Eranda Maldeniya is Director of Product, Options and AI at Enterprise Analytics, with 20+ years of expertise in enterprise programs.
Enterprise AI adoption typically begins with the improper query: “Which mannequin ought to we undertake?” A greater place to begin is, “Which enterprise course of are we attempting to enhance?” That is particularly vital for organizations that depend on advanced enterprise platforms similar to enterprise useful resource planning (ERP) programs.
I’ve seen organizations choose an AI mannequin first after which seek for use instances to suit it. The identical sample is now occurring with agentic AI, the place groups search for locations to deploy brokers earlier than defining the enterprise downside. The outcome could be unclear outcomes, wasted time and issue proving the worth of the funding.
The workflow is the true unit of worth.
Enterprise AI is usually evaluated on the activity stage. A mannequin might extract info from an bill, summarize upkeep directions or predict the probability of a machine breakdown on the store ground. These capabilities could be helpful, however such actions alone don’t assure a greater enterprise final result.
For instance, extracting bill knowledge is just one a part of the accounts payable (AP) course of. In apply, the tougher problem is the complete AP workflow. Invoices should be collected from the AP mailbox, separated from supporting paperwork, matched to buy orders, checked for taxes and expenses, assigned the proper fee phrases and routed for approval. If these steps stay handbook, higher extraction alone won’t create a lot worth for the finance group.
I’ve seen a number of AP automation tasks the place AI-enabled options had been deployed on high of ERP programs however targeted primarily on knowledge extraction. Though the extraction labored, the broader course of remained handbook, making it tough to exhibit significant worth to the finance group.
For this reason AI ought to be assessed by the operational affect it creates. Did it scale back knowledge entry time? Did it scale back the error charge? Did it resolve exceptions quicker? Did it enhance the standard or pace of a choice?
ERP programs could be the inspiration for AI.
AI can generate an perception, prediction or advice, however enterprise programs are the place that output turns into operational. ERP platforms typically maintain a lot of the enterprise context, controls and transaction historical past wanted to behave on that output.
If AI will not be related to the prevailing workflow, somebody nonetheless has to interpret the outcome and full the motion manually in one other system. The perception could also be helpful, however the course of stays fragmented.
Organizations typically have three choices: use AI already embedded within the ERP platform, join an exterior AI instrument or construct a customized resolution for a selected workflow. There isn’t any common best choice. The correct selection depends upon what suits the enterprise, what the answer can realistically ship, what it prices and the way will probably be used, not on which know-how sounds most superior.
Begin with the issue, not the know-how.
Each enterprise perform has operational challenges, similar to handbook knowledge entry, inaccurate information, delayed closing, poor reconciliation, restricted stock visibility or recurring gear failures. As soon as the issue is clearly outlined, the group can decide the correct resolution. In some instances, AI could also be acceptable. In others, typical automation, course of redesign or higher system integration might ship extra worth. This avoids forcing AI into issues which will have a less complicated resolution.
Contemplate a treasury group that manually enters foreign money charges into the ERP system every day. That is primarily an integration downside, not an AI downside. Then again, repeated machine breakdowns on the store ground might require a special method. If upkeep groups battle to establish early warning indicators, AI-based anomaly detection may assist assist preventive upkeep and scale back unplanned manufacturing delays.
When AI is the correct resolution, the aim ought to be to make use of the smallest efficient functionality. A specialised mannequin could also be higher suited than a big general-purpose mannequin for duties similar to doc extraction, classification or anomaly detection. It may be quicker, extra predictable and cheaper to function. Do not deploy essentially the most superior AI in the event you can resolve the issue reliably and measurably.
Measure the workflow earlier than and after.
A measurable AI rollout begins earlier than the know-how is chosen. The group first must outline how the workflow performs right this moment utilizing measures similar to cycle time, error charges, handbook effort or price per transaction. With out this baseline, it’s tough to show that AI has improved the method.
McKinsey’s five-layer framework helps this method by linking technical efficiency to consumer adoption, operational KPIs, strategic outcomes and monetary affect. A mannequin may go nicely technically, however it creates little worth if individuals don’t use it or if the workflow doesn’t enhance.
Every initiative ought to due to this fact have one clear enterprise final result. This might be a better straight-through processing charge, quicker bill registration, fewer planning exceptions or shorter contract assessment occasions. Technical measures similar to accuracy, response high quality and token utilization nonetheless matter, however they need to assist the enterprise final result reasonably than exchange it. In my expertise, a proof of worth is extra helpful than a proof of idea as a result of it exhibits not solely that the know-how works, however that the enterprise performs higher.
The Takeaway
Enterprise AI ought to start with a enterprise downside, a clearly understood workflow and an final result that may be measured. The correct resolution could also be embedded AI, related AI, a specialised mannequin or no AI in any respect. What issues is whether or not it suits the method and improves enterprise efficiency.
From what I’ve seen, one of the best outcomes don’t at all times come from essentially the most highly effective mannequin. They arrive from making use of the correct functionality to the correct workflow and proving that the enterprise performs higher.
Forbes Technology Council is an invitation-only group for world-class CIOs, CTOs and know-how executives. Do I qualify?
