This is an anonymised case study from one of our 2026 engagements with a MENA-based B2B SaaS company. Names and specific numbers have been adjusted to protect the client; the methodology, the deliverables, and the directional ROI are faithful to the actual engagement.

The aim of publishing this: show what an AI-driven automation engagement actually looks like from initial brief to measured outcome. If you are evaluating similar work, you should have a realistic mental model of the inputs, the work, and the realistic returns.

The client at engagement start

B2B SaaS, mid-market segment, MENA-headquartered with regional revenue
ARR ~USD 8M, growing 35% YoY
Sales team: 4 SDRs + 2 AEs + 1 head of sales
Their problem: SDR cost was ~USD 240K/year fully loaded for a function that was, on close inspection, 70% manual data entry and follow-up. They could not afford a 5th SDR but they were leaving demos on the table.
Their existing stack: Salesforce CRM, Outreach for sequences, ZoomInfo for enrichment, Slack for internal coordination.

The brief

“We want to free our SDRs from manual work so each one can run 2x the pipeline they do today, without sacrificing quality. We have a 12-week timeline before our Series B raise and need measurable productivity gains by then.”

What we delivered (12 weeks)

Three workstreams, run in parallel:

Workstream 1 — Inbound triage automation (weeks 1–4)

Built an n8n workflow that:

Watches the marketing-form intake (HubSpot → Salesforce)
Enriches each lead via Clearbit + ZoomInfo (cost-throttled — only enriches if our scoring model gives the lead ≥ 50/100)
Classifies into ICP-tier-A / B / C / out using Claude 4.7 with the company’s actual qualification rubric in the system prompt
For tier-A: auto-creates a Salesforce opportunity, assigns to the round-robin SDR, drafts the first outreach message (the SDR reviews + sends — they do not auto-send)
For tier-B: enters the standard 14-touch nurture sequence
For tier-C / out: politely declined with a templated email

Workstream 2 — Outbound list orchestration (weeks 3–8)

Custom n8n workflow with web research nodes that scan public sources (LinkedIn, Crunchbase, the prospect’s own website, recent press) and produce a 3-sentence “why this prospect, why now” briefing for the SDR
The briefing is attached to the outbound sequence as context; the SDR uses it to personalise the first email
LLM eval: 80 sample briefings rated against SDR-written counterparts on a 1–5 scale. AI: avg 3.8. Human (SDR): avg 3.7. The AI is not better; it is as good, at zero marginal cost.

Workstream 3 — Post-demo follow-up (weeks 5–10)

After a demo, an SDR fills a 30-second form in Slack (slash command); the form data hits an n8n workflow
The workflow drafts a personalised follow-up email + a one-page recap document referencing the customer’s specific use cases from the call (using meeting transcript via Gong/Fireflies as input)
SDR reviews and sends. Median SDR time per follow-up: 22 minutes → 4 minutes.

What we did NOT deliver

Worth listing what was explicitly out of scope:

We did not let the AI send emails autonomously. Every customer-facing message goes through a human reviewer. The reason is not performance; it is brand risk.
We did not replace the CRM (Salesforce stayed). Augment, not replace.
We did not touch the AE workflow at this stage. Phase 2 if Phase 1 wins.
We did not promise specific revenue lift. We promised productivity lift, measured in time saved + pipeline volume.

The measured outcome (week 12)

The numbers below are directional but accurate within ±10%:

Metric	Before	After (week 12)	Delta
SDR pipeline created per month	38 opps	71 opps	+87%
Median SDR time per outbound prospect	38 min	11 min	-71%
Median SDR time per inbound triage	12 min	2 min	-83%
Median SDR time per post-demo follow-up	22 min	4 min	-82%
Lead-to-opportunity conversion rate	14%	19%	+5pp
Customer-feedback score on “personalisation of outreach”	3.2 / 5	3.6 / 5	+0.4
Tooling cost added (n8n self-hosted + Claude API + enrichment)	—	~USD 1,400 / mo	—

Translation: each SDR now produces the pipeline of ~1.8 SDRs at zero added headcount. Annualised, that is the productivity-equivalent of 3.2 SDRs from a team of 4 — without hiring. At loaded cost of ~USD 60K/SDR/yr, that is ~USD 192K in productivity gains for ~USD 17K in annual tooling cost. Engagement-side fees are deliberately not stated; ask if relevant.

The conversion-rate lift (+5pp) was the surprise. We did not promise it. The hypothesis is that better-personalised outreach (the AI-drafted briefings) lifted both acceptance of meetings and quality of accepted meetings.

What we got wrong

Three honest pieces of post-mortem:

Underestimated change-management. Week 1–3 had near-zero adoption because SDRs did not trust the AI’s lead-tier classifications. The fix was making the system prompt and scoring rubric visible to the SDRs so they could see why a lead got the tier it did. Adoption hit 100% by week 5.
Overengineered briefing detail. First version of the outbound briefing was 8 sentences. SDRs ignored it because they did not have time to read. Trimmed to 3 sentences + 1 conversation-starter; usage tripled.
Slack form was a UX bottleneck. First version had 12 fields. Usage was low. We cut to 4 fields + auto-extraction from the meeting transcript. Usage went from 30% of demos to 95%.

How we would replicate this

If you are running a similar function and want to do this yourself, here is what we would tell you:

Build the eval set first, on real qualification decisions from your team. 50–100 examples is enough.
Start with one workflow, not three. Pick the SDR’s most-hated manual task; automate that. Earn trust before scoping more.
Self-host n8n if you can. Egypt has data-residency expectations on customer data that n8n Cloud (EU only) does not always meet. Self-hosted on a Hetzner or OVH box runs ~USD 30/month.
Keep humans in the loop on customer-facing outputs. The AI drafts; humans review-and-send. The 5–10% of cases the AI gets wrong are exactly the cases that would break trust the most if they slipped through.

If you would like us to scope a similar engagement for your team, contact us at contact@kalastor.net — initial scoping conversations are free and take ~30 minutes.

Disclosure: numbers in this case study have been adjusted to anonymise the client. The methodology, the workstream definitions, the timing, and the directional ROI ratios are faithful to the actual engagement.