Where the AI signals are — and what they're not telling you

Over the past several months, our agentic pipeline has been continuously scanning vendor announcements, release notes, cloud provider blogs, and industry publications for signals of AI deployment across enterprise operations. We have now accumulated more than 1,700 tagged use case signals mapped to the 13 end-to-end domains of the Business Flows architecture.

When you plot them — by domain, by signal count, and by the effort and potential ratings the sources themselves assign — a picture emerges. It is not quite the picture vendors want you to see.

Signal count is a measure of attention, not readiness

The single largest domain in our dataset is Lead-to-Order (L2O) with 485 signals. Campaign management, lead scoring, opportunity tracking, AI-assisted quoting — the coverage is enormous. This reflects real market dynamics: sales AI is the most visible, the most funded, and the easiest to demo. A GenAI copilot that drafts a quote or summarises a customer call takes minutes to prototype and hours to present at a conference.

But 485 signals does not mean 485 deployments. It means 485 published claims. The CRM vendors, hyperscalers, and SaaS platforms compete ferociously to announce sales AI capabilities. Sales is customer-facing, measurable in pipeline metrics, and beloved by marketing teams. It generates blog posts.

Note the gap between the two logistics-adjacent domains: I2O (Inbound-to-Outbound, 48 signals) covers the actual warehouse floor — inbound goods, cross-docking, transportation management, inventory counting. It has one tenth the signal density of L2O. A scheduling algorithm running inside a WMS makes a worse press release than an AI that summarises a sales call. The numbers reflect that reality faithfully.

At the far end: Plan-to-Produce (Pl2P) has 21 signals and Forecast-to-Plan (F2Pl) has 34. Production scheduling, capacity optimisation, demand-driven MRP, S&OP integration with AI — these are domains where AI arguably has its deepest long-term impact on capital efficiency. They are barely represented. Not because nothing is happening, but because what is happening is harder to package into a press release and far harder to demonstrate without a real production environment.

The silence in Pl2P and F2Pl is itself a signal.

The "High potential" problem

Across our full dataset, the share of signals rated "High potential" by their sources ranges from 50% (Record-to-Report) to 82% (Quality Inspection). Almost every domain lands above 65%. This is not a coincidence — no vendor launches a capability and calls it "moderate potential." The rating inflation is structural.

What makes this interesting is where the inflation is most egregious, and where it is arguably earned. The chart below plots each domain by how often its sources claim "High effort" versus "High potential."

Lead-to-Order (L2O) scores 78% High potential at medium effort. The use cases are real — AI-assisted quoting, lead scoring, opportunity summarisation — but the bar for a "High potential" claim is low when the audience is a sales VP. Every CRM vendor has an incentive to frame their AI as transformational, and the pipeline metric makes it easy to attribute uplift.

Quality Inspection (I2Q) also scores 82% High potential — but only 4% of its signals are rated Low effort (compared to 36% in R2R). Vision-based defect detection, causal AI for root cause analysis, AI-assisted CAPA workflows — these are technically harder, operationally stickier, and when they work, they produce measurable quality yield improvements that compound. The potential rating is inflated by the same vendor optimism, but the underlying cases are closer to the floor of what enterprise AI actually needs to do.

Record-to-Report is the quiet winner

R2R stands out as the domain where signal-to-noise is most favourable. Only 7% of signals are rated High effort. 36% are rated Low effort. The use cases — automated period close, intercompany reconciliation, journal entry anomaly detection, AI-assisted financial consolidation — operate on structured, well-governed data that already exists in ERP systems. The business case is measurable in working days per close cycle. The integration path is known.

This is why R2R is where we see the most actual deployments behind the signals, rather than the most ambitious claims.

Plant maintenance and production planning: the underserved opportunity

The lowest signal counts belong to A2D (37 signals) — predictive maintenance, asset performance management, work order intelligence — and the two planning domains. This is structurally predictable. AI for predictive maintenance requires sensor data, historian integration, asset master data discipline, and tolerance for probabilistic outputs in safety-relevant contexts. It is not a GenAI story. It is a time-series modelling story that requires operational technology integration. Specialist vendors are doing real work here, but it does not generate the same volume of blog posts as an AI-powered sales copilot.

The same applies to F2Pl. Demand planning AI — probabilistic forecasting, supply constraint propagation, S&OP integration — is genuinely complex. The practitioners who work on it are not writing press releases. They are running backpropagation through hierarchical time-series models before a planning cycle. The signal count does not reflect the sophistication or the stakes.

What the charts actually tell you

Almost everything lands in a narrow band around "Medium effort, Medium-to-High potential." This is not because all enterprise AI has the same complexity profile. It is because the ratings come from sources with an incentive to neither oversell difficulty nor undersell ambition. The charts are a map of vendor positioning, not a map of deployment reality.

The genuinely differentiated insight from this data: the domains with the fewest signals and the highest rated effort are probably where enterprise AI will create the most durable competitive advantage — precisely because they are hard to replicate from a press release.

L2O is crowded. R2R is accessible. Pl2P and A2D are where the next wave of differentiation is building, quietly, in production environments that do not make the news.

Where does this leave you?

Most organisations entering an AI programme start where the signals are loudest — sales, customer service, finance automation. That is not wrong. R2R and P2P are accessible entry points with measurable payback. But if your competitive position depends on operational excellence in manufacturing, planning, or asset-intensive processes, the crowded domains are not where the differentiation is.

The hard question is not "which AI use cases exist?" — the 1,700 signals answer that. The hard question is which use cases are right for your process landscape, your data maturity, and your organisation's capacity to absorb change — and in what sequence.

That mapping exercise is precisely what bpExperts does. We work from the Business Flows architecture as a common language, overlay it against your actual end-to-end process coverage, and identify where AI can land with the lowest friction and the highest sustainable return — not just the highest vendor claim.

If the charts in this article made you think about your own domain coverage, that is a good starting point for a conversation.

---

This analysis is based on 1,700+ signals collected by our agentic pipeline across vendor publications, cloud provider blogs, and industry sources, mapped to the Business Flows end-to-end domain architecture. Signal counts reflect publication volume, not verified deployment counts. Effort and potential ratings are as attributed by original sources.