1. Why I am writing this
I have spent eighteen years on the inside of three different waves of business transformation. As an internal auditor and, later, a process architect at BRD–Société Générale, I watched a major retail bank work through Lean and Six Sigma in the 2010s, making each step of every core process a little faster and a little cleaner. As COO at Modex Technologies, I led the transition of a fintech company from on-premise software to a software-as-a-service delivery model, which was a different exercise in redesigning how work is delivered. Today, as CEO at Safeflows, I am building a go-to-market engineering company that helps B2B firms generate predictable pipeline through AI, data and orchestration workflows. Across those three waves, the same pattern keeps repeating: the organisations that win are not the ones that apply better tools to the old motion. They are the ones that change the motion itself. I believe B2B sales is now at one of those moments, and that most sales leaders are responding to it the wrong way. They are adding AI tools on top of a process that was designed for a world that no longer exists. My argument in this article is that sales is becoming an engineering discipline, and that the leaders who treat it that way will own the next decade of pipeline. The leaders who treat AI as an upgrade to the old playbook will spend the next five years wondering why their pipeline is shrinking even though their activity numbers keep going up. I want to set out why I believe this, what we have learned at Safeflows from working with mid-market B2B clients across very different industries, and what I think this means for sales leaders preparing for 2028.
2. The old playbook stopped working
For most of my career, sales improvement was treated as a process-optimisation problem. The toolkit was familiar to anyone who spent time in a regulated industry: process mapping, KPIs and KRIs, root-cause analysis, continuous improvement. Apply Lean to lead-to-cash. Apply Six Sigma to the funnel. Get cleaner data into the CRM. Write better email templates. Run more sequences. Add headcount when you need to scale. It worked, for a while. Now it does not, and the reason is structural. First, tactical playbooks have been commoditised. The same outbound templates, the same sequencing logic, the same prospecting tools are available to every competitor, and generative AI has accelerated this further; any tactical edge erodes within months. Second, AI has shifted where the bottleneck sits. Research and content production used to be the constraint; today they are nearly free. What is genuinely hard now is identifying the right account at the right moment and orchestrating a coherent response across the four or five channels a buyer actually uses. Third, buying journeys are fragmented. Prospects move across email, LinkedIn, content, communities and CRM touchpoints, and the linear funnel of the previous era is a poor model of how purchasing actually happens.
Deloitte’s most recent State of AI in the Enterprise survey of 3,235 senior leaders across 24 countries gives a sense of the gap. Seventy-four per cent of companies plan to deploy agentic AI within two years, and twenty-five per cent already say AI is having a transformative effect on their business, more than double the share recorded a year earlier. But eighty-four per cent of those same companies have not redesigned the work itself around AI; they are layering new technology onto legacy processes (Deloitte, 2026). When I look at the sales teams I have spoken to in the past year, this is exactly what I see. Everyone has AI now. Almost no one has redesigned what they actually do.
3. What I mean by sales as an engineering discipline
When I say sales is becoming an engineering discipline, I mean something specific. I do not mean that sales representatives should learn to code, and I do not mean that AI will replace them. I mean that the unit of work in sales is shifting from the individual representative to the system that organises their work. We design and run that system the way a software team designs and runs a product, with data models, instrumentation, governance, version control and continuous improvement. At Safeflows we describe what we build for clients as a four-layer system. The labels are simple, but each layer represents a category of decision that most teams skip.
The data layer is where most teams begin in the wrong place. The Ideal Customer Profile (ICP) is not a one-page document; it is a structured data model with best-fit accounts, mapped buying committees and signal-rich attributes. Most teams attempting AI-enabled outreach start by writing personalised messages without first engineering the underlying account model. That is the equivalent of optimising the user interface of a product whose data schema has not yet been defined. I see this mistake almost every week.
The signal layer is where the timing decision lives. Hiring patterns, funding events, technology adoption, market moves and community engagement are the events that tell us when an account becomes relevant. Without a signal layer, outbound is a guessing game with better grammar.
The orchestration layer is where workflows automate scoring, routing, enrichment and handoff, so that every account moves through the funnel according to its fit and signal strength rather than the calendar of whichever representative is available. Deloitte (2025) maps this to what they call the Orchestrate level of the agentic autonomy ladder, where adaptive contextual autonomy enables multi-step workflows with human exceptions. Gartner has forecast that roughly one third of enterprise applications will embed autonomous agents by 2028 (Keen, 2025); I believe sales orchestration will be one of the first and largest domains for that embedding.
The insights layer is what most teams ignore until it is too late. Dashboards, quality-assurance loops, post-meeting analysis and reporting turn execution into learning. Without an insights layer, you do not have a system; you have a sequence of campaigns. If I map our clients and prospects to Deloitte’s autonomy ladder, almost all B2B sales organisations I see today are at levels 0 to 1, using AI assistants to draft emails or summarise calls while leaving the underlying motion untouched. A small minority operate at levels 2 to 3, executing bounded workflows end-to-end and orchestrating multi-step sequences. The frontier is level 4, where the system itself becomes the source of advantage, and that is where I believe the leaders of the next decade will live. The most common mistake I see — and one we made ourselves at Safeflows before correcting course — is that teams start with message copy, before the ICP, the signals, the data workflow and the CRM architecture have been engineered. The discipline I am describing reverses that sequence.
4. What we see at Safeflows
I want to be honest about the evidence behind what I am writing. Safeflows is two years old, and we have worked closely with a handful of mid-market B2B clients across very different industries: industrial equipment, computer-aided engineering software, wellness technology, travel partnerships, advisory services and an early-stage technology startup. The numbers I share here are real, but the sample is small. I am not claiming a statistical result; I am describing a pattern. The pattern is this: outcomes correlate with the depth of system engineering, not with the volume of outreach activity. When we begin an engagement with explicit ICP construction, signal mapping and data enrichment, we see measurable results. In one industrial-equipment engagement, four verticals were activated and qualification speed roughly doubled. In an engineering-software case, eighteen qualified meetings were generated across two channels in the first cycle. In a wellness-technology engagement, more than 120 target accounts were mapped and fourteen qualified conversations were initiated across two markets. When engagements begin with outreach copywriting instead, before the underlying system has been built, results are inconsistent regardless of effort. This matches the broader evidence. Deloitte (2026) reports that only twenty-one per cent of companies have a mature governance model for autonomous agents, and only one in four has moved 40 per cent or more of their AI experiments into production. Gartner is blunter: Keen (2025) reports a forecast that more than 40 per cent of agentic AI projects will be cancelled by the end of 2027, mainly because the underlying engineering and governance are not in place. That last point deserves more attention than it gets. Governance in the agentic era is not a compliance task you handle after deployment. It is part of the design from day one: deliverability monitoring, message audit trails, escalation pathways for AI-drafted communication, and clear boundaries on where agents may act autonomously versus where they must seek human confirmation. In our own work we have learned that this is the difference between a system you can run for a year and a system that gets your domains blacklisted in six months.
5. What I think this means for sales leaders
If you lead a sales organisation, here is what I think this transition asks of you. The role of the senior sales leader shifts from manager of activity volume to architect of a revenue system. New functions emerge at the intersection of sales, data engineering and AI operations: go-to-market engineers who design the data and signal layers, revenue operations architect who build the orchestration, and the equivalents of the Agentic Process Architect and Multi-Agent Systems Engineer roles described by Deloitte (2025), now applied to the revenue function. At Safeflows we have started hiring for exactly this profile, and the talent is harder to find than I expected, because the role does not yet have a settled name in the market. The integration decision is harder to defer. Deloitte (2025) describes three patterns: agents overlay, where agents are added on top of existing workflows; agents as a service, where embedded agentic features are consumed from CRM and outreach platforms; and agents by design, where the workflow is re-architected from the ground up around autonomous agents. Most B2B sales organisations should start with overlay and as-a-service for quick wins, but durable advantage will accrue to those that ultimately move to agents-by-design for the core revenue motion. The longer that decision is deferred, the more the legacy motion becomes the constraint on what you can build. The risk of doing nothing is not stasis; it is accelerating disadvantage. Deloitte (2026) finds that fifty-four per cent of organisations expect to move 40 per cent or more of their AI experiments into production within the next three to six months. The sales organisations that will define the next decade of B2B pipeline are the ones engineering the system now, not the ones optimising the activities the system will replace.
6. Where I think this goes
Every wave of digital transformation I have lived through — enterprise resource planning, software-as-a-service, cloud — has eventually demanded that the work itself be redesigned, not merely augmented. The current wave, defined by autonomous AI and the architectures that support it, is making the same demand of the B2B sales function. I want to be explicit that this is not an argument for taking the human out of sales. Quite the opposite: the value of judgement, relationship-building and strategic thinking goes up, not down, when the system absorbs everything routine. What changes is what we ask the human representative to do, and what we build around them so they can do it. The systems I am describing free representatives from the work that does not benefit from being human in the first place, and concentrate their time on the work that does. The shift is neither complete nor inevitable. It will be uneven across industries and organisations, and it will require investment of a kind that has historically been reserved for product development. But the direction is clear, and the window in which early movers can establish a durable advantage is narrowing.
I believe the future of sales is being engineered right now. The question I would ask every revenue leader — and the question I ask myself most weeks at Safeflows — is whether we are among those doing the engineering, or among those competing with the systems that others have built.
References
Deloitte (2025) Agentic enterprise 2028: A blueprint for cost savings, job creation, and faster growth through agentic AI. Deloitte Consulting LLP. Available at: https://www.deloitte.com/us/en/what-we-do/capabilities/applied-artificial-intelligence/articles/agentic-ai-enterprise-2028.html (Accessed: 24 May 2026).
Deloitte (2026) State of AI in the Enterprise: The untapped edge. Deloitte AI Institute. Available at: https://www.deloitte.com/us/state-of-ai (Accessed: 24 May 2026).
Keen, E. (2025) Gartner Predicts Over 40% of Agentic AI Projects Will Be Canceled by End of 2027. Gartner Newsroom, 25 June. Available at: https://www.gartner.com/en/newsroom/press-releases/2025-06-25-gartner-predicts-over-40-percent-of-agentic-ai-projects-will-be-canceled-by-end-of-2027 (Accessed: 24 May 2026).
ABOUT
- Cristian Bărbuia, Founder & CEO, Safeflows, ROCA EMBA Student

