AI SDRs: redesigning SDR and account executive roles

Salesforce's State of Sales survey of 4,050 sales professionals, fielded in August and September 2025, found that 87 per cent of sales organisations use some form of AI for tasks such as prospecting, forecasting and lead scoring. Some 55 per cent of sales professionals use AI for prospecting, with another 38 per cent planning to. The technology is no longer a pilot topic; it is a baseline.

Agents are the next step. Salesforce reports that 54 per cent of sellers say they have used agents and nearly nine in ten plan to by 2027. Sellers expect agents to cut prospect research time by 34 per cent and email drafting by 36 per cent. Those two tasks are the core of traditional SDR work, which is why the role is the first to be redesigned.

McKinsey estimated that generative AI could open up an incremental $0.8 trillion to $1.2 trillion in productivity across sales and marketing. Yet its survey found only 21 per cent of commercial leaders had fully enabled enterprise-wide adoption in B2B buying and selling, with 22 per cent having only piloted specific use cases. Adoption is wide but shallow, which leaves room for leaders who redesign the operating model rather than add tools.

Terminology is part of the confusion. An AI SDR can mean a copilot that suggests messages to a human, a workflow that sends sequences automatically, or an agent that decides whom to contact and what to say. Each carries different risk and different labour savings. Leaders should insist on a plain description of what the system does without human approval, what it does with approval and what it never does. Without that clarity, teams adopt tools under the same label that behave very differently, and results cannot be compared.

It is also worth separating tasks from jobs. A task, such as summarising a prospect's last annual report, may disappear. A job, such as building a relationship with a finance director who is sceptical of vendors, rarely does. Sales roles are bundles of tasks, and AI removes the easy parts of the bundle first. What remains is harder, more skilled and more valuable. That is why role design, not tool choice, determines whether a productivity gain turns into revenue.

The implication for a CRO is that the tooling decision is the easy part. The harder decisions are who supervises the agents, who handles the replies, and what the human team does with the time it recovers. What to do: list every task in the SDR and AE workflow, mark which ones agents can already do reliably, and assign an owner to each task that remains.

The business case starts with time. Salesforce found that sellers devote nearly one full day of the workweek to prospecting, yet 48 per cent say they lack the bandwidth to do adequate cold outreach. The average seller spends 40 per cent of their time selling, and reps lose about two hours a week to manual data entry.

AI SDRs address that gap by working at scale on tasks humans do slowly. Salesforce reports that its own agents contacted 130,000 leads and created 3,200 opportunities in four months. We treat any vendor-reported figure with caution, since outcomes depend on lead quality, offer and market, but the order of magnitude explains the interest: an agent can follow up every lead, including those a human team would never reach.

Top performers are using the capability. Salesforce found top-performing sellers are 1.7 times more likely to use prospecting AI agents for outreach than underperformers. That pattern is consistent with how earlier tools spread: leading teams adopt first and use the time they save for higher-value work, while weaker teams use the same tool to do the same activity faster.

Volume is not the only benefit. A human team follows up some leads promptly and others days later, or never. Agents can respond to every inbound enquiry within minutes and re-engage dormant contacts that a human team has stopped working. Salesforce describes using agents to revisit very large lead databases that no SDR team could attempt. The value lies in coverage and speed, provided the message is relevant and the offer is worth responding to.

Cost matters, but only after quality. A fully loaded SDR costs more than median on-target earnings once management, tools, training and turnover are included, and a team with 40 per cent annual attrition keeps paying ramp costs again. AI reduces some of those costs and adds others: licences, data, prompt design, review time and monitoring. The right comparison is cost per accepted opportunity, with an honest view of the human time the agents still require.

The implication is that the capacity AI frees must be redeployed deliberately. If it is not, the organisation will simply send more email. What to do: decide before deployment what the recovered hours will be used for, such as deeper account research, more discovery calls or faster follow-up on inbound leads, and measure that use in the first quarter.

The Bridge Group's 2025 SDR benchmark gives the clearest picture. Average SDR tenure is 1.9 years, ramp time is 3.0 months and only 60 per cent of SDRs reach quota, the lowest figure in the study's history. Median annual attrition was 40 per cent in 2024, made up of 13 per cent involuntary departures, 11 per cent voluntary and 16 per cent promotions.

Promotion is the career path that makes the SDR job attractive, and it has weakened. The Bridge Group notes that promotions have halved since the post-COVID boom, from 34 per cent in 2020. Median SDR on-target earnings are $80K, unchanged since 2022. A role with falling promotion rates and flat pay is a hard sell to ambitious people, even before AI starts to take the easiest tasks.

Activity levels are high. Median total daily activities are 112: 44 phone, 41 email, 19 LinkedIn and 8 text or other. Yet the median SDR-to-AE ratio is 1 SDR to 2.4 AEs, which implies that each SDR must generate enough qualified pipeline for more than two AEs. When volume is the main lever, the model is exposed to any tool that raises volume cheaply.

These figures are medians, and a CRO should read them as reference points rather than targets. The Bridge Group's interquartile range for attrition, 21 per cent to 57 per cent, shows how widely teams differ. A team with strong product-market fit, a sharp ideal customer profile and inbound demand will behave differently from one selling a complex, long-cycle product into a narrow market. Compare your own numbers before drawing conclusions.

The ramp figure deserves attention. At 3.0 months, ramp is the lowest since 2010, which suggests that tools and enablement have already shortened time to productivity. Agents can shorten it further by giving new SDRs ready research and message drafts. They can also make ramp matter less: if the agent does the first pass, the human needs less product knowledge to start and more judgement to improve. Hiring profiles should reflect that change.

The Bridge Group also reports that 2025 was the first year AI SDRs appeared as a distinct category, at 1 per cent of respondents. That figure is small, and it will not stay small. The implication is that the human SDR model is not stable, and that AI should be introduced into a redesign, not on top of the existing structure. What to do: use these benchmarks to compare your own tenure, ramp, attainment and promotion numbers before choosing how far to change the role.

Buyer research shows why more outreach does not equal more pipeline. In Gartner's survey of 632 B2B buyers, conducted in August and September 2024, 61 per cent prefer an overall rep-free buying experience, and 73 per cent actively avoid suppliers who send irrelevant outreach. Cheap, high-volume messaging that misses the mark is not neutral; it is a reason to be excluded from a shortlist.

Consistency matters as much. Gartner found that 69 per cent of B2B buyers report inconsistencies between the information on a sales organisation's website and what sellers tell them. AI-generated messages, written without access to current positioning, pricing or product facts, are likely to add to that inconsistency. A buyer who has to reconcile three versions of the truth is more likely to disengage.

Buying groups are also large. Forrester's 2026 State of Business Buying says the typical decision includes 13 internal stakeholders and nine external influencers, and that procurement professionals are decision-makers in 53 per cent of buying cycles. A single contact reached by an automated sequence is rarely enough. Multi-threading across a committee is a skilled, human task, and it becomes more valuable when first touch is automated.

There is a positive reading of the buyer data. Buyers who value self-service do not dislike help; they want help when they ask for it and accurate information when they do not. A seller who arrives with specific, correct knowledge of the buyer's situation is welcome. AI that supports that kind of preparation, rather than substituting for it, aligns with what buyers say they want.

Relevance has operational consequences. It requires current account intelligence, a clear definition of the ideal customer and the discipline to leave out accounts that do not fit. Many AI SDR programmes struggle because the target list is too broad. Narrowing the list reduces volume and raises quality, and it cuts the cost of data and sending. The SDR orchestrator's most valuable decision is often whom not to contact.

The implication is that the SDR's value shifts to judgement: choosing which accounts and which people deserve attention, reviewing what an agent drafts, and handling the reply. What to do: set explicit quality gates on agent output, give SDRs authority to stop a sequence, and measure complaints, opt-outs and reply sentiment as seriously as reply rate.

Gartner predicts that by 2028 AI agents will outnumber sellers by 10 times, yet fewer than 40 per cent of sellers will say agents improved productivity. The gap between adoption and perceived benefit is the design problem. Tools added to unchanged roles produce activity, not results; tools embedded in redesigned roles produce capacity that can be pointed at the right work.

We would define the redesigned SDR as an orchestrator. The role owns target account selection, agent configuration and prompts, review of drafted messages, handling of replies and the handoff to the AE. It needs skills that traditional SDR hiring did not emphasise: research judgement, writing quality, data discipline and the ability to hold a short, useful conversation with a senior buyer.

The AE moves towards deal architecture. With first-touch automated, the AE spends more time on discovery, on mapping the buying group, on commercial negotiation and on building consensus. Gartner's earlier forecast, that 60 per cent of B2B seller work would be executed through conversational interfaces by 2028, up from less than 5 per cent in 2023, suggests the interface layer will become routine. The human differentiator is judgement under uncertainty.

McKinsey's finding that only 21 per cent of commercial leaders had fully enabled enterprise-wide adoption shows how few organisations have redesigned work at scale. Early movers will learn faster, but will also make visible mistakes. A pilot pod limits the cost of error and produces internal evidence that sceptical sellers will trust more than a vendor's argument for change.

Pod design needs discipline. Keep it small, give it a clear scope of one segment and one offer, and avoid contaminating the comparison group. Because sales cycles can be long, track leading indicators in the first quarter, such as accepted-opportunity rate, reply quality and AE satisfaction, and leave the final judgement on win rate until the cycle completes.

Career paths need attention too. If SDRs become orchestrators, the route to AE changes from proving volume to proving judgement. Promotion criteria should reflect that, and promotion rates, which the Bridge Group says have halved since 2020, should recover if the role is worth having. A role with a visible path attracts the people who will make the agents effective.

The implication is that job descriptions, hiring profiles and career paths need rewriting. What to do: draft the new role definitions now, pilot them with a small pod of one SDR orchestrator, one agent set-up and two AEs, and compare the pod's accepted-opportunity rate, cycle time and win rate with the existing structure over two quarters.

Incentives drive behaviour more than technology. If SDRs are still paid on meetings booked or touches made, an agent that produces more meetings will be used to do so, regardless of quality. The right measure is qualified opportunities accepted by AEs, weighted by stage and by later win rate. That aligns the orchestrator with the AE's pipeline and with revenue.

Gartner's survey of 210 chief sales officers found 60 per cent say their revenue number is largely driven by elements outside their control. That is an argument for building quotas around inputs that the team can influence, such as account coverage, accepted opportunities and cycle time, and for reviewing them more often. Annual quotas set on volume assumptions will be wrong when the cost of volume falls.

The Bridge Group reports median quality conversations of 4.1 per day per SDR, the first rebound in its study's history. If agents handle first touch, the same SDR can focus on a smaller number of higher-value conversations. A new ratio of SDRs to AEs, perhaps fewer SDRs and more specialised ones, is likely to emerge. We would not set it in advance; we would let pilot data determine it.

Manager capability is part of the redesign. Managers who coached on activity counts must coach on conversation quality, targeting and reply handling. That requires message and call review time that many frontline managers do not have, so freed capacity should include manager time. Without coaching, the new roles revert to old habits within a quarter, because habits follow what is inspected.

Acceptance rate is the most honest metric available. If AEs accept a high share of SDR-sourced opportunities and those opportunities convert, the system works. If acceptance is low, the problem is targeting or qualification, however many meetings were booked. Publishing acceptance rate by source, including agent-sourced opportunities, also gives AEs a voice in the design and surfaces problems early.

Transition should be honest. Some SDR roles will not survive in their current form, and employees notice when leaders avoid the topic. Where reductions are likely, say so early, offer retraining for orchestrator and AE roles and use natural attrition, which the Bridge Group puts at a 40 per cent median, to adjust headcount without redundancies. Clear communication keeps the best people, who have options.

The implication is that compensation design is the lever that determines whether AI creates quality or noise. What to do: replace activity targets with accepted-opportunity targets, adjust on-target earnings to reflect the higher skill level of the new role, and give SDR orchestrators a share of closed-won revenue from the opportunities they originate so that quality is rewarded.

Salesforce found that 51 per cent of sales leaders with AI say disconnected systems are slowing down their AI initiatives. It also reports that top performers prioritise data hygiene more than underperformers, at 79 per cent against 54 per cent. Agents act on the data they are given. Duplicate contacts, stale titles and inconsistent positioning produce poor messages at scale.

Governance needs a named owner. Someone must decide which segments agents may contact, what claims they may make, how replies are handled, what happens to opt-outs and how personal data is stored. Email and privacy laws in the United States and elsewhere apply to automated outreach just as they do to human outreach, and the sender, not the tool, is accountable. Legal review of templates and suppression logic is a precondition for scaling.

Measurement should look beyond reply rate. Track accepted opportunities per hundred accounts contacted, opt-out and complaint rates, meeting show rate, pipeline created by source and win rate by origin. Without those measures, a programme can look successful while eroding domain reputation and brand. A quarterly review of sample conversations by the sales leadership team is a useful check on tone and accuracy.

Brand risk deserves explicit treatment. A single poorly worded agent message sent to a target account's board member can undo months of relationship building. Segment the contact list so that strategic accounts, existing customers and senior executives are handled by humans, with agents limited to lower-risk segments. Review the boundaries quarterly, since most reputational incidents come from boundary failures where an agent contacts someone it should not.

Vendor selection is a governance decision too. Ask suppliers where data is stored, how consent and suppression lists are handled, what logging exists and whether you can export all conversation records. Prefer platforms that integrate with the CRM rather than copy data into a separate system. Disconnected systems are a problem Salesforce's respondents cite often, and adding another silo makes it worse.

The implication is that CROs should treat AI SDRs as an operating system, with data standards, controls and metrics, not a feature. What to do: audit data quality in the CRM before deployment, set up a governance group with sales, legal and marketing, define the stop conditions that pause outreach automatically, and review them every month.

Karter McKinley helps sales leaders redesign commercial roles and operating models around AI: baselining the current team, designing SDR and AE roles, quota and compensation structures, and the data and governance needed to run agents safely. We work alongside sales operations and legal teams to put the design into practice and to measure outcomes. This briefing is general information, not legal advice.