Name: Automation results benchmark: before and after figures from seven delivered Otimiz projects
Creator: Otimiz
Published: 2026-09-24

Every figure on this page comes from a published case study of a real, named client project we delivered: four for Engage Media, one for Stand Up Recruitment and two for Mallory Portraits. I put the before and after of each one into a single table so you can compare them side by side, and every row links back to the full write-up it came from.

How this was compiled

Source. The seven case studies on this site that describe work for a named client. Only figures that trace to a verified source are used, each copied exactly as its case study states it. Nothing was rounded, added or re-estimated for this page.
Excluded. The three illustrative example builds on the case studies page (real estate, financial services and e-commerce). They describe typical projects rather than a specific client, so their figures do not belong in a table of delivered results.
Compiled. September 2026.
Size. Seven rows. One carries a measured before-and-after time and two carry a scale figure. The other four have no numbers: rows without numbers are included where no verified figure exists, and describe the outcome or what was built instead. Each row says which kind it is.

Seven rows is a small dataset. Read it as seven worked examples of what automation changed in a real business, not as statistics about automation in general. Where no verified figure exists, the row says what changed in words instead of inventing a number.

🗂️The data

Seven delivered projects, before and after

Measured first. Measured means a verified before and after time, Scale only means a size figure but no time saving, and Qualitative means no verified figure exists, so the row describes the outcome or what was built in words. Approximations such as "~" and "about" are the case studies' own wording.

Before and after results from seven delivered automation projects, with the source case study for each row
Process	Client	Before	After	What changed	Source
Monthly invoicingMeasured	Engage Media	~3 days every month of senior finance time across 35 client workbooks: order data extracted by hand, line items hand-calculated in QuickBooks, planned vs actual spend reconciled from vendor PDFs.	The monthly close takes about 30 minutes of review. Invoice drafts stop at a human approval gate and never auto-post to accounting.	Workbooks made machine-readable first, then a master tracker across all 35 clients, weekly invoice drafts, and planned vs actual reconciliation that flags variances.	FinOps invoice automation
Vendor data ingestionScale only	Engage Media	Data from ~20 active vendors as Excel, CSV, PDF screenshots and ZIPs of billboard photos, ingested into the reporting dashboards 100% manually.	An automated, validated pipeline. Low-confidence extractions go to human review, so nothing uncertain reaches a client dashboard.	Files routed by type, read with AI, de-duplicated, campaign names resolved against a canonical list, then merged into one reporting warehouse.	AI vendor reporting pipeline
Dashboard commentaryScale only	Engage Media	Commentary written by hand every month for every client dashboard, across 26+ clients, and no longer matching the data as soon as a client changed the date range.	Manual writing eliminated. The commentary matches whatever date range the client selects, and a person's edits are preserved when the workflow re-runs.	Data from nine ad platforms in one warehouse, a monthly workflow drafting commentary per client and per dashboard section, and the dashboard looking up the matching text.	AI reporting system
Weekly ad-ops analysisQualitative	Engage Media	Weekly manual analysis of search-term reports, RSA asset performance, and audience plus creative fatigue, pulled together by hand across Google Ads, Meta and the warehouse. The first thing to slip on a busy week.	An automated weekly report of specific actions: negatives to add, RSA pins to set, fatigued creatives to refresh. The AI recommends, the team decides.	One weekly workflow pulling Google Ads, Meta and the warehouse, three AI analyses with deterministic thresholds, and an error log so one platform failing never kills the run.	Ad ops optimization engine
Candidate intake, screening and schedulingQualitative	Stand Up Recruitment	Intake, CV screening, shortlist building and interview scheduling all done by hand across separate tools, with interviews booked over back-and-forth email.	Recruiters open the day with a shortlist to review instead of an inbox, and shortlisted candidates receive a booking link instead of back-and-forth email.	Applications from job boards and the careers inbox written to one structured record per candidate, each CV parsed and ranked against the role's must-have criteria, the recruiter notified with the candidate's full context.	Stand Up Recruitment
Partner event researchQualitative	Mallory Portraits	Manual, one-query-at-a-time research each cycle. Low coverage, repeated outreach to the same organizations, and no audit trail of what had been checked.	A weekly, deduplicated, audit-logged pipeline of vetted opportunities every Monday. Every entry is held for a four-day human review before any outreach.	A scheduled agent that searches the web for upcoming regional fundraising events, filters them for fit, de-duplicates against the CRM and the sheet, and logs every skip.	AI event discovery agent
Donations program CRMQualitative	Mallory Portraits	A legacy CRM, a separate task tool and several disconnected subscriptions. No centralized funnel, no clear ownership, and outreach sent by hand.	Four disconnected tools became one CRM funnel: a seven-stage pipeline with a named owner at every stage, and automated email and SMS outreach.	Everything consolidated into one GoHighLevel pipeline with two entry tracks, a manual pause tag for human control, and a documented handover so the team runs it themselves.	GoHighLevel donations pipeline

Four patterns in the rows

These follow from the seven rows above and nothing else. With a dataset this small, treat them as observations, not rules.

1. A person kept the final call in six of the seven

Invoice drafts stop at an approval gate. Uncertain vendor extractions go to a review queue. The ad-ops team decides which recommended moves to make. Recruiters review a ready-made shortlist. Both Mallory Portraits builds hold research-sourced entries for a four-day human review before any outreach. The seventh row, dashboard commentary, has no approval step, but it preserves a person's edits when it re-runs. How we design those checkpoints is in our guide to human-in-the-loop automation.

2. AI reads and drafts, rules decide

Five of the seven case studies say this explicitly: invoicing, vendor ingestion, dashboard commentary, ad-ops analysis and candidate screening all use AI to extract, draft or score, while the routing, maths, thresholds or storage stay deterministic. The donations pipeline case study describes no AI step at all, and it still replaced four tools. The reasoning behind that split is in AI routing vs deterministic rules.

3. Messy data got fixed before it got automated

In four rows the first job was structure: the invoicing workbooks were made machine-readable before anything read them, data from around 20 vendors was mapped into one shape, applications were written to one record per candidate, and four tools were consolidated into one CRM. Where the input was a mess, cleaning it up was part of the result, not a step before it.

4. Only one row has a verified time saving

Monthly invoicing is the one row with a verified before-and-after time: ~3 days a month to about 30 minutes of review, on work repeated across 35 client workbooks. Two more rows carry a scale figure (~20 vendors, 26+ clients) with no time attached, and four have no verified number at all. That is not evidence that nothing was saved in those projects. It means the time was never measured in a way we can publish. If you want a before-and-after figure for your own project, measure the current process before the build starts.

How to cite this page

If you reference these figures, please link to this page or to the case study each figure comes from, which carries the full context.

Otimiz, "What automation actually saved: results from our delivered projects", September 2026, https://otimizagency.com/guides/automation-results-benchmark/

Bastien Daumas

Founder, Otimiz

Ex English-French translator who built a translation SaaS, automated the backend with no-code, and turned that into an AI automation agency. Based near Nice, working with B2B clients globally. No false promises, just fewer repetitive tasks and concrete results.

🔗Keep reading

Related guides and case studies

Not ready to talk yet? Get a free written automation audit: three automations worth building for your business, with rough hours on each. No call. Request your audit.

What automation actually saved: results from our delivered projects