When EBITDA Lies: Why the Metric Fails for Banks Lloyds & AI

For a bank, EBITDA has nothing to say. The cost:income ratio, net interest margin and the provision lines do – and what they say about Lloyds’ £2bn AI programme is that it has a savings number, no headcount number, and no AI governance or model-risk provision at all.

The banking & financial services paper in the Bloor Finance series: a CFO-perspective benchmark of Lloyds Banking Group plc (LSE: LLOY) against the UK bank peer set on AI in banking, on the 2025 audited record and the half-year results to 30 June 2026, published between 28 and 31 July 2026.

Prepared by Darren Sack · Research Director · Bloor Finance Desk · Bloor Research International · September 2026

The opening paper in this series made one claim and asked twelve questions. The claim: wherever an organisation embeds AI, cost leaves the wage bill – where EBITDA (earnings before interest, tax, depreciation and amortisation) sees it – and reappears as depreciation, interest, leases and provisions, where it does not. The technology paper showed the swap from labour debt to compute debt at Oracle. This is the banking paper the opening paper promised, and it turns the question round. In a bank the cost being removed is almost entirely people, and the metric that would flatter its removal does not exist.

Five weeks after the opening paper went to press, the sector spoke. Lloyds launched Accelerate 2030 on 30 July with a £2bn savings ambition and no headcount forecast. Barclays reported a 14.8% return on tangible equity and a £2bn three-year efficiency target. NatWest reported 19.7% and a cost:income ratio of 46%.

The rest of the AI-in-banking story arrived in the same weeks. Standard Chartered had already put a number on it in May: 15% of corporate-function roles by 2030. HSBC is reported, and has not confirmed, to be reviewing around 20,000. Morgan Stanley’s May note put the European figure at around 200,000 roles by the end of the decade. Every one of those numbers is a wage-bill number, and every one of them lands on the line a bank actually reports.

EBITDA – earnings before interest, tax, depreciation and amortisation – is operating profit with the financing and accounting charges added back. For most companies it is a rough proxy for operating cash flow. For a bank it measures nothing, because the items it removes are the business itself.

EBITDA was built for businesses where interest is a financing choice and depreciation is the echo of a past capital decision. A bank is neither. Interest is not a cost of financing the business; it is the business. Net interest income of £7.3bn is the largest line on Lloyds’ income statement, and the £17.5bn of interest expense the Group paid in 2025 is not below any line – it is the raw material of the product. Strip it out and there is nothing left to measure.

For a bank, EBITDA is not distorted – it is meaningless.

That is why banks report the gauges they do: net interest margin, cost:income ratio, jaws, return on tangible equity, capital generation, and the provision lines that carry conduct, credit and – in principle – model risk. These are the honest gauges the opening paper asked every CFO to read. In banking they are not an alternative to EBITDA. They are the only instrument on the panel.

Two details make the point sharper than the principle.

First, Lloyds does carry depreciation of exactly the kind EBITDA exists to remove – £841m of it in the first half, up 18%, on a fleet of leased cars rather than servers, including a £41m charge in the second quarter for falling used-car values. Second, the Group’s own bridge from underlying to statutory profit lists the adjustments an EBITDA culture would hide: £34m of restructuring, £65m of amortisation of purchased intangibles from the Schroders Personal Wealth and Curve deals, up from £40m a year earlier. Small numbers, honestly shown.

The question for the rest of this paper is whether the large number – the AI programme – is shown as honestly.

Read without EBITDA, Lloyds’ half-year is strong. Statutory profit before tax rose 23% to £4.3bn; profit after tax to £3.1bn. Return on tangible equity was 17.1% against 14.1% a year earlier, and 17.0% in the second quarter alone, ahead of the full-year guide of more than 16%. Net income on the underlying basis rose 9% to £9.7bn: underlying net interest income up 9% to £7.3bn on a banking net interest margin of 3.19%, up 15 basis points, and underlying other income up 11% to £3.3bn.

Below the income line the story is one number. Operating costs were £4,876m against £4,874m in the first half of 2025 – flat to within £2m. Add £39m of remediation and total costs were £4,915m, again flat. With income up 9% and costs unchanged, jaws were positive by about nine percentage points and the cost:income ratio fell from 55.1% to 50.4%, reaching 49.0% in the second quarter against a full-year guide of below 50%.

Every basis point of that improvement came from the denominator. The numerator did not move.

Capital followed. The Group generated 108 basis points of capital in the half – roughly £2.6bn on £241.8bn of risk-weighted assets – and, after a 30% higher interim dividend of 1.58p (£918m) and a new £1.0bn buyback on top of the £1.75bn programme announced in January, ended the period at a pro forma CET1 ratio of 13.1%, close to its 13.0% target. The structural hedge, the quiet engine of the income line, earned £3.4bn in six months on a £246bn notional, against £2.6bn a year earlier, with guidance of more than £7bn for 2026 and more than £8bn for 2027.

That is the honest reading. It is also the reading that makes the next question urgent. When the cost line is flat and the income line is doing all the work, an AI programme that promises to take £2bn out of costs has to show up somewhere that is not the denominator.

Lloyds Accelerate 2030 is a growth strategy with a £2bn gross cost-savings programme inside it, whose proceeds Lloyds says will finance higher investment. It names AI as the lever and gives no headcount figure. In the opening paper’s terms, that is labour debt – fixed payroll for processed work – being addressed without being disclosed.

The targets are precise: return on tangible equity of around 20% in 2030 and above 18% in 2028; capital generation above 225 basis points in 2030; a cost:income ratio below 45% in 2030 ‘with year-on-year reductions’; and ‘around £2 billion of gross cost savings over the plan’, to be delivered by ‘extending existing levers, such as our digital transformation and tech modernisation, and leveraging new areas, in particular realising AI value to drive a productivity step-change’.

The Group says it has already generated more than £2bn of gross cost savings under the 2022–2026 plan; Accelerate 2030 is the same figure again, this time with AI named as the lever.

Here the AI story is a labour-debt story dressed as transformation.

Three disclosures show where the cost lives. The Group reports a more than 45% improvement in Retail customers served per full-time employee since 2022. It says generative AI is expected to deliver ‘over £100 million of benefit in 2026’, roughly double the c.£50m reported for 2025 – about 1% of a cost base guided below £9.9bn. And 232 branches are closing in 2026.

Asked on the results call what £2bn of technology-driven savings means for jobs, the chief executive said the Group does not put targets around numbers of staff, and that AI ‘is going to impact work’. The paper takes him at his word on both counts. A savings number with a named technology and no headcount reduction figure is the shape the opening paper described: cost leaving the wage bill, destination unstated.

The destination is written into the plan. ‘Higher investment will increasingly be financed by the capacity created by the gross cost saves.’ That sentence is the whole thesis of this series in one line. Savings are gross; investment is higher; the net is what reaches the cost:income ratio.

The Group has committed more than £13bn to digital investment, has hired around 11,000 technology and data specialists, and is recruiting more than 1,000 AI roles in 2026 alone. None of that is wrong. It is simply the reinvestment channel, and it is the reason a flat cost line and a £2bn savings programme can both be true at once.

The opening paper set a test: removed cost should reach the cost:income ratio and positive jaws within two reporting periods, or it has been renamed. Lloyds’ three most recent halves give the raw material. Operating costs were £4,874m in the first half of 2025, £4,887m in the second, £4,876m in the first half of 2026. Net income went £8,914m, £9,387m, £9,747m. Cost:income, excluding remediation, went 54.7%, 52.1%, 50.0%.

The test is passed on jaws – but the cost line has not fallen. Over twelve months and two reporting periods, the cost base moved by £2m.

The question is not whether AI removes cost. It is whether the removed cost is really gone, or merely reinvested and renamed.

The staff-cost note answers the question the headline cannot. Total staff costs fell from £2,472m to £2,387m, a 3% reduction, which on its face is the labour-debt story working.

Look inside it. Salaries and social security rose 4%, from £1,908m to £1,989m. Pensions rose from £270m to £276m. The entire reduction – and more – is the line ‘restructuring and other staff costs’, which fell from £294m to £122m. The Group’s own commentary says as much: costs were flat because of ‘a lower severance expense and plateauing investment as this strategic cycle culminates’. The flat cost line is a severance-timing story. The underlying payroll grew.

Put the two halves of the note together and the labour-debt ratio the opening paper asked for can be estimated. Staff costs were 49% of underlying operating costs in the first half of 2026, down from 51%; salaries alone were 41%, up from 39%.

That is not a wage bill that AI has begun to remove. It is labour debt still on the books. It is a wage bill whose one-off exit costs happened to be lower this half than last. Whether the £2bn programme changes that is precisely what the Q3 statement in late October and the 2026 full-year results in January will begin to show – and the paper’s test is that the salaries line, not the restructuring line, must be the one that moves.

There is one more channel to watch, because it is the one EBITDA-style thinking hides in every industry: capitalised software. Development spend that is capitalised leaves operating costs and returns as amortisation. Lloyds’ CET1 deduction for goodwill and other intangibles was £5.8bn at 30 June 2026; the software element is not separately disclosed in the half-year release and is reported here as n/d. It belongs on the dashboard.

NatWest is already below 46%; Barclays carries the same £2bn ambition over a shorter period; Lloyds is the only one of the three to price its AI benefit in pounds. None of the three discloses a headcount forecast, and none discloses a provision for AI governance or model risk.

Set Lloyds against the two UK-listed peers that reported in the same week and its distinctiveness is not returns or capital. NatWest’s return on tangible equity is higher and its cost:income ratio already below 46%; Barclays’ returns are lower and its cost:income ratio higher, with the investment bank in the mix.

Lloyds is distinctive in one respect: it is the only one of the three to attach a quantified, recurring pounds-and-pence line to AI. Every one of the three has a large gross efficiency ambition; none of the three discloses a headcount forecast; and none of the three discloses a provision, or a capital line, specific to AI or model risk. That last row is empty across the whole table. The empty row is the finding.

Honest gauges – Lloyds against the UK bank peer set (half-year to 30 June 2026, results published 28–31 July 2026)

Bank	RoTE H1 2026 (H1 2025)	Cost:income H1 2026 (H1 2025)	CET1 (pro forma)	AI / efficiency ambition	Disclosed headcount change	AI or model-risk provision
Lloyds	17.1% (14.1%); Q2 17.0%	50.4% (55.1%); Q2 49.0%	13.6% (13.1%)	c.£2bn gross savings by 2030; C:I <45%; GenAI >£100m benefit 2026; 232 branches	n/d – ‘no targets around numbers of staff’	n/d
NatWest	19.7% (18.1%)	46.0% (48.8%), ex litigation & conduct	13.2%	c.£250m gross cost reductions in H1; simplification and technology investment	n/d in results	n/d
Barclays	14.8% (13.2%); Q2 16.1%	55% (n/d); Q2 54% (59%)	14.3% (14.0%)	£2bn gross efficiency savings over three years; C:I high-50s 2026, low-50s 2028	n/d in results	n/d
HSBC (context)	n/d in this paper	n/d in this paper	n/d in this paper	Reported review of c.20,000 roles (c.10%) – not confirmed by the company	Not confirmed	n/d
Standard Chartered (context)	n/d in this paper	n/d in this paper	n/d in this paper	15% of corporate-function roles by 2030 (>7,000 on Reuters’ calculation) – confirmed at investor day	15% of corporate functions	n/d

Lloyds, NatWest and Barclays figures are from each bank’s half-year results announcement. NatWest cost:income excludes litigation and conduct as the bank reports it; Barclays cost:income is the Group ratio as reported. Pro forma CET1 is after announced buybacks where the bank states it. HSBC and Standard Chartered are context only: HSBC’s figure is a Bloomberg report carried by Reuters (19 March 2026) that the company has not confirmed; Standard Chartered’s is the company’s own investor-day statement (19 May 2026) with the role count as calculated by Reuters. n/d – not disclosed in the cited document; never treated as zero. No bank is ranked; the comparison is indicative.

Two contrasts deserve a line each. NatWest reaches 46% without a quantified AI line, which suggests the ratio is a function of mix and scale before it is a function of technology; Lloyds’ 45% target for 2030 is, on that reading, catching up rather than leaping ahead. Barclays, whose group ratio is diluted by the investment bank, has the same £2bn gross number as Lloyds over a shorter period and had delivered roughly £150m of it by the first quarter.

Gross efficiency ambitions of £2bn are, it turns out, the going rate. What distinguishes the banks is what they say happens to the money afterwards – and Lloyds is the only one to say, in its own results release, that it is financing higher investment.

Lloyds holds £1.95bn of provision against motor-finance conduct and discloses nothing against AI governance or model risk. Neither do its peers. The only capital line that touches AI at all – operational-risk capital – is computed from income, not from the models the bank runs.

Banks are, of all businesses, the ones that know what a late provision looks like. Lloyds’ motor finance provision – for commission arrangements on car loans – stands at £1,950m in total after an £800m charge in the third quarter of 2025 – the largest single item in a 2025 remediation bill of £968m, and the reason return on tangible equity for 2025 printed at 12.9% rather than the 14.8% the Group reports excluding it.

The half-year release records the current state: no further charge; the FCA’s final redress rules published in March 2026; four legal challenges lodged; an Upper Tribunal hearing not expected before December 2026; and the provision remaining, in the Group’s words, its best estimate. That is how conduct liability arrives – years after the conduct, in one large tranche, with the amount still contested.

Now look for the equivalent line for the technology that Accelerate 2030 places at the centre of the bank. The half-year release lists twelve principal risks, including model risk and operational risk, and commits the Group to ‘safe and responsible use of models and tools such as artificial intelligence’. It contains no quantified provision for AI governance, model risk or algorithmic conduct. Nor does the peer set.

The nearest thing on any UK bank’s balance sheet is operational-risk capital: for Lloyds, £27.8bn of risk-weighted assets at the end of 2025, about 12% of the total, which at a 13% CET1 target implies roughly £3.6bn of capital held against operational risk of every kind. But that charge is computed from a rolling three-year average of income, not from any assessment of the models the bank runs; the Group’s own Pillar 3 disclosure attributes the year’s increase to income growth. It rises when the bank earns more. It does not rise when the bank deploys an agent. AI governance, in other words, has a policy but not a number.

The opening paper called this Ghost Liability: an obligation the balance sheet does not yet see because nothing has forced it to. The motor-finance line shows what forcing looks like. The AI line shows a bank that has quantified the benefit of the technology to the pound – £50m, then £100m, then £2bn – and has not quantified the risk of it at all. That asymmetry is the finding, and reporting the provision as n/d, rather than as zero, is the only honest way to record it.

In banking the remedy is not to add gauges but to read the existing ones for the right thing. The banking edition of the depreciation-aware dashboard, applied to Lloyds’ first half, reads as follows.

– Cost:income ratio and jaws, together – and which side moved. Lloyds: 50.4%, jaws about +9 points; the cost line moved £2m, the income line £833m. Improvement that is all denominator is income, not efficiency.

– Return on tangible equity against the cost of equity, and against the plan. Lloyds: 17.1% now, above 18% guided for 2028, around 20% for 2030; cost of equity is not disclosed and is reported here as n/d.

– The labour-debt ratio: labour cost as a share of operating costs, split into salaries and exit costs. Lloyds: 49% (51% a year ago); salaries 41% and rising; restructuring and other staff costs £122m against £294m. The salaries line is the one that must move.

– Restructuring and severance as a share of stated savings – the price of the saving, and whether it is inside or outside the underlying figure. Lloyds: £34m below the line plus £122m within staff costs; savings for the half not disclosed, so the ratio is n/d.

– Capitalised software as a share of operating costs – the reinvestment channel that turns today’s saving into tomorrow’s amortisation. Lloyds: n/d in the half-year release; intangibles deducted from CET1 £5.8bn.

– Net interest margin and the structural hedge, so that income tailwinds are not credited to the cost programme. Lloyds: 3.19%; hedge income £3.4bn in six months, up £0.8bn, with more than £2bn of additional hedge income guided over the plan.

– Capital generation against distributions. Lloyds: 108 basis points generated; £918m of dividend and £1.0bn of buyback announced in respect of the half.

– The provision panel, read as a set: conduct provisions on the balance sheet, operational-risk capital, and the AI-governance and model-risk line. Lloyds: £1.95bn; £27.8bn of RWAs; n/d.

Read together, these separate an efficiency programme from an income cycle – which, for a UK bank in 2026, is the distinction that matters.

For a bank CFO the work is specific. Publish the staff-cost note with the same prominence as the cost:income ratio, and let the salaries line be the scorecard for the AI programme. State savings gross and net, and say where the gross goes. Disclose capitalised development spend alongside operating costs. Name the cost of equity.

And put a number – any number, with a method – against AI-governance and model risk, because the alternative is the number the regulator or the courts will eventually supply. The motor-finance provision is the sector’s own evidence that liabilities nobody quantified do not stay unquantified.

Lloyds is not the weak case. On the honest gauges – return on tangible equity, cost:income ratio, net interest margin, CET1 ratio – it is a strong one, and EBITDA would have told you none of it. Rising returns, disciplined capital, a credible plan, and more transparency about its AI economics than either peer. That is why it is the right case. If the flat cost line of the UK’s most AI-forward high-street bank turns out, on inspection, to be a severance-timing effect on a growing payroll, the sector’s £2bn ambitions should be read the same way until the salaries line says otherwise.

Watch for the remaining When EBITDA Lies industry papers as they publish. For boards that want an early diagnostic read of their sector’s honest gauges before their industry paper lands, the Bloor Finance Desk is available for a confidential conversation.

Every figure in this paper is drawn from a named, dated document; the principal ones are listed below with links, and a full Source Note listing the origin, pinpoint and status of every figure accompanies this paper. Ratios, jaws, the labour-debt ratio and the operational-risk capital estimate are Bloor arithmetic on reported figures. Headcount, cost of equity, capitalised software and any AI governance or model-risk provision are not disclosed in the cited documents and are shown as n/d, never as zero. No bank is ranked. This is analysis for a CFO audience and is not investment advice.

1. Lloyds Banking Group — 2026 Half-Year Results, 30 July 2026. Includes Accelerate 2030.
   Results news release (PDF)
2. Lloyds Banking Group — Form 6-K, 30 July 2026. See Note 6, Operating expenses, for staff costs.
   SEC filing
3. Lloyds Banking Group — 2025 Results, 29 January 2026. Covers the motor-finance provision and risk-weighted assets by risk type.
   Results news release — SEC
4. Lloyds Banking Group — 2025 Year-End Pillar 3 Disclosures, 17 February 2026. Covers operational-risk capital.
   Pillar 3 disclosures (PDF)
5. Barclays PLC — H1 2026 Interim Results Announcement, 28 July 2026.
   Interim results (PDF)
6. NatWest Group — Interim Results 2026, 31 July 2026.
   Results announcement — Investegate
7. Reuters, citing Bloomberg — HSBC workforce review, 19 March 2026. The report describes a review at an early stage, with no final decisions made.
   HSBC weighs deep job cuts as AI overhaul unfolds reuters.com
8. Reuters — Standard Chartered investor-day announcement, 19 May 2026.
   StanChart to cut over 7,000 jobs, boost AI to replace ‘lower-value human capital’ reuters.com
9. Morgan Stanley — European banking research note, May 2026. I could not locate a public link to the original note. This is secondary reporting on its forecasts:
   Business Standard — AI could lead European banks to cut up to 20% jobs: Morgan Stanley, 28 May 2026. World News – Business Standard
10. City AM — Lloyds chief executive’s remarks on staffing and branch closures, 30 July 2026.
    ‘It’s going to impact work’: Lloyds to cut £2bn in costs with AI cityam.com

Print Article for this blog Download here