CTOs turn to ‘lean’ AI to overcome implementation challenges

It’s amaz­ing the sway that two let­ters can have. Investors’ ears prick up. Rivals fear oblit­er­a­tion. Share­hold­ers smell progress. But for all of the out­sized val­ue that a sim­ple men­tion of AI can bring to the board­room, there are only a few com­pa­nies real­ly able to max­imise its com­mer­cial poten­tial.

This is because apply­ing AI effec­tive­ly requires a large amount of infra­struc­ture, both cul­tur­al and mate­r­i­al. For advances in machine learn­ing (ML) and oth­er fron­tier fields to be tru­ly inte­grat­ed into a busi­ness, there needs to be a con­sid­er­able quan­ti­ty of data, a clear process in place and skilled prac­ti­tion­ers who can put these to good use. Many busi­ness­es, even larg­er ones, do not have those ingre­di­ents to hand. For oth­er, small­er enter­pris­es, there sim­ply isn’t the time or mon­ey to cre­ate an effec­tive inter­nal AI oper­a­tion.

As a result, spe­cial­ist com­pa­nies, often star­tups, are offer­ing a lean­er, out­sourced AI oper­a­tion as a ser­vice. In the process, these spe­cial­ist com­pa­nies are expand­ing the applic­a­bil­i­ty and pow­er of AI in all of our lives. The more that busi­ness­es are able to tap into these tech­nolo­gies, the more cus­tomers they’ll reach.

“By out­sourc­ing AI to high­ly spe­cialised tech com­pa­nies, not only indi­vid­ual com­pa­nies but also the entire indus­try can ben­e­fit as the AI mod­els can reach some lev­el of gen­er­al­i­sa­tion,” says Ramakr­ish­na Nan­junda­iah, co-founder of Phan­tas­ma Labs, which pro­vides AI-based mod­els for auto­mo­tive com­pa­nies and smart fac­to­ries. “Gen­er­al­i­sa­tion is hard to achieve if indus­tries do not share insights on the range of use cas­es and the intend­ed val­ue expect­ed from the AI,” he explains.

By work­ing across indus­tries, such spe­cial­ist com­pa­nies are more like­ly to make real advances in the field, poten­tial­ly to the ben­e­fit of all. By focus­ing exclu­sive­ly on AI tasks, these com­pa­nies are far bet­ter equipped to try out new tech­niques and approach­es, and become true experts at apply­ing break­through meth­ods to a vari­ety of real-world cas­es.

Outsourcing AI expertise

This might strike many CTOs as anath­e­ma to their modus operan­di. Aren’t they, after all, the har­bin­gers of inno­va­tion? Out­sourc­ing an AI project requires a lev­el of self-aware­ness and humil­i­ty, tap­ping into a risk-averse men­tal­i­ty and admit­ting that such quests may be beyond an enterprise’s core capa­bil­i­ties. Many busi­ness­es can­not jus­ti­fy expen­sive out­lays that result in fail­ure. And as a fron­tier tech­nol­o­gy, AI projects are far from guar­an­teed to suc­ceed.

A decid­ing fac­tor will often be the matu­ri­ty of the company’s exist­ing data cap­ture and man­age­ment sys­tems. The unthink­ing accu­mu­la­tion of data for data’s sake is insuf­fi­cient to fos­ter a com­pe­tent AI oper­a­tion. The appro­pri­ate data has to be cap­tured and then processed, tagged, stored and updat­ed reg­u­lar­ly in order for it to be use­ful.

“Before con­sid­er­ing AI adop­tion, it’s impor­tant to know where your data lies,” says Chris Royles, field CTO for Europe, the Mid­dle East and Africa at Cloud­era. “Here, hav­ing a robust data man­age­ment strat­e­gy is key. Since algo­rithms are essen­tial for machine-learn­ing process­es, mean­ing­ful results can only be achieved by lever­ag­ing high-qual­i­ty data.”

Data is impor­tant but it’s unlike­ly that an indi­vid­ual busi­ness start­ing an AI oper­a­tion from scratch will have all of the nec­es­sary data points. For­tu­nate­ly for many, own­ing pro­pri­etary data and pur­su­ing a big data strat­e­gy is far less of a mis­sion-crit­i­cal moat than it might have been a few years ago, when data was her­ald­ed as “the new oil”.

“Data becomes redun­dant over time,” explains Nan­junda­iah. “The Covid cri­sis has shown us that indus­tries can­not rely on past data alone for prepar­ing for future emer­gent sce­nar­ios. There has to be a dif­fer­ent way in which we can pre­pare indus­tries for Black Swan events.” 

Tech­niques such as rein­force­ment learn­ing are the next nat­ur­al evo­lu­tion of AI, he says, “where we can build val­ue-deliv­er­ing AI mod­els with­out the need for big data”.

Cutting costs

Anoth­er fac­tor moti­vat­ing CTOs to look beyond an inter­nal oper­a­tion will be cost. AI spe­cial­ists such as machine-learn­ing engi­neers are very well paid. Salaries above £100,000 are expect­ed but, unless com­bined with an effec­tive or well-informed team around said engi­neer, their tal­ents will often end up frus­trat­ed.

Indeed, hir­ing a cou­ple of experts inter­nal­ly may be insuf­fi­cient if there is a lack of doc­u­men­ta­tion, agreed-upon frame­works and tac­it knowl­edge already in place. An out­sourced team, how­ev­er, will be bet­ter able to tap into a gen­er­al suite of best prac­tice, expe­ri­ence and spe­cial­ties as well as being on top of all the impor­tant reg­u­la­to­ry and AI safe­ty require­ments.

Cloudera’s hybrid data cloud allows com­pa­nies to tap into AI- and ML-pow­ered data ser­vices, regard­less of their own data capa­bil­i­ties. Their advice for CTOs look­ing to adopt a lean­er, out­sourced approach to AI is to use the oppor­tu­ni­ty to explore these new tech­nolo­gies at their own pace – and real­ly appre­ci­ate the val­ue.

“Arti­fi­cial intel­li­gence can infuse your busi­ness with the infor­ma­tion and agili­ty need­ed to adapt to mar­ket changes and deliv­er inno­v­a­tive solu­tions,” says Royles. 

When out­sourc­ing, he says, it’s wise to play it safe by start­ing small and imple­ment­ing one use case at a time, “keep­ing in mind that the sequence you cre­ate may inform the use cas­es that fol­low. Then you keep build­ing on.”

A key con­sid­er­a­tion for any CTO fac­ing the ques­tion of whether or not to out­source an AI project is how ready they are inter­nal­ly to cul­ti­vate their own team and AI exper­tise. It’s an expen­sive and resource-heavy task that requires a real open­ness to fail­ure and exper­i­men­ta­tion as well as a clear sight of the nec­es­sary goals. At least to begin with, out­sourc­ing the first few projects might be the most effi­cient way to dis­cov­er how they want to inte­grate AI into the exist­ing tech­nol­o­gy stack.