Artificial Intelligence project reviews

Real feedback from clients who have shipped AI systems with us. We have included case studies below the testimonials so you can see the numbers behind the words.

By the numbers

38
Projects delivered
94%
Client retention rate
12
Industries served
4.8
Average rating (out of 5)

What clients say

★★★★★
"We asked ValueAI Seed to build a churn prediction model for our subscription platform. They delivered a working API in seven weeks. Within three months of using it, our retention team was reaching at-risk customers 11 days earlier on average, and monthly churn dropped from 4.2% to 3.1%."
Portrait of Rachel Okonkwo
Rachel OkonkwoHead of growth, StreamLoop
★★★★★
"Our legal team was spending roughly 600 hours a month reviewing supplier contracts for non-standard clauses. The NLP system ValueAI Seed built flags those clauses automatically and has cut review time by about 65%. The model handles English and German contracts equally well."
Portrait of Thomas Engel
Thomas EngelGeneral counsel, Brexworth Group
★★★★☆
"Honest and direct. During the discovery sprint they told us one of our three proposed use cases was not feasible with the data we had. That saved us from wasting budget. The two projects we did go ahead with are both live and performing above the targets we agreed."
Portrait of Amir Patel
Amir PatelCTO, Finsemble
★★★★★
"We run 14 bakery production lines. The computer vision system they installed catches packaging defects that our previous camera system missed entirely. Reject rates have fallen from 3.8% to under 1%, which at our volume translates to about £220,000 in annual savings."
Portrait of Sian Griffiths
Sian GriffithsOperations director, Hartwell Bakeries
★★★★★
"They integrated a document AI pipeline into our claims processing workflow. Adjusters now spend their time on judgement calls instead of re-keying data from PDFs. Processing time per claim dropped from 25 minutes to about 8."
Portrait of Marcus Hale
Marcus HaleVP of operations, Coverly Insurance
★★★★☆
"Good communication throughout. Weekly demos meant we always knew where things stood. The demand forecasting model they built for our warehouse network improved stock allocation accuracy by about 18 percentage points compared to our old spreadsheet method."
Portrait of Diana Kowalski
Diana KowalskiSupply chain manager, Northgate Distribution

Case studies

Reducing packaging defects at Hartwell Bakeries

Hartwell operates 14 high-speed bakery lines producing biscuits, crackers and cereal bars. Their existing rule-based camera system caught obvious defects (missing labels, crushed boxes) but missed subtle issues like misaligned seals and colour shifts in printed packaging.

We trained a YOLOv8 object detection model on 12,000 labelled images captured over four weeks from their production cameras. The model runs on NVIDIA Jetson Orin modules mounted beside each line, processing frames at 90 fps with inference latency under 35 ms.

After a two-week shadow period, the system went live. In the first quarter it flagged 4,200 defective packages that the old system would have missed. Reject rates fell from 3.8% to 0.9%.

0.9% reject rate (was 3.8%) ~£220k annual savings 35 ms inference
Bakery production line with camera inspection system

Contract clause detection for Brexworth Group

Brexworth's legal department reviews around 800 supplier contracts per month, each between 15 and 60 pages. Lawyers were manually scanning for 23 types of non-standard clauses: liability caps, indemnity carve-outs, auto-renewal terms and data processing obligations, among others.

We fine-tuned a transformer model on 2,400 annotated contract sections. The system reads each uploaded PDF, extracts text via OCR where needed, classifies every paragraph against the 23 clause types, and highlights matches in a review interface. Lawyers now focus only on flagged sections.

Average review time per contract dropped from 45 minutes to 16 minutes. The model achieves 93% precision and 89% recall across clause types, with a human reviewer confirming every flag before action is taken.

65% less review time 93% precision English + German
Lawyer reviewing AI-highlighted contract clauses on screen

Claims automation at Coverly Insurance

Coverly processes around 3,000 motor insurance claims per month. Each claim arrives as a mix of PDFs (police reports, repair estimates, medical certificates) and structured form data. Adjusters were spending roughly 25 minutes per claim just extracting and verifying data before they could begin the actual assessment.

We built a document AI pipeline that ingests each PDF, identifies the document type, extracts key fields (policy number, incident date, repair cost, injury description) and validates them against Coverly's policy database. Claims where all fields match and the total falls below the auto-approval threshold are routed directly to payment. The rest go to an adjuster with pre-filled forms.

Processing time per claim dropped to an average of 8 minutes. About 40% of claims now qualify for straight-through processing with no human touch at all.

8 min per claim (was 25) 40% straight-through 97.4% field accuracy
Claims processing dashboard on an office computer

Your project could be next

Tell us what you are working on. We will let you know whether AI is the right approach, and if so, what a realistic timeline and budget look like.

Start a conversation