Book a Demo Try Speak AI Free Login
Case study · Public sector & economic development · An economic development organisation

How an Economic Development Organisation Automates Enquiries, Meetings and Calls With AI

An economic development organisation uses Speak AI to sort and draft replies for its shared enquiries inbox, record and summarise its meetings, and answer website visitors by voice, so a small team handles public questions without losing accuracy.

Book a DemoTry Speak AI Free
As of October 2026Measured from Speak usage records. Voice figures include internal test conversations.
  • 80+
    Voice agent conversations
  • About 4 hours
    Of voice conversation
  • 34
    Meetings recorded by the assistant
  • 22+
    Hours of meetings recorded

Summary

The organisation's public inbox collects thousands of messages, mostly automatic replies, newsletters and notices, with a few real questions from job seekers and businesses buried among them. One staff member answered by hand. The team wanted faster first replies without a single wrong answer going out in its name.

It now uses three Speak AI tools together. Inbox automation sorts mail and prepares draft replies that staff review before anything is sent. A voice agent on the website answers visitor questions. A meeting assistant records recurring meetings and writes summaries with action items.

Usage is measured. The voice agent has held more than 80 conversations, and the meeting assistant has recorded 34 meetings, more than 22 hours. Drafts stay under human review, and the knowledge sources are limited to what the team can verify.

A woman in a navy cardigan reads her laptop at a tidy desk in a bright office.

Three lines to take away

Before

A small team answered a crowded public inbox by hand, and common questions went unanswered outside office hours.

With Speak

Speak AI sorts the inbox and drafts replies for review, answers visitors by voice from approved sources, and summarises meetings.

After

Staff review drafts instead of writing from scratch, and visitors get an answer straight away.

A public inbox that needed a person for everything

Most of what arrived did not need a reply, but someone still had to find the few messages that did.

A tired woman with a phone on her shoulder looks at a crowded laptop screen at a cluttered desk.

Real questions buried

Automatic replies, newsletters and notifications made up most of the mail. A job seeker or a business with a site question sat among them until someone found it. Staff also had to tell genuine enquiries from the daily stream of automatic out-of-office replies to the organisation's own news digest.

Answers written from scratch

The same questions about jobs, support for new businesses and who to contact came in again and again, and each reply was typed by hand. Quality depended on who was free and how much time they had that day.

Nothing outside office hours

The website offered a contact form and a phone number. A visitor with a quick question after hours had no one to ask.

The cost: The inbox held a backlog of several thousand unread messages, and the one person answering had little time left for higher-value work.

The turning point

Draft first, send later

The idea clicked when the team saw that automation did not have to send anything. Speak could sort the inbox, file the noise out of sight, and leave a draft reply next to each real enquiry. A staff member would still read, adjust and send every one. The team would also see why a draft was written, so trust could build message by message rather than on a promise.

The main objection was accuracy. The team worried more about a wrong answer than a slow one, and it rejected a plan to feed the system old documents because outdated material would produce outdated replies. The agreed design used only the live website and current job listings as sources, with web search off, and flagged anything uncertain for a person instead of guessing.

That same rule shaped the voice agent. It answers from the organisation's own content and says so when it does not know.

How they use Speak AI

Four pieces work together, each limited to sources the team trusts.

  1. Step 1

    Sort the inbox

    New mail is classified as it arrives. Automatic replies, newsletters and notices are filed out of the inbox, and nothing is deleted, so every decision can be reversed.

  2. Step 2

    Draft replies

    Real enquiries from job seekers and businesses get a draft written from the website and current job listings. Staff review and send, and drafts are paired with the final version to improve later ones. Job seekers are pointed to current openings, and business enquiries are flagged with a suggested owner.

  3. Step 3

    Answer visitors

    A voice agent on the website answers questions about jobs, starting a business and who to contact. Where it lacks an answer it points to the contact page.

  4. Step 4

    Record meetings

    The meeting assistant joins scheduled video meetings. An automation then writes a short summary with action items for the team. The summary sits beside each recording, so no one has to write it up.

Honest friction: Access to the mailbox took several weeks because it sat with an outside IT provider, and the first test failed on a permission setting. Early voice conversations showed gaps, such as local statistics the agent had no source for, and contact details were captured badly until the instructions were fixed. Visitor use started modestly, and the team is still deciding how widely to promote it. Recorded meetings were mostly in 2025, and recent use has moved to the inbox and the voice agent.

An older man wearing earbuds speaks to a laptop at a wooden table in morning light.

Book a Demo

Results

Measured use across voice and meetings

The voice agent has held more than 80 conversations, about four hours in all, and the meeting assistant has recorded 34 meetings, more than 22 hours. Both tools run alongside the inbox work, which is reviewed by staff rather than counted in usage records.

2
Automations running: meeting summaries and personalised email drafts
About 2 minutes
Median length of a voice conversation
60+
Voice conversations that ran a minute or longer
How we calculated this

An estimate, not a measurement

The usage numbers on this page are measured from Speak records. This block is a separate estimate of the manual effort the same workload would take. It uses published time benchmarks at the low end of their ranges and official wage statistics at the 25th percentile. Most teams already have transcripts, so Speak’s value here is the analysis layer.

Headline for this study

Starting from: already using a transcription service

Counts only the analysis, grading and notes effort. The conservative baseline.

At least 5 hours
about US$300 at the low end

Up to about 30 hours and US$1,400 at the high end of the published ranges.

Alternative baseline

Starting from: no transcription today

Baseline (a) plus manual transcription. The upper bound.

At least 80 hours
about US$2,000 at the low end

Up to about 160 hours and US$5,200 at the high end of the published ranges.

Which baseline we lead with: No call or dossier shows the before-state was listening back or typing notes by hand, so we use the conservative baseline that counts documentation effort only.

Task Manual time (published range) With Speak (our assumption)
Documenting meeting notes and actions 0.7 to 1.82 hours per hour of meetings
Allocation of Physician Time in Ambulatory Practice: A Time and Motion Study in 4 Specialties		 0.3 hours per hour of meetings
Our assumption: 0.3 hours to read and correct an automatic summary per hour of meeting.
Manual transcription (baseline b only) 4 to 7 hours per hour of meetings
Realities Toolkit #08: Transcribing your own qualitative data		 0.25 hours per hour of meetings
Our assumption: 0.25 hours to check an AI transcript.

Workload (measured): 20 hours of meetings. Meeting-assistant recordings (not deleted) for the account, total duration of 34 recordings, pulled 2026-10-11.

Hourly cost: US$26.90 (25th percentile) to US$33.41 (median) for First-line supervisors of office and administrative support workers; transcription is costed at US$16.42 to US$19.43. Both are multiplied by 1.43 to include benefits and employer costs. US wages are used as a neutral reference.

Hours saved = workload x (manual hours – hours with Speak). Value = hours saved x hourly wage x 1.43.

This is equivalent manual effort, not a measured saving. It covers only the recorded meetings, and not every meeting would have been documented in full by hand. The benchmark is a clinical documentation study used as a proxy.

As of October 2026. This estimate is not customer-reported, is not audited, and does not include software costs.

What the team found, in short

Paraphrased from our own conversations with the team. Not quotes.

Drafting for human review gives most of the speed of automation with none of the risk of an unchecked reply going out.

Paraphrased from our calls with the team

Limiting an assistant to sources the team can verify matters more to a public body than the breadth of what it can answer.

Paraphrased from our calls with the team

Questions a voice agent cannot answer are a map of what the website and knowledge base are missing.

Paraphrased from our calls with the team

What is next

Three colleagues gather around a laptop at a meeting table in a bright office.
  • Wider inbox coverTeams like this extend the same sort-and-draft pattern to other staff mailboxes once the shared inbox runs well. Each added mailbox gets its own review step before anything is sent.
  • Closing knowledge gapsReviewing unanswered visitor questions on a schedule shows which facts to add to the knowledge base.
  • Lead captureA voice agent can take a name and email for follow-up on high-intent questions, which makes enquiries countable.

More customer stories

Customer name withheld at their request. Figures are from Speak usage records, as of October 2026.

Automate your public enquiries

Talk to us about sorting your inbox, drafting replies for review and answering visitors by voice.

Book a DemoTry Speak AI Free