Fulfillment Re-Imagined - Part 2

Previously in the Future of Retail series, we looked at the common business models that define supply chain and logistics today.  Based on those models, we looked at the  functional design and common failures associated with today’s ecommerce.  In this installment, we will start exploring the future as human involvement recedes.

Transition from Human Driven Transaction to Autonomous Fulfillment

We are already witnessing the beginnings of a major transformation in retail.  While the recent explosive impact of generative AI and the availability of AI tech has been big news to many, AI’s influence on retail is far from new.  The emergence of AI has happened in parallel to the evolution of ecommerce and been pivotal.  

Simple Intro to AI

Without digging too deeply into the science of AI, an overview is helpful to understand how this impact will continue to reshape retail as we know it.  AI can be broken down into capabilities and functionalities.  The capabilities are: 

Narrow AI (ANI) - ANI encompasses all of the AI in use today.  Narrow AI solutions are for specific tasks and the solution’s is limited to addressing these tasks.  These solutions require human training and while known as ‘weak’ AI, solutions can be quite powerful.  

General AI (AGI) - AGI is still theoretical but there is active research which may lead to real solutions.  Solutions in this area would attempt to deliver human-level intelligence.

Super AI (ASI) - ASI, or artificial superintelligence,  is also theoretic and the stuff science fiction captures.  It’s a “hypothetical form of AI that surpasses human intelligence across all fields, from creative arts to scientific research” (per USC).  

Use of AI in Retail Today

There is already a near-unlimited list of existing AI solutions powering retail today.  Some of the categories and examples are: 

Classification

The complexity of digital catalogs has outpaced the ability to leverage humans to support all the associated activities.  Further, the accuracy of humans and vulnerabilities of companies which fail to accurately maintain product details, has advanced the use of AI classification.

Amazon was one of the companies that pioneered the use of AI to classify Hazmat and automate classification for product shipping into international markets.  In both cases, AI classification accelerated the availability of products: (1) enabling access and fulfillment of products  with lithium ion batteries,  and (2) identifying international demand for products and dictating Amazon’s international marketplace launch plan.  

There are numerous other non-Amazon examples where classification has improved various retailer’s catalog, enforced consistency operationally across diverse product skus, and enabled faster product listings.  The emergence of regulatory restrictions for data (such PHI data for HIPPA)  and product usage (such as country of origin) have further driven the need for speed and accuracy making this an ongoing area of focus.

Forecasting

AI has been used in forecasting for decades.  It is involved in demand forecasting, inventory modeling and positioning, and network modeling to name a few.  McKinsey Group has estimated that applying AI forecasting for supply chain management can result in a 20-50% defect reduction.  At this point, all major retailers leverage it for varying forecasting purposes and the availability of solutions is reaching a wider range of retailers.

Optimization

Transportation and warehousing are heavily improved by a variety of AI solutions today.  Warehousing AI solutions address labor planning, inventory positioning, picking and fulfillment optimization.  Numerous solutions are available for translating warehouse forecasts into labor plans by optimizing based on role, task and schedule.  Early innovators like Kiva used machine learning and other AI approaches to dynamically position warehouse inventory to achieve optimal pick paths more than a decade ago. 

Similarly, transportation optimization is in use throughout retail logistics today.  Its commonly used for route planning and optimization and carrier selection.  A variety of enterprise solutions are available today for fleet management, last mile optimization and selecting the optimal carrier at a package level.

Personalization

Personalization has evolved with the digitization of retail and it should be no surprise that AI has been a vital part of this journey.  One of the early focuses was how to use customer insights to influence purchasing.  Collaborative Filtering was introduced in 2003 by Amazon.  This solution has been a primary driver for recommendations and reset early assumptions about how to accurately predict customer demand. Rather than focusing on similarity between customers, it focuses on product insights.  By looking at purchase histories across customers interested in same skus, it builds predictions based on the frequency of recurring items in historical data.  This proved to be far more accurate than trying to leverage direct customer insights.  

AI-driven approaches continue to modernize retail as retailers have seen the power of personalization to drive convergence.  Over 70% of US ecommerce retailers believe that AI-driven personalization and generative AI will affect their business (based on a Bolt survey in 2023).  Therefore, more money and effort continues to drive AI advances in personalization.  Beyond collaborative filtering, other solutions add more traditional AI approaches like neural networks and account for variation like time period such as time to increase the likelihood of recent items.  

Marketing

AI has already redefined marketing over the last decade and continues to outpace other applications of AI.  The aforementioned personalization has been one component heavily involved in digital marketing (I’ve separated it as its become an integral part of most retail stacks for more than marketing and therefore stands on its own).  However, there are many other aspects of marketing that leverage AI to connect with consumers.

Customer segmentation and grouping are fundamental  to connecting with pools of  “similar” consumers. Affinity marketing has origins that date back to the 1980s.  However, affinity profiling which uses deep insights and machine learning to identify customer segmentation, has been at the core of digital marketing since ecommerce began.  Obviously the wealth of data and multitude of observations of populations across a number of sources can create more accurate audience campaigns with very precise attribution.  This has spun out into varying types of audiences depending on data sources, algorithm and purpose. Two popular groupings are frequently used:  (1) affinity audiences that Google Ad uses which are built using a fairly stable set of observations (visited websites, interests, location and app downloads) and (2) interest-based audiences (interests, hobbies, activities, etc) used by Meta. However, the precision and approaches continue to evolve.  Additional precision for segmentation is being delivered using natural language prompts and generative AI, for example.

Another common AI usage in marketing is for tailoring and refining digital campaigns and even non-marketing digital treatments.  Companies like Salesforce and Amazon have different approaches.  Salesforce offers tools to create, test and optimize campaign variants and use predictive AI to improve success.  Weblab was a very early innovation at Amazon.  It was used to run concurrent treatments or “experiments” to determine the more successful approach and is now offered via AWS.  

Lead scoring is yet another area where AI has been instrumental in modernizing marketing.  By analyzing customer data, scoring is identified which can predict which leads are most likely to convert into sales.  There are a plethora of 3P and inhouse approaches that nearly every retailer utilizes today.

Audit

Shrinkage and anomaly detection is something AI does well.  By looking at existing data, AI can identify and predict trends that lead to poor behaviors (overstocking or understocking) and suggest operational challenges (potential theft based on the loss rates), vendor fraud (products never delivered) to name a few.  There are numerous commercial solutions available and nearly every major retailer relies on some form of AI for audit purposes.

Support

Automated chat is a commonly used solution powered by AI.  Nearly 37% of ecommerce global marketers  are already using AI to power CS and support  (February 2024, Klaviyo AI Trends Report).  The explosion of generative AI is accelerating retail usage adding natural language interfaces and AI-powered virtual assistants.

How AI Will Reinvent Retail

So AI is already commonplace in retail today, why will it change retail dramatically and why now?  These seemingly reasonable questions assume that AI investment is just starting.  As we saw, this is not the case… the truth is that AI is already reinventing retail, it will just accelerate.  The reason for this acceleration is:

1. Increased investment in AI
2. Availability of AI and sophisticated technology
3. Business challenges with the existing model

The first two should be apparent to all.  The rapid availability of technology and AI has followed the emergence of cloud computing and large technology investments over the last two decades. These changes have been coupled with the reduction in hardware costs and improvements in digital access and bandwidth globally.   Without going too deep into the business challenges, the rapid rise of ecommerce has exposed large flaws in the current retail model:

1. Rising consumer expectations (speed, accuracy, price)
2. Slim retail margins 
3. Explosive growth of delivery 
4. Excessive waste (packaging and fuel)

Unchecked, these are resulting in another extinction event for many retailers and carriers that can’t keep pace.  

Automated Buying

So focusing on AI’s impact on purchasing, what changes and why would this happen quickly?  A quick look back on the impact of AI on marketing demonstrates how quickly AI reinvented that space. This was at a time when much of the technology wasn’t readily available.  Therefore, it's safe to assume AI-driven purchasing and fulfillment are likely to happen rapidly given the availability of data, technology and AI solutions.  

Current State:  Retailer Readiness

eCommerce today is a generally series of specific activities which could easily be automated via a workflow management system However, these activities are generally implemented through disparate systems and even by different companies for most of today’s retailers.  Most retailers are ill prepared for the emergence of AI-driven ordering and systemic changes which will be fast follows.

With the expansion of AI, retail will see the rise of anonymous purchasing solutions (LINK ).  These solutions will replace or vastly reduce traditional buying.  The entire retail selling and fulfillment process will be reshaped dramatically.  These changes will include transformational changes in the structure of ecommerce, emergence of data and AI authorities, and new protocols for managing the interactions associated with buying. 

Future State:  Retailer Tech Requirements

The transition to AI-driven purchasing will be dramatic and the retail technology requirements will be dramatically different from the current state.  As captured in previous installments, ecommerce tech currently manages workflows.  In the future, we will see it shift towards a state more closely resembling automated bidding and auctions like a stock trading platform.  One benefit of this purchasing shift will enable more economically viable fulfillment.  In fact, we are likely to see the transformation of consumer fulfillment more closely resembling B2B delivery:  greater consolidation, fewer trips and greater efficiencies throughout the fulfillment journey.

Major Components of Automated Buying

We discussed the fundamentals of anonymous purchasing solutions (APS) previously.  We will now translate these into a possible design which enables its implementation. 

There are four key components will power the continuous automation of retail: 

• Customer Domain - there will be a full understanding of the customer used to inform and automate decisions.  This will have a deep understanding of every aspect of the areas which shape a customer’s retail behavior: existing possessions, activities, interests, career, finances and commerce decisions.  

• Drivers - temporal or events that influence decisions will be vital to making accurate decisions.  Among these will be time and schedule, major events, social influences and personal events or situations.

• Observers - this is one of two independent groups of consumer authorities.  Access and availability of this data would be driven by a variety of forces from government to market. Likely mapping tools like Google Maps and other tools not purposely advertising their collection intents would shape these sources.  This data would shape an understanding of a consumer's existing preferences, financial performance, education and travel patterns.

• Reporters - this is the second group of consumer authorities.   This would include existing authorities like credit agencies and background services to aggregators of data from various public and private data sources. 

Alchemy of a Purchase

The automation of retail will leverage these four components to invoke a continuous, concurrent network of automated purchasing.   The actual tools that power the individual consumer purchasing may be diverse, segmented auto-shoppers separated by need (like existing app stores) or large homogenous solutions (like marketplace automations) may dominate.  Overtime, its likely that the shopping capabilities will become extensions of existing platforms such as a consumer’s home, vehicle or a location based platform (like a common purchase marquee).  Regardless of implementation, the underlying goals will be the same:  capture a purchase intent for a consumer based on the knowledge of what the consumer wants and needs given a set of drivers, reporters and observers.  

For example, a consumer consumes three 12-packs of diet coke during the summer and always pays for it immediately except when traveling to his parents home in the second week of August.  The automated purchasing solution (APS) would understand these constraints, look for the lowest cost that keeps the consumer supplied according to the demand pattern.  

In the previous example, the general domain is well understood but there are likely to be many ways it could be achieved.  However, there is likely to be differences between those ways and the automated purchasing solution must evaluate all options and choose the ‘best’ way… perhaps multiple times.  So how would this be accomplished?

Scope of an Automated Purchasing Solution

APS will own capturing the consumer insights that shape the purchase.  These need to weighed or adjusted by drivers.  The APS will also need to seek out insights from relevant reporters and observers.  This will ensure greater accuracy and success for the transaction.  Once this information is combined, it results in the basis for an offer.

Customer Intent Universe

The combination of information garnered from the four components to create an offer occurs in the customer internet universe.  This is loosely a means to convert the information into a formalized request or an offer.   The Customer Intent Universe dictates protocols which enable automated purchase agents to find ways to actualize the consumer request.  

This would be one of the first major changes between today’s retail technology stack.  Rather than surface a website, retailers would need to be ready to respond to automated purchase agents.   There will likely be numerous implementations of purchasing platforms until dominant winners emerge.  Winners would likely occur based on the participation of retailers responding to authorized purchase agents on a given platform.  Ultimately, this will determine how many concurrent platforms an APS surfaces requests to.

However, even in a situation with a single purchasing platform, an APS is likely to try multiple requests for the same purchase.  As mentioned, there are a number of ways most purchases could be framed (time, cost, item type) and each could impact the final purchasing decision.

Actualized Offers

The automated purchasing agent (APA) has two activities:  (1) translate the various inputs into an actualized offer, and (2) respond with qualified bids.  We will discuss the second part after reviewing the automation of fulfillment.

As mentioned, as automated shopping matures, there are likely to become hardened interfaces and structures for an offer.  It will be the APA’s responsibility for submitting the actualized offer to a negotiation platform and managing the collection of responses.  

The filtering of qualified offers and resulting action will be encapsulated in the APA.

Automated Fulfillment Solution

Fulfillment will become the flipside of each automated purchase.  Unlike today’s retail world where offers are fairly static and encapsulate each step of fulfillment, fulfillment bids against an offer will become real time calculations.   These fulfillment bids will capture 

Offer Candidate

Fulfillment systems will need to review a multitude of parallel offer candidates that arrive in realtime and respond within a prescribed time window driven by automated purchasing agents.  In their most simple form, an offer candidate will include:

• The desired items - these may be as specific as today’s sku (i.e. brand) or as general as a loose item description
• Time - the time its needed, how long the offer is valid, etc
• Cost - what is the desired price range or pricing constraints

These will be converted into a formalized request by an automated offer agent (AOA). An AOA  tenders the request to fulfillers via a Fulfillment Interchange and collects qualified bids to return back to the requesting automated purchasing agent (APA) via the Negotiation Interchange.  

Ultimately the Automated Offer Agent and the Fulfillment Interchange manage any inconsistencies across fulfillers and qualify bids.  This will be critical based on the diversity of entities likely to participate as fulfillers - especially during the early transition.

Transformation of the Retail Stack

So if today’s retailer is managing workflows, tomorrow’s retailer will be managing concurrent bids.  This transition will require an overhaul of the way the existing foundation pieces interact:

• Inventory:  will need to respond to signals that it collects based on offers.  It's likely to continue to have independent forecasts and will be receiving updates from Procurement and Fulfillment systems to actualize realtime placement and availability.  
• Fulfillment: these systems will need to be able to respond in real time regarding inventory position, fulfillment constraints including timing, availability and cost across inhouse and external service providers and likely provide a menu of options (cost for fastest / slowest, cost for bulk, etc.).  It will also have to be able to provide real time updates to inventory systems, procurement and financial systems.  It will be the heart of business success.
• Finance:  will move into a dominant and dynamic role in modern retail.  The reliance on buffer pricing decisions will result in lost opportunities as competition scales with more immediacy.  Finance systems will take on real time pricing decisions, weigh fulfillment costs, procurement cost risk and questions typically decided through human decisions currently.

Benefits:  while there will be a great deal of information exchanged in real time, fulfillers will see more predictability and ability to consolidate.  This will enable larger loads and better transportation costs.  Similarly, the discipline of automation will force more predictable labor patterns.  Finally tighter inventory management will allow better purchasing discipline and the ability to move away from unnecessary overbuying.

Challenges:  few retailers can make these changes alone.  The emergence of standards and capable solutions will be required.  There will be a period of turmoil as dominant solutions emerge.  This will also result in potential disruption for companies that made the wrong solution decisions.

Ultimately, this is one vision of the future but one based on similar disciplines and likely to be close.  It will be an exciting time for consumers and those companies ready to make the transition.