tial to trans­form how we do some­thing, peo­ple get baf­fled. The mar­ket is matur­ing so rapid­ly and there are now so many more organ­i­sa­tions who can help you to imple­ment AI strate­gies with­out delay or com­plex­i­ty.

The impor­tant thing is to inter­view and select ven­dors and part­ners care­ful­ly. Have they done this before? Do they know your mar­ket? If you get the right peo­ple and they have the right expe­ri­ence, you will get suc­cess.

What are the key steps a CIO must start with to unlock busi­ness val­ue and ben­e­fits from AI?

I firm­ly believe to suc­ceed with AI you need the right infra­struc­ture, and that means hav­ing plat­forms that can deal with the mas­sive amount of data and then help you to find the vir­tu­al nee­dle in that haystack.

Peo­ple wast­ed a lot of time try­ing to get their exist­ing infra­struc­ture to sup­port AI, but you need new data repos­i­to­ries that are opti­mised for AI, and they need to be inte­grat­ed with the busi­ness. That inte­gra­tion process has matured sig­nif­i­cant­ly in recent years, and there are now many more bridges between AI plat­forms.

What’s the most excit­ing appli­ca­tion of AI you’ve seen recent­ly?

We recent­ly helped build a num­ber of AI sys­tems to sup­port crit­i­cal phar­ma­ceu­ti­cal and life sci­ences research. This is an excit­ing growth area, where com­pa­nies are using AI super­com­put­ers to dis­cov­er new types of ther­a­pies and to improve the design of anti­bod­ies to help treat inflam­ma­to­ry dis­eases such as asth­ma and COPD. Dig­i­tal biol­o­gy is attract­ing both estab­lished organ­i­sa­tions and emerg­ing star­tups because it changes the eco­nom­ics of drug dis­cov­ery.

What would you say is the biggest mis­take com­pa­nies are mak­ing with AI right now?

I think it’s try­ing to use exist­ing tech­nol­o­gy for new pur­pos­es. A lot of tech­nol­o­gy exists with the pur­pose of keep­ing the lights on, man­ag­ing the data­base, and tak­ing care of oper­a­tions. It’s not often seen as an enabler. Now things have changed, because AI is a busi­ness enabler, and the vol­ume and type of data required to sup­port that enable­ment is com­plete­ly dif­fer­ent to what we’ve seen before. It’s not that it’s much hard­er, it just needs dif­fer­ent tools and plat­forms. Don’t take my word for it. Do the research. Go and see the suc­cess­ful imple­men­ta­tions, see who has deliv­ered, and ask them how they achieved those bet­ter out­comes.

Over the next five years, what’s the biggest change we’ll see in how we approach AI projects?

I think with­in five years we will start to see cus­tomised imple­men­ta­tion of AI tools and plat­forms because organ­i­sa­tions will want AI that meets their spe­cif­ic needs. Beyond that, I’m excit­ed to see AI used in intel­li­gent homes. I don’t just mean auto­mat­ed smart lights or music, but things like med­i­cine built into intel­li­gent dwellings to enable us to live health­i­er and more active lives.

To find out more, vis­it ddn.com/ai-for-business

---

Pro­mot­ed by DDN