Ashford & Dwight Talk to us

Data engineering · Analytics · Applied AI

Data you can defend.

We build the pipelines, models and reporting that turn scattered business data into numbers your team trusts and decisions you can explain. Our methods come from quantitative finance research, where every figure has to survive scrutiny.

Talk to us See our services
Example client workspace Illustrative data
Workspace/Overview
Daily run · ready 06:12 Live
Data ready by
06:12daily
On schedule
Tests passed
412 / 412
Forecast accuracy
95.2%
Sources synced
3 / 4
Bank feeds retrying

Forecast accuracy, last 12 weeks

Weekly accuracy
96% 95% 94% 93% 92% W1 W4 W7 W10 W12 95.2%

Sources

3 of 4 synced
  • ERPsyncedlast load 05:51
  • CRMsyncedlast load 05:49
  • Billingsyncedlast load 05:50
  • Bank feedsretryingattempt 2 of 3

Lineage

flowingwaiting on retry
SOURCES STAGING WAREHOUSE OUTPUTS ERP CRM Billing Bank feeds stg_erp stg_crm stg_billing stg_bank fct_orders dim_customers fct_cash demand_forecast board_pack kpi_dashboard

Recent runs

Today
JobStatusDurationFinished
ingest.erp_dailyok3m 12s05:51
model.demand_forecastok11m 04s06:02
report.board_packok48s06:12
ingest.bank_feedsretry 2/3––

Run log

tail
  1. ingest.erp_dailyok 3m 12s
  2. tests.staging412/412
  3. model.demand_forecastok 11m 04s
  4. ingest.bank_feedsretry 2/3
  5. report.board_packok 48s
  6. data ready
  • Strategy
  • Engineering
  • Analytics & BI
  • Machine learning
  • Managed operations

Services

From raw data to a decision you can explain.

We scope each engagement to the problem in front of you. Most clients start with one service and add others as the work proves itself.

01

Data strategy & audit

Find out what your data can support before you spend on it.

  • Source and system inventory
  • Data quality assessment
  • Prioritised roadmap with costs
02

Data engineering

Pipelines that load, clean and version your data on schedule.

  • Warehouse design
  • Automated ingestion
  • Tests on every load
03

Analytics & dashboards

Reporting your team opens every morning, with one definition per metric.

  • Agreed metric definitions
  • Dashboards and board packs
  • Self-serve access
04

Predictive modelling

Forecasts and risk models, backtested before anyone relies on them.

  • Demand and revenue forecasting
  • Credit and churn risk
  • Backtests and error tracking
05

Applied AI

Language models put to work on your documents and workflows, with measured accuracy.

  • Document extraction
  • Search over internal knowledge
  • Evaluation against labelled samples
06

Managed analytics

We keep it running after launch, and change it as the business changes.

  • Pipeline monitoring
  • Model retraining
  • Monthly review

How we work

Four stages. Each ends with something you can use.

01 / 04

Discover

2–4 weeks

We map your sources, interview the people who own them and test what the data can support.

DeliverableData audit and roadmap

audit / source_inventory.csv Discover
SourceOwnerTablesField coverageStatus
ERPFinance3898%mapped
CRMSales ops2191%mapped
BillingFinance1496%mapped
Bank feedsTreasury672%in review
Sources mapped
3 / 4
Owners interviewed
3 / 3
Roadmap items
7
Output: audit.pdf · roadmap.xlsxWeek 3

02 / 04

Build

4–10 weeks

We build the warehouse and pipelines, with tests and documentation from day one.

DeliverableTested, documented pipelines

models / staging / stg_orders.sql Build 
-- one row per order, cleaned from the ERP exportwith source as (    select * from {{ source('erp', 'orders') }}),cleaned as (    select        order_id,        customer_id,        cast(order_date as date)  as order_date,        round(amount_net, 2)      as amount    from source    where amount_net is not null)select * from cleaned
$ dbt test --select stg_orders PASS unique_stg_orders_order_id 0.41s PASS not_null_stg_orders_amount 0.22s PASS relationships_customer_id 0.57s Done. 3 of 3 passed

03 / 04

Model

3–8 weeks

We build forecasts, risk models or AI tools and backtest them against your history.

DeliverableBacktested models with error reports

models / demand_forecast · backtest Model
Weekly units (thousands) actualforecast80% interval
60 55 50 45 40 W1 W7 W13 W19 W25 train holdout
HorizonMAPEBiasIn interval
1 week3.1%+0.2%8 / 8
4 weeks4.8%−0.4%8 / 8

04 / 04

Operate

Ongoing

We monitor, retrain and report, and adjust as your questions change.

DeliverableMonthly operating report

monitoring / daily_run Operate
Data ready
06:12
Tests passed
412 / 412
Next retrain
01 Dec
Rows loaded, last 14 days (millions)
14 days ago today 1.98
  1. schedule.dailystart
  2. ingest.erp_dailyok 3m 12s
  3. model.demand_forecastok 11m 04s
  4. ingest.bank_feedsretry 2/3
  5. report.board_packok 48s
Monthly operating report · drafteddrift checks: within limits

Most of our work is for financial services, real estate, private capital and family offices, and mid-sized companies.

Principles

How we hold ourselves to account.

  • Every number traces back.

    Each figure we deliver links to its source rows and every transformation applied to it.

  • Your data stays yours.

    It lives in your cloud account. We hand over the code and the documentation.

  • No black boxes.

    Every model ships with its assumptions, its backtest and its known limits written down.

What decision is your data not supporting yet?

Send us the question. We'll reply with a short brief on what it would take to answer it.

Talk to us Our services