A five-person Bristol sustainability consultancy was losing 26 hours every week to carbon data collection, ESG report drafting, and chasing clients for emissions figures. That admin was capping the firm at nine ESG reports a month, while a growing pipeline of UK businesses facing mandatory sustainability disclosure requirements sat waiting. Three AI agents changed the maths: 67% more reports delivered, 22 hours recovered weekly, £80 a month to run.
The Admin Bottleneck Inside a UK Sustainability Practice
The firm — a specialist ESG and sustainability consultancy based in Bristol — helped UK SMEs and mid-market businesses produce their mandatory sustainability reports. Clients came to them for two reasons: Scope 1, 2, and 3 emissions calculations for net zero roadmaps, and structured sustainability disclosure reports aligned to the UK's Sustainability Disclosure Requirements (SDRs) and TCFD framework.
The technical side required their consultants. Understanding a client's emissions profile, identifying reduction opportunities, interpreting the regulatory disclosure standards — none of that was automatable. The admin surrounding that work was a different matter entirely.
Every client engagement followed the same sequence. The consultancy needed emissions data from the client: energy bills, travel records, supply chain spend, fleet fuel consumption. Getting that data meant sending a structured data request, waiting, chasing, validating what came back, and following up on gaps. For a typical mid-market client, data collection took six to eight hours across two to three weeks. Four to five of those hours were the consultants' time — sending reminders, fielding queries about what was needed, and manually validating submissions against expected ranges.
Once the data was in, writing the report took another six to eight hours per client. The format was largely standardised — executive summary, emissions inventory by scope, year-on-year comparison, reduction targets, and the regulatory disclosure narrative. The consultants described it as "writing the same document with different numbers each time." Accurate, but consuming.
On top of data collection and report writing: three to four hours per week monitoring regulatory updates. UK SDR requirements were evolving. TCFD guidance updates landed quarterly. The FCA's sustainability rules for investment products had knock-on implications for some of their corporate clients. Staying current was not optional — a report filed under an outdated framework could expose a client to compliance risk.
Total weekly admin drain: 26 hours. Nine ESG reports a month was the ceiling — not from lack of expertise, but because the surrounding admin was consuming the time the consultants needed to take on more.
We were turning away enquiries from businesses with genuine compliance deadlines. Not because the work was beyond us — because every report we took on committed us to twenty-plus hours of non-billable admin before we could even start the analysis. That's not a sustainable model when the whole sector is growing.
— Director, Bristol Sustainability Consultancy
The Three Agents They Built
The build took five weeks from scoping to all three agents running in production. The work was done in n8n, hosted on a low-cost VPS, using Claude for report generation and monitoring summaries. The data collection agent used the firm's existing client portal to receive submissions. No custom development was required.
Agent 1: The Data Collection Agent. When a new ESG engagement is confirmed, the firm sends the client a structured emissions data request — a multi-section form covering energy consumption, business travel, supply chain spend, fleet usage, and waste. When the form is submitted, the Data Collection Agent reads the response automatically. It validates each submitted figure against expected ranges for the client's sector and size (built into the agent's knowledge base), flags any implausible values for consultant review, identifies missing sections, and sends automated follow-up emails to the client contact requesting the specific missing data. Follow-up cadence: every three working days until the submission is complete. When all sections are validated, the agent generates a clean, structured data summary and notifies the lead consultant. Time to complete validated data collection: reduced from four to five consultant hours to under forty minutes of oversight.
Agent 2: The ESG Report Agent. When the validated data summary is saved to the client folder, the Report Agent triggers automatically. It reads the client's data, identifies the applicable disclosure framework from the project metadata — TCFD, UK SDR, or a bespoke format — loads the correct report template, and generates the full draft: executive summary, Scope 1/2/3 emissions inventory, year-on-year comparison, reduction targets based on the net zero pathway, and the regulatory disclosure narrative. The agent flags sections where client data is unusually sparse or where the regulatory language requires a judgement call, so the consultant knows exactly where to focus their review. Draft report ready in under eight minutes. Human review and edit: thirty to forty-five minutes per report. Previously: six to eight hours.
Agent 3: The Regulatory Monitoring Agent. This agent runs every weekday morning. It scans a curated list of sources — DESNZ publications, FCA regulatory updates, FRC sustainability guidance, ISSB standards updates, and the UK Green Finance Strategy communications — and identifies any new or amended guidance relevant to the firm's core frameworks. When something relevant appears, it generates a structured summary: what changed, which clients or report types are affected, and whether any action is required before the next client deliverable. Delivered to the lead consultant by 9am via email and Slack. Time previously spent on manual regulatory monitoring: three to four hours per week. Time now: five minutes to read the daily brief.
What Each Agent Actually Does, Step by Step
All three agents sit inside a single n8n instance. They share connections to the client portal (a Typeform-based data collection system), the document management system (Google Drive), and the team's communication tools (Gmail and Slack). LLM calls go to Claude via the Anthropic API. Each agent is a separate workflow — they don't interfere with each other and can run simultaneously during busy periods.
The Data Collection Agent is triggered by a webhook from the client portal on every new or updated submission. It runs a structured validation prompt against the submitted data, comparing figures against a sector benchmark table embedded in the system prompt. For out-of-range values, it generates a targeted clarification email — not a generic "please review" message, but a specific request: "Your gas consumption figure of 450,000 kWh appears high for an office of your stated floor area. Could you confirm whether this covers multiple sites?" This specificity dramatically reduces back-and-forth compared to generic follow-ups.
The Report Agent is triggered by a file-created event in the validated-data Google Drive folder. It reads the clean data summary, identifies the reporting framework from project metadata, and runs a structured generation prompt against the firm's master report template — a detailed system prompt that encodes the firm's house style, preferred phrasing for regulatory disclosures, and the specific structure of each section. Sections requiring consultant judgement are flagged inline with a comment: [CONSULTANT REVIEW: Net zero target year is more aggressive than sector average — please confirm approach]. The draft saves automatically to the client folder and sends a Slack notification to the lead consultant.
The Regulatory Monitoring Agent is a scheduled workflow that triggers at 8am on working days. It fetches the latest publications from each monitored source using RSS feeds and HTTP GET requests to government publication pages, runs them through a relevance filter prompt, and generates a morning brief. When nothing relevant has landed, the brief says so — no alert fatigue from false positives. For a deeper look at building automated monitoring systems like this, the AI agent observability guide covers the same trigger-and-summarise pattern applied to live agent monitoring.
The Results After 90 Days
Three months after all three agents went live, the numbers were clear:
- Report capacity: 9 per month → 15 per month. A 67% increase in throughput with the same five-person team. The additional capacity came entirely from recovered time — the consultants were no longer spending six to eight hours on data collection admin and report drafting for every engagement.
- 22 hours recovered per week. Across the team, the data collection, report writing, and regulatory monitoring that previously consumed 26 hours now requires four hours of human oversight — validating agent outputs, reviewing flagged sections, and making editorial decisions on the regulatory narrative.
- Data collection cycle: 2–3 weeks → 8–10 working days. The automated follow-up cadence and specific clarification messages cut the client data collection cycle nearly in half. Faster data in means faster report delivery — which clients noticed.
- Zero regulatory monitoring gaps. In the three months prior, the firm had one instance where a client report was drafted against guidance that had been updated a month earlier — caught in internal QA before delivery, but requiring a rewrite. Since the Monitoring Agent went live, every relevant update has been flagged within 24 hours of publication.
The firm took on six additional clients in the first full month the agents were running. At an average engagement fee of £2,200 for a full ESG report and net zero roadmap, that represented £13,200 of additional revenue capacity in month one. They're currently hiring a second senior consultant — not to replace the admin the agents now handle, but to do the analytical and client-facing work that the new capacity has created demand for. That's the right reason to hire: for the work that genuinely needs a person.
What It Actually Cost
The build was completed over five weeks. The running cost breaks down as follows:
- n8n self-hosted (VPS): £12/month
- Anthropic API (Claude) for report and monitoring outputs: £42–£55/month at current volume
- Typeform Pro (client data collection portal): £13/month
- Total monthly running cost: approximately £80/month
The payback period on the build cost was five weeks. At that point, the additional revenue from the capacity increase exceeded the cost of building the system. The agents have run with no meaningful maintenance since launch — the firm updates the regulatory monitoring source list quarterly, and the Report Agent's system prompt is reviewed when a major new disclosure standard is published.
For the technical approach to keeping API costs flat as report volume grows, the AI agent cost optimisation guide covers the caching and routing strategies that keep LLM spend from scaling linearly with throughput. For the testing approach used before going live with client deliverables, the evaluation framework post covers the golden-dataset method — running the Report Agent against twelve completed historical reports before deploying it on live client work.
If you run a consultancy — sustainability, compliance, professional services, or any practice where your revenue-generating work is surrounded by structured, repeatable admin — and you want to see what a similar build would look like, book a free 30-minute call. We'll map your current process, identify the highest-value automation targets, and give you a clear picture of what the build would cost and recover.