The managing director of a 12-person consultancy is looking at a sales board with 25 active opportunities and three SDRs running outbound. Her CRM has 8,000 contacts, last refreshed unevenly. Last quarter’s reply rate was 2 percent. A vendor on Tuesday demoed Apollo and promised 8 to 12 percent reply rates with a two-week setup.
She has £1,500 of monthly software budget and a CFO asking what the SDRs will do with the freed time. She is not deciding whether AI belongs in her sales motion. She is deciding which two of the deployable jobs to hand over first, and whether her contact list is clean enough to deploy any of them at all.
What jobs does AI do well in sales today?
Six. Lead qualification at roughly 50 percent cost reduction versus manual, recovering 30 to 40 percent of SDR time. Outbound email sequencing with personalised emails hitting 26 percent higher open rates and 10 percent more replies. CRM data hygiene at a 90 percent reduction in manual maintenance. Conversation intelligence on recorded calls. Meeting summarisation and CRM auto-population. Proposal and RFP automation, with named services-firm cases cutting time-to-submit by around 70 percent.
The pattern across these six is that AI handles structured, repetitive, evidence-backed work where the right answer is recoverable from data. Gong’s analysis of 7.1 million opportunities shows AI-using sellers generate 77 percent more revenue, with the lift coming from preparation and admin time, not from AI carrying the conversation. HubSpot’s 2026 State of AI in Sales report puts teams running five-plus AI tools at roughly 12 hours of recovered selling time per rep per week. The named precedent for proposals is Inventive AI’s 70 percent time-to-submit reduction, which is the post to read alongside this one if you sell on RFPs.
Where are the leaders using it?
HubSpot Breeze ranks third in UK SME adoption after ChatGPT and Canva AI, with Starter from £13 a month and Breeze AI built into Sales Hub. Apollo combines a 224 million contact database with multi-channel sequencing at £39 per user per month. Clay enriches outbound at £115 to £495 monthly using waterfall sourcing across 150 providers. Lavender coaches cold-email reply rates inline. Syrvi AI is the UK specialist, with case data showing 240 percent lead increases inside 90 days.
Gong sits at the enterprise end, £5,000 platform fee plus £1,300 to £1,600 per user per year and £15,000 to £65,000 implementation, named here as a ceiling reference rather than an SME recommendation. Fireflies and Avoma cover meeting summarisation at £10 to £19 and £19 to £79 per user per month, the price point that actually works at 5 to 15 person sales teams. Bain reports AI-enabled teams achieving 30 percent-plus win-rate improvements; the SyncGTM benchmark on full-stack AI is a 41 percent lift in revenue per rep, from £1.24m to £1.75m, on 18 percent fewer activities. The figures are real, but they are ceiling numbers, not floor numbers, and they assume clean data underneath.
Where does AI fall short in sales today?
Three places. Multi-step CRM tasks, where Carnegie Mellon documents 70 percent failure on multi-step office work and Salesforce finds advanced agents succeed on only 30 to 35 percent of multi-turn CRM tasks. Cold-email deliverability at volume, where Microsoft and Google now reject unauthenticated mail and Apple Mail Privacy Protection has rendered open rates unreliable. And multi-stakeholder enterprise selling, where the close depends on relationship work AI cannot do.
The failure modes are predictable. AI sees a £50m funding round and assumes growth mode; it could be debt during a downsizing. AI sees “VP of Sales” and treats a 50-person startup identically to a 5,000-person enterprise. Generic AI-generated outreach at volume gets systematically deprioritised by inbox providers because engagement now drives placement, and templated personalisation reads as templated. Real-time negotiation is the sharpest boundary. Red Bear’s research is direct on this point: the common wrong turns are behavioural, conceding too early, over-disclosing, failing to test buyer claims, negotiating with non-decision-makers. AI flags those risks in a pre-call brief; it cannot stop a seller collapsing under price pressure live.
What does a 90-day starter rollout look like?
Three phases. Weeks 1 to 4 are assessment and a data-hygiene sprint on the top 5,000 contacts. Weeks 5 to 12 are pilot deployment of one or two tools against a defined bottleneck, run in parallel to human selling. Weeks 13-plus are measurement against baseline, with scaling only if the pilot saved 5 to 10 hours per rep per week. The affordable starter stack is £69 a month.
That stack is ChatGPT Team at £25 per user, Make.com at £9, Notion AI at £8 per user, around £40 per person on a three-person team. The data-hygiene phase is the load-bearing one. AI qualification accuracy drops from 92 to 96 percent on clean data to 68 to 78 percent on dirty data. Teams that skip the sprint waste three to six months recovering, then come back to the same hygiene work they avoided in week one. Layer in tools by bottleneck: Apollo at £39 per user for outbound-heavy SDRs, Clay at £115 to £495 for enrichment, Fireflies at £10 to £19 for meeting capture, Inventive AI for RFP-driven services firms. HubSpot-native teams add Breeze inside an existing licence. The total ongoing for a five-person team typically lands between £250 and £1,000 a month, the right band for £1m to £10m turnover, and the spend should be set against recovered selling hours, not the headline reply-rate promise from the demo.
What should you ask before you commit?
Five questions separate a serious vendor from a marketing pitch. Reply-rate uplift on my actual contact mix, not your reference-case mix, with a sandbox pilot on my CRM. Per-seat cost at my projected volume, with the override floor if a month spikes. How the platform documents decisions for the ICO. SPF, DKIM, and DMARC posture. Whether it can explain why it scored a lead the way it did.
Black-box lead scoring fails the ICO’s transparency principle and creates a compliance gap. The EU AI Act’s transparency rules apply from August 2026 to any AI system deployed on EU customer data, which catches many UK firms with European clients. Insist on the sandbox pilot before signing, with a defined success metric, a defined timeframe, and a written rollback path if the numbers do not land. Pair this brief with proposal-ai-win-rate-myth and where-to-apply-ai-first before you commit. AI in sales multiplies the trust signals in both directions, which is why the procurement questions matter more than the demo.
If you would like to walk through which two workflows to start with for your firm specifically, book a conversation.
