The Most Valuable AI Work is Sometimes the Least Glamorous

TECH | January 1, 2026

The Most Valuable AI Work is Sometimes the Least Glamorous

I spend a lot of time with leadership teams who want “an AI strategy” that shows up in front of customers: smarter websites, clever chatbots, generative campaigns. Those things can matter and within MCD we've built quite a few of those experiences for clients. But if you look at where AI is actually moving the needle into 2026, most of the measurable ROI is being created far away from the homepage and inside operations, workflows, or the “boring” back-of-house machinery that actually runs the business.

The uncomfortable truth is most companies won’t win their category because they launched the flashiest AI-native experience. They’ll win because they quietly rewired how work gets done.

The hype vs. the value: AI in the wild

If you zoom out and look at the research, a pattern emerges:

McKinsey’s 2025 State of AI survey shows that while AI adoption continues to spread, the companies actually getting measurable value are the ones embedding AI deeply into operations and processes, not those doing one-off pilots or “innovation theater.”
Boston Consulting Group’s 2025 analysis found that only about 5% of companies are meaningfully benefiting from AI. Those “future-built” firms are seeing clear returns in cost reduction and productivity, while ~60% see little or no value despite big investments.
Gartner’s 2025 guidance to CMOs notes that, in practice, most AI planning is company-centric right now—focused on productivity and operational gains rather than front-of-house transformation.

We’re seeing the same pattern play out in our own client work. While the initial question is often, “How can we use AI to wow our customers?”, the real value lands when we help leadership understand how to redesign and automate the internal workflows that actually run the business. In other words, while the market may talk about AI as a consumer experience revolution, the value is showing up in very unsexy places like shared drives, ticket queues, spreadsheets, ERP systems, and customer operations.

Why back-of-house AI has disproportionate impact

Operational AI has a few structural advantages that consumer-facing AI often lacks.

1. Clear baselines and measurable outcomes

If you automate a finance reconciliation workflow, a claims process, or a content operations pipeline, you usually start with a known and somewhat deterministic process. For example:

Known volumes (tickets per month, files per week, approvals per campaign)
Known cycle times (how long it takes now)
Known cost per hour or FTE

When you introduce AI into that environment (say a set of agents that pre-classifies tickets, drafts responses, or flags exceptions) you can measure impact in simple, concrete ways. Now the number of tickets one person can handle in a day doubles, or the percentage of items that require a human to touch them at all drops from 100% to 40%. Those changes translate directly into hard numbers: hours freed, FTE capacity gained, backlog reduced, SLA compliance improved.

This is why back-office automation so often lands in the 20–40% productivity improvement range: not because back-of-house AI is special, but because the work lends itself to being counted, timed, and improved in visible increments. Those are the kinds of outcomes you can walk into a boardroom with and defend.

2. Less brand risk, more room to iterate

When you put generative AI directly in front of customers, the tolerance for hallucinations, bias, or inconsistency is essentially zero. It immediately triggers the need for governance, content controls, legal review, and strict guardrails. In practice, that means many customer-facing AI initiatives often stall out before they ever reach production. I’ve watched promising prototypes, pilots, and proofs of concept get blocked or abandoned more times than I can count. Moreover, there is no one to blame for these stalls when you consider a company’s reputation and brand trust are on the line. This creates and entirely different risk profile.

However, when you put AI inside internal workflows as a triage engine, a routing layer, a summarizer, or a drafting assistant you can build in natural human-in-the-loop checkpoints along the way.

AI proposes, humans approve.
AI drafts, humans edit.
AI routes, humans audit.

This allows your teams to reduce risk and lets you improve the system week by week instead of waiting for a “perfect” consumer launch.

3. Compounding gains across departments

Back-office efficiency is rarely confined to one team. Once we introduce AI into internal workflows, the impact tends to ripple across functions and compounds over time. A good example starts in support operations: AI that can summarize cases, suggests responses, and classifies reasons for contact, reduces handle times and improves resolution rates. That immediately benefits service, but it also creates a rich, structured signal for product and sales: recurring complaints, common objections, early churn indicators. What starts as a support productivity play becomes a feedback loop that can shape roadmaps and strategy.

We see a similar chain reaction with data hygiene. When AI is used to clean, normalize, and classify data at the point of entry, analytics teams spend less time fixing inputs and more time generating insight. Product and CX teams then base decisions on timely information instead of patchy reporting. The visible win might be “better dashboards,” but the real value is faster, more confident product and experience decisions across the organization.

Even something as mundane as internal knowledge management can create cross-department gains. When AI turns scattered policies, playbooks, and project docs into a conversational knowledge layer, new hires ramp faster, teams stop reinventing the wheel, and leaders spend less time answering basic questions. The net effect is that the same headcount can handle more work, at higher quality, because friction is reduced at dozens of small but critical points in the system. That is what I see as “compounding gains” in practice.

Why agencies are uniquely positioned to do this work

From an agency perspective, I think we are exceptionally suited to bring more value to client teams looking to boost their efficiency with AI.

While consulting firms and system integrators are already going after enterprise AI transformation, agencies have three differentiating advantages that sets us apart:

We understand the full customer journey At MCD, we don’t just work within one area of the organization, but instead are able to see into the campaign funnel, the service touchpoints, the content, and the people behind them. That makes it easier to spot the friction and the places where internal inefficiency quietly destroys customer experience.
We already integrate messy stacks CMS, CRM, CDP, e-commerce, analytics, loyalty platforms, custom APIs and the list goes on. MCD lives and builds in these worlds every day. Translating that into an AI-ready architecture (with decoupled services and shared data models) is a natural extension of what we already do.
We’re good at building things people actually use The biggest killer of internal AI tools is adoption. MCD's is used to designing interfaces, workflows, and content that people want to use, not just systems that “technically work.”

The agencies that lean into this will increasingly look less like “creative vendors” and more like product and operations partners with a deep AI layer.

Dogfooding the same AI fabric inside the agency

We’re not only doing this for clients.

While all this sounds great on paper, I'd be a hypocrite if it wasn't something that my team and I haven't also put into practice within the agency itself. We’ve been dogfooding the same AI and agent concepts in our own back-of-house operations and putting the ideas in this article into practice everyday.

Two areas where this has been particularly powerful:

Resource management and staffing
Conversational access to project data

AI agents for production and retrospective analysis.

Like every agency, we live and die by how well we allocate people to projects. Historically, that’s involved spreadsheets, capacity reports, and PM intuition. We have to coordinate between time-tracking platforms, sales tools, and active project plans while relying on manual reconciliation when someone is overbooked, underutilized, or on the wrong type of work.

To streamline these processes the team has launched an internal “production agent” that sits on top of our project scopes, SOWs, time-tracking tools, and employee skillset data to allow a straightforward view into the entire ecosystem for all relevant departments.

For example instead of digging across all these systems, our resource leads can ask:

“Who are the senior React/Next.js engineers with automotive experience who have availability in the next 3 weeks?”
“Given the current pipeline, where are we most at risk of overloading our design team next quarter?”
“Show me which projects have creeping scope but flat staffing in the last 6 weeks.”

All the while, behind the scenes, agents are:

Reading structured and unstructured data (SOWs, GitHub Projects, time logs, skills profiles).
Normalizing it into a shared model (roles, skills, timeframes, allocation).
Surfacing recommendations or flags.

This isn’t replacing human judgment; it’s removing the sludge work so humans can exercise that judgment more often and earlier. And because we’ve built these agents and data models around the way our agency actually runs, we’ve been able to take the same approach one level up and wire our project data into a conversational interface that our internal teams can use for retrospectives and gauging the health of a project. For example, instead of going to each department manually or building ad hoc reports, we can now ask:

“What’s the current budget burn vs. timeline for our major CMS rebuilds?”
“Show me projects in the last 18 months where we underestimated integration work by more than 20%.”
“List all projects where we used X framework, and summarize the outcomes in a paragraph.”

For leadership, this transforms AI from “something we sell” into a product we can improve on.

This entire layer is something we’ve deliberately built inhouse rather than buying an off-the-shelf tool and hoping AI plug-ins would close the gaps. We designed the agents, data model, and orchestration around how our agency actually works: the way we scope, staff, track, and deliver projects. That gives us a few important advantages over canned solutions.

We can swap in new models as they emerge without waiting for a vendor release, tune behaviors to the nuances of different clients and project types, and spin up new agents as other systems come online. In practice, it means the intelligence layer can evolve at the speed of the business instead of at the speed of a product roadmap.

Where this is going in 2026

If there’s a single thread running through all of this, it’s that AI becomes meaningful when it is anchored to how work actually gets done. The companies seeing real returns aren’t the ones with the flashiest chatbot on their homepage; they’re the ones quietly rewiring reconciliation flows, support queues, content pipelines, and staffing decisions so that humans spend more time on judgment and less on sludge.

That doesn’t mean the front-of-house work is a distraction. Far from it. At MCD, we’re doing a lot of work on the customer-facing side as well in areas like generative AI for product placement, AI-assisted creative to lower production costs, smarter journeys that adapt in real time based on context and behavior. I fully expect 2026 to be a year where confidence in these kinds of experiences rises as governance, controls, and patterns mature. But I also believe the brands that benefit most from those surface-level experiences will be the ones that have already done the work underneath. Clean data, owned services, and agent-driven workflows that make the customer-facing AI trustworthy and scalable is a winning strategy.

If you’re reading this and thinking, “We know there’s opportunity here, but we’re not sure whether to start with the customer experience or the back-of-house,” that’s the conversation I’m most interested in having. Somewhere in that intersection is a practical, defensible way to turn AI from a slide in your strategy deck into a fabric that quietly makes the whole business and the experiences your customers see—run better.