Buy AI or Build It? The 2026 Answer for Logistics Leaders

AI coding tools have made building look easy. Getting AI into production in logistics is still hard. Here is how to decide where to build, where to buy, and where to partner.

When this article first ran in December 2025, the answer was clear: in logistics, buying AI beat building it almost every time. Ten months on, the question has shifted.

AI models are improving every few months, and AI coding agents mean a small team can now prototype in days what used to take a quarter. McKinsey's State of AI in 2026 survey, published in August, found that 32% of organisations have decided against buying at least one software product or feature because they could build it in-house with AI coding tools. Among AI high performers, it is nearly half.

Yet the results have not moved. In the same survey, only 37% of organisations report any EBIT impact from AI, unchanged from last year, and just 6% qualify as high performers. In logistics the gap is wider still. BCG's 2026 AI in Logistics Executive Survey found that 97% of logistics executives rank AI as a strategic priority, but only 13% say it is delivering measurable financial impact.

The real problem: building is easier than ever, and value is as hard to capture as ever. The question for 2026 is not "build or buy?" It is "who gets this into production, and who keeps it running?"

What Has Changed Since 2025

Four shifts have redefined the decision since the original article ran.

• Prototypes are cheap. AI coding agents have collapsed the cost of a first version. A working demo is no longer evidence that a workflow is ready for production.
• Running AI is now a cost line. One in five organisations told McKinsey that AI operating costs, including tokens, have constrained their use. Building is one bill. Running it reliably every day is another.
• Large companies are pulling ahead on agents. 40% of organisations with revenue above $1 billion are scaling AI agents, up from 27% a year ago. Among smaller organisations the figure is flat at 22%.
• Logistics customers are starting to ask. BCG and Alpega found more than 40% of shippers now consider a logistics provider's AI capabilities when choosing partners, and around 60% of logistics providers say integrating AI into their existing systems is their main investment priority for the next one to two years.

The evidence that still holds

MIT's GenAI Divide research (2025) analysed 300+ publicly disclosed AI initiatives. It found that buying from specialised vendors and building partnerships succeeded about 67% of the time, while internal builds succeeded about a third as often.

That research predates the latest wave of coding agents, but its explanation matters more now, not less. Projects did not fail because the models were weak. They failed on integration, and because tools did not learn and adapt to the way the business actually works. Faster prototyping does not fix either of those.

Why the Gap Between Demo and Production Is Wider in Logistics

Logistics amplifies every challenge that sits between a promising prototype and a workflow that runs every day.

Your systems were never designed to talk to each other

CargoWise, TMS, WMS, ERP, customs platforms, carrier portals, shared inboxes, PDFs, spreadsheets and WhatsApp all sit in the same workflow. BCG points to siloed legacy systems as one of the main reasons logistics AI has not paid back. A prototype that works on clean sample data rarely survives contact with a real inbox.

The work is mostly exceptions

Every customer has its own surcharge rules, routing guides and document formats. Overweights, hazmat, mismatched invoices and missing commodity codes are where operational risk lives. Production AI has to recognise uncertainty and route it to the right person, not guess.

The rules keep moving

Duty rules, customs requirements, carrier surcharges and customer formats change constantly. Someone has to own monitoring, testing and updates after go-live, every week, not just at launch.

Enterprise-grade means governed

Customers and auditors expect controls, audit trails, data security and clear accountability. McKinsey found high performers are much more likely to actively manage AI risks. That is work an internal side project rarely budgets for.

Cheap models can be expensive

In our experience, teams choosing cheaper models to save money often spend far more fixing their mistakes. In logistics operations, accuracy pays for itself.

The real failure mode is rarely the model. It is the absence of operational design around the model.

Building Works, If You Are C.H. Robinson

The strongest case for building in logistics is C.H. Robinson. It owns its application layer and has used AI to automate manual processes across workflows. The company reports productivity gains of more than 60% since the end of 2022, and in Q2 2026 it grew adjusted operating income by 20% year on year, while the Cass Freight Shipment Index fell year on year for the 15th consecutive quarter.

That is real ROI. It is also the product of a multi-year transformation with a large in-house engineering and data science organisation behind it. Most freight forwarders, 3PLs, carriers and shippers cannot, and should not try to, replicate that. They need the same outcome without building the same organisation.

Is Checking Faster Than Doing? A Simple Test

At the Leaders in Logistics conference in London in October 2026, Ripple CEO Adrian Smith shared a practical filter drawn from McKinsey's guidance on the economics of agentic AI: divide the time it takes a person to check the work by the time it takes to do it. If the AI succeeds more often than that ratio, the workflow is worth automating.

If a task takes an hour and checking it takes six minutes, the AI only needs to be right more than 10% of the time to save effort. Many core logistics workflows pass easily because they are slow to prepare and quick to check.

Workflows that typically pass the test

• Quoting and rate requests
• Customs entries and HTS classification
• Vendor onboarding and compliance checks
• Audits and incident reporting

Once a workflow is running, judge it on outcomes, not cost per token. The fully loaded cost includes human review, exceptions and rework, so measure what you already measure: touches, turnaround, cost per entry and quality.

Six Decisions That Separate the 6%

Based on the McKinsey Live webinar on the economics of agentic AI and Ripple's experience working with logistics operators, here are six decisions that distinguish companies getting real returns from AI from those still waiting.

1. Pick two or three lighthouses. Choose workflows that matter to the business and rebuild them with AI, rather than spreading chatbots across everyone.
2. Redesign, do not bolt on. Nearly three-quarters of high performers have fundamentally redesigned workflows around AI, compared with about a quarter of everyone else. AI runs the standard work; your people supervise and own the exceptions.
3. Own your data. Process documentation, emails and operational history become one of your most valuable assets as AI joins your processes.
4. Make AI spend visible. Know which workflow, agent and model is driving cost.
5. Engineer the economics. Match the model to the task and put limits on loops, rather than defaulting to the most advanced model everywhere.
6. Manage AI on returns. The goal is not to cut AI spend. It is to know where that spend delivers a return, and back it.

High performers are also twice as likely to say their organisations have defined processes to measure the impact of AI initiatives.

Build or Buy: A Framework for 2026

The old choice was binary. In 2026 there are three realistic routes, and most logistics businesses will use more than one.

Build in-house

• Best fit: AI is core to how you compete, not just how you operate
• What it requires: Permanent AI engineering capability, logistics domain expertise, production support, monitoring, controls and ongoing maintenance

Buy off the shelf

• Best fit: The workflow is standard and the product fits it as it stands
• What it requires: Tolerance for vendor roadmap constraints and low configuration risk

Buy a specialist AI partner

• Best fit: The workflow is specific to how you operate and spans your existing systems, inboxes and documents
• What it requires: Fast production deployment, enterprise-grade controls and a vendor accountable for accuracy and outcomes

Building a prototype has never been cheaper. Running AI reliably inside a live logistics operation has not got any easier. That is where most of the 87% of logistics businesses not yet seeing financial impact are stuck, and where the right specialist makes the difference.

What Production Looks Like

Two live deployments show what happens when a workflow that passes the check-versus-do test runs end-to-end inside existing systems.

W.W. Rowland Trucking

A Houston-based drayage and LTL carrier with 10 terminals. Quoting lived in a shared inbox alongside work orders and PODs. Ripple now finds the RFQs, reads the requirements, applies each customer's surcharge rules and sends the quote, routing overweights, hazmat and residential deliveries to the team for review.

• Quote response time cut from hours to seconds
• 95% less team time on manual quoting
• More than 90% of quotes correct from go-live

Pentagon Freight Services

Around 140 UK customs declarations a day, each needing 30 minutes to two hours of manual preparation from PDFs, spreadsheets and scanned invoices. Ripple reads the documents, applies customs logic centrally and hands a ready entry into the existing customs system, while operators keep the checks and exceptions.

• Up to 1,400 hours a week removed from customs preparation
• Up to 35 FTEs of capacity redeployed to checks and compliance
• 60% lower clearance costs, with software costs 40% below the legacy system it replaced

These are not pilot wins. They are production outcomes, running inside the systems both businesses already use.

A Practical Plan for the Next 90 Days

If you are deciding where to start, use a narrow, operationally honest sequence.

1. Pick one workflow that is high-volume, slow to do and quick to check.
2. Baseline it using the KPIs you already report: response time, touches, cost per entry, error rate.
3. Run it inside your existing systems, with your people owning exceptions from day one.
4. Measure in weeks, not quarters, against that baseline.
5. Scale to two or three lighthouses once the numbers move.

This is where most AI programmes stall: they launch on enthusiasm and die on maintenance. The companies turning AI into financial results are the ones that treat production as the product.

Bottom Line

In 2026, anyone can build an AI prototype. Very few logistics businesses are turning AI into financial results. The companies closing that gap are not the ones with the most pilots. They pick the workflows that matter, redesign how the work gets done, and measure AI on outcomes.

Your competitors are not debating whether to build or buy. They are deciding how fast they can get to production, and who will keep it running when they do.

Frequently Asked Questions

Should logistics companies build AI or buy it?

Most logistics businesses should buy a specialist solution. Building only makes sense when AI is core to how you compete and you have permanent engineering, support and governance capability in-house.

What makes AI hard to productionise in logistics?

Logistics workflows span legacy systems, shared inboxes, PDFs and exception-heavy work. A prototype can look good and still fail in production if it cannot handle integration, changing rules, audit trails and operational controls.

What is the check-versus-do test for AI automation?

Divide the time it takes to check AI output by the time it takes to do the task manually. If the AI succeeds more often than that ratio, the workflow is worth automating. Most core logistics tasks pass easily because they are slow to prepare and quick to check.

When should a logistics business buy a specialist AI platform?

When the workflow is specific to your operation, spans your existing systems and documents, and needs to go live quickly with enterprise-grade controls and a vendor accountable for accuracy and outcomes.

What are the six decisions that separate AI high performers in logistics?

Pick two or three lighthouse workflows; redesign around AI rather than bolting it on; own your data; make AI spend visible; match models to tasks; and manage AI on financial returns, not activity metrics.

About Ripple

Ripple designs, builds and runs continuously improving enterprise-grade AI solutions for logistics and supply chain operations.

Our custom AI automates time-consuming work across emails, documents and existing systems, helping teams handle more work, reduce costs and serve customers faster.

Built around each customer's specific business requirements, our AI agents handle repetitive tasks, apply business rules and route exceptions requiring human judgement. We engineer the integrations, controls, monitoring, compliance and security needed to make AI reliable in production.

Our proprietary architecture is LLM-agnostic, allowing us to use the right AI models for each task without tying customers to a single provider.

Our solutions cover customs and trade operations, operational compliance and risk management, and quote-to-cash and workforce processes.

We can also replace legacy software that is no longer fit for purpose or is too costly. Under our legacy replacement offer, we build the replacement without an upfront development fee and commit to running it for less, subject to agreed scope and terms.

Ripple is live in production and active deployment with logistics businesses including JAS Worldwide, Pentagon Freight Services, W.W. Rowland Trucking, Bergen Logistics, Novex, Polo 4PL and RAM International Shipping. Results achieved across deployments include 10x throughput, 95% less manual effort, 99%+ accuracy and up to 75% lower technology costs.

Book a call with Ripple or contact jonny@useripple.io.

Sources

• McKinsey, The state of AI in 2026: On the road to ROI (25 August 2026)
• McKinsey Live, Improving the economics of agentic AI: A guide for business leaders (2026)
• BCG, Why AI isn't delivering ROI in logistics (2026 AI in Logistics Executive Survey)
• BCG and Alpega, AI Is Already Moving the Logistics Industry Forward (March 2026)
• MIT NANDA, The GenAI Divide: State of AI in Business 2025, via Fortune (August 2025)
• C.H. Robinson, Q2 2026 earnings summary (July 2026)
• Ripple, W.W. Rowland Trucking customer story
• Ripple, Pentagon Freight Services customer story‍